Merge remote-tracking branch 'origin/master' into fix/workspace-blank-session-reuse-membership

# Conflicts:
#	packages/client/runtime/README.i18n.yaml

.agents/notes/implemented/architecture/2026-07-30-web-config-plane.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/architecture/2026-07-30-web-config-plane.md
-2026-07-30-web-config-plane.md: 6d1a8c242c1888ee4fca9e21ebc814f7a345d633
-2026-07-30-web-config-plane.zh.md: 184299cf0c44b992e55a268c990b468c6d13f947
+2026-07-30-web-config-plane.md: 5225460be1d66b85a05ff2fd5ae2826b0e6c41d7
+2026-07-30-web-config-plane.zh.md: 53a21ddf31640d963c413e1793276de694547311

.agents/notes/implemented/feature/2026-07-06-sandbox.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-06-sandbox.md
-2026-07-06-sandbox.md: 4355ab57374f77f1733fff39e2fb4ccadcedaf6b
-2026-07-06-sandbox.zh.md: bcfadae1d5b4373e1b3def0dcf09c9652c184891
+2026-07-06-sandbox.md: 6934691e9b3f37e50bb02dfec0240a03ec9b5f30
+2026-07-06-sandbox.zh.md: cc5b7e5be0f697d77732efbf55667c258b69a95f

.agents/notes/implemented/feature/2026-07-06-sandbox.md

@@ -36,9 +36,9 @@ Four `cordis.yml` entries turn an unconfined coding agent into the sandboxed pro
   name: '@deepseek-ai/dsh-permission'      # one product-facing select over both mechanism knobs
 ```
 
-The swap is invisible to every consumer of `ctx.bash`: the bash tools, hook commands, and background tasks run exactly as before, spawned through the wrapped argv the provider returns. Deleting the `sandbox` and `permission` entries and replacing `bash` with `@deepseek-ai/dsh-bash-local` is the opt-out — execution is unconfined again and the escalation fields vanish from the tool schema, because they are capability-gated on the mounted executor, not on configuration. Omitting only `approval` keeps confinement but fails every escalation closed with its own error text; `permission` also requires the approval seam and a confining executor, so a partially composed preset layer fails loud at load.
+The swap is invisible to every consumer of `ctx.bash`: the bash tools, hook commands, and background tasks run exactly as before by directly spawning the wrapped argv the provider returns. Deleting the `sandbox` and `permission` entries and replacing `bash` with `@deepseek-ai/dsh-bash-local` is the opt-out — execution is unconfined again and the escalation fields vanish from the tool schema, because they are capability-gated on the mounted executor, not on configuration. Omitting only `approval` keeps confinement but fails every escalation closed with its own error text; `permission` also requires the approval seam and a confining executor, so a partially composed preset layer fails loud at load.
 
-Misconfiguration fails loud: `mode` outside the closed vocabulary is rejected at plugin load, and a host with no usable backend throws the structured `SANDBOX_UNAVAILABLE` — at `confine()` before the command ever spawns — rather than degrading to unconfined execution. `runnerCommand` on `dsh-sandbox-local` is the operator's explicit assertion of a bwrap-compatible runner (chain and probes skipped); it doubles as the deterministic fake-runner seam for keyless tests.
+Misconfiguration fails loud: `mode` outside the closed vocabulary is rejected at plugin load, and a host with no usable backend throws the structured `SANDBOX_UNAVAILABLE` at `confine()` rather than degrading to unconfined execution. If the selected runner rejects with attributable `ENOENT` or `EACCES`, the consumer reports the same infrastructure error from the spawn channel before any command starts; other spawn errors retain local command-start semantics while still running nothing. `runnerCommand` on `dsh-sandbox-local` is the operator's explicit assertion of a bwrap-compatible runner (chain and probes skipped); it doubles as the deterministic fake-runner seam for keyless tests.
 
 Denied file effects return a `[sandbox: file access denied under <mode> mode]` marker and instructions not to work around the denial. A confining executor adds paired `sandbox_permissions` and `justification` fields for one approved retry that must be strictly wider than the session's effective mode. A grant widens only that retry; rejection executes nothing, returns `the user rejected escalating this command to "<mode>"`, and permits no re-ask. The owner-derived runtime context states the current file policy without replacing those enforcement boundaries. When `dsh-permission` is composed with a UI adapter, one preset selects both knob values; unmatched values fold to `custom`. The [ACP automation composition](../../../../examples/acp-agent/README.md) does not mount that UI service and selects its deployment mode explicitly.
 
@@ -50,7 +50,7 @@ OS subprocess confinement applies to the bash executor, including hook commands,
 
 #### The seam: `ctx.sandbox`
 
-`dsh-sandbox` owns the vocabulary and the `SandboxProvider` contract: `confine(argv, policy)` returns the argv to spawn INSTEAD of the caller's own — wrapped so the process and everything it spawns run confined — plus the `enforcement` completeness the selected backend achieves, its denial dialect (`denialSignatures`, the stderr substrings that backend's kernel prints on a denied file effect), and its runner-failure dialect (`runnerFailureSignatures`, how the runner ITSELF failing — and therefore the command never running — identifies itself); with no usable backend it throws the fail-closed `SANDBOX_UNAVAILABLE` error, never a silent unconfined passthrough. The vocabulary: `SandboxMode` (`read-only` / `workspace-write` / `danger-full-access`, FILE effects only — network and process visibility are not claimed), `SandboxEnforcement` (`full` / `partial`), `SandboxExecutionPolicy` (the complete per-capability-call mode + workspace root), and `SandboxPolicy` (the confined provider subset).
+`dsh-sandbox` owns the vocabulary and the `SandboxProvider` contract: `confine(argv, policy)` returns the argv to spawn INSTEAD of the caller's own — wrapped so the process and everything it spawns run confined — plus the `enforcement` completeness the selected backend achieves, its denial dialect (`denialSignatures`, the stderr substrings that backend's kernel prints on a denied file effect), and its structured runner-failure evidence (`runnerFailureRules`, optional allowed exit codes plus fatal per-line signatures after exact informational-line exclusions); with no usable backend it throws the fail-closed `SANDBOX_UNAVAILABLE` error, never a silent unconfined passthrough. The vocabulary: `SandboxMode` (`read-only` / `workspace-write` / `danger-full-access`, FILE effects only — network and process visibility are not claimed), `SandboxEnforcement` (`full` / `partial`), `SandboxExecutionPolicy` (the complete per-capability-call mode + workspace root), and `SandboxPolicy` (the confined provider subset).
 
 Policy rides each CALL, not the provider: two consumers may confine under different policies at the same instant (bash under `read-only` while a confined child agent keeps its state directory writable), and an approved escalated retry is a new call with a wider policy — inexpressible under a config-fixed provider mode.
 
@@ -60,17 +60,17 @@ Left open, for the phase that needs them: whether network restriction arrives as
 
 #### Local backends and the shipped launcher
 
-`dsh-sandbox-local` selects one platform runner per provider lifetime and caches the verdict. Linux functionally probes `bwrap` then Landlock; macOS uses Seatbelt. Unsupported platforms and unusable runners fail closed. Each wrap carries backend-specific denial and runner-failure signatures so `dsh-bash-sandbox` can distinguish a denied file effect from a broken sandbox. `runnerCommand` skips selection as an operator assertion of a bwrap-shaped runner, but missing or unexecutable commands still classify as sandbox failure and never run the payload unconfined.
+`dsh-sandbox-local` selects one platform runner per provider lifetime and caches the verdict. Linux functionally probes `bwrap` then Landlock; macOS uses Seatbelt. Unsupported platforms and unusable runners fail closed. Each wrap carries backend-specific denial signatures and runner-failure rules so `dsh-bash-sandbox` can distinguish a denied file effect from a broken sandbox. Landlock runner failure requires exit 125 plus a non-notice `landlock-run:` line; the exact partial-enforcement line is informational even when a child exits 1, 2, or 125. Bubblewrap and Seatbelt remain signature-only because neither public contract reserves a launcher-failure status. `runnerCommand` keeps its operator-facing `runnerFailureSignatures` config, requires non-empty single-line entries, and maps them into one internal fatal rule. The consumer directly spawns every returned argv, so a missing runner, a non-executable runner, or an executable script whose shebang interpreter is unavailable fails through the attributable `ENOENT`/`EACCES` spawn channel while a successfully launched child exit 126 or 127 remains ordinary. An operator-configured script necessarily owns its interpreter startup before it applies its profile.
 
-The launcher is a ~300-line C program (plain C11 over the raw Landlock UAPI — no libraries beyond a statically linked musl, so the audit surface is that one file plus the kernel's stable syscall contract): `--ro <path>` / `--rw <path>` grants, `--`, the wrapped argv; it installs the ruleset on itself and `exec`s (rulesets are inherited across `execve`, and it sets `no_new_privs` before restricting); `--probe` enforces a maximal ruleset in a short-lived child and exits 0 only when the kernel actually enforces; launcher failures exit 125 without exec'ing.
+The launcher is a ~300-line C program (plain C11 over the raw Landlock UAPI — no libraries beyond a statically linked musl, so the audit surface is that one file plus the kernel's stable syscall contract): `--ro <path>` / `--rw <path>` grants, `--`, the wrapped argv; it installs the ruleset on itself and `exec`s (rulesets are inherited across `execve`, and it sets `no_new_privs` before restricting); `--probe` enforces a maximal ruleset in a short-lived child and exits 0 only when the kernel actually enforces; every launcher failure exits 125 without running the child and prints a fatal `landlock-run:` line. A successfully exec'd child may also return 125, so status alone is not launcher evidence. An older ABI prints the exact `landlock-run: partial enforcement (older Landlock ABI)` notice before it executes the child, so that line is not fatal evidence.
 
-The Landlock launcher source and package workspace live at `native/landlock-run`, next to the harness consumers. The standalone [`node-addon-landlock-run`](https://github.com/deepseek-harness/node-addon-landlock-run) repository is the release mirror used to pack and publish the npm package family; `native/README.md` owns the export procedure. Platform binaries are selected by npm, and the entry package owns path resolution, probing, and CLI flags while the harness maps sandbox modes to grants. Versioning the entry point with its binaries keeps probe parsing and launch syntax aligned.
+The Landlock launcher source and package workspace live at `native/landlock-run`, next to the harness consumers. The standalone [`node-addon-landlock-run`](https://github.com/deepseek-harness/node-addon-landlock-run) repository is the release mirror used to pack and publish the npm package family; `native/README.md` owns the export procedure. Platform binaries are selected by npm, and the entry package owns path resolution, probing, CLI flags, the fatal prefix, and the partial-enforcement notice while the harness maps sandbox modes to grants. Versioning the entry point with its binaries keeps probe parsing and launch syntax aligned.
 
 Backend profiles share the mode contract but differ in necessary host grants. Landlock and Seatbelt allow only `/dev/null` in read-only mode; workspace-write also permits their required host temp roots. Each wrap carries backend-specific denial signatures. Landlock reports partial enforcement on older ABIs that cannot govern every operation, while successful bwrap and Seatbelt profiles report full enforcement.
 
 #### The bash consumer
 
-`dsh-bash-sandbox` extends `LocalBashExecutor` and hands `ctx.sandbox` the exact `['bash', '-c', command]` argv it is about to spawn. A denial is an orthogonal result fact, conservatively classified from the active runner's stderr dialect. A runner failure outranks denial: foreground execution throws `SANDBOX_UNAVAILABLE`; a settled `BashProcess` stamps `sandbox.runnerFailed`, and the bash producer renders it through generic `task_output`.
+`dsh-bash-sandbox` extends `LocalBashExecutor`, hands `ctx.sandbox` the exact `['bash', '-c', command]` argv, and directly spawns the provider result. This leaves shell semantics and `BASH_ENV` on the inner Bash after the shipped native runner establishes confinement. A provider error propagates unchanged. A pre-process rejection is runner-owned only when the caller-owned workdir is independently usable and Node reports `ENOENT` or `EACCES` with positive provenance for provider argv[0]; a bare `syscall: 'spawn'` without an exact error path, other codes, invalid workdirs, resource failures, unrelated syscalls, and unstructured rejections retain local command-start semantics. Foreground execution converts runner-owned rejections to `SANDBOX_UNAVAILABLE` with the original detail; an asynchronous background rejection stamps `runnerFailed: true`, `denied: false`. A `SubprocessService` that synchronously throws the same provenanced shape makes background start throw `SANDBOX_UNAVAILABLE`, while other synchronous errors propagate unchanged. After a process starts, foreground and background use one runner-failure classifier that requires the rule's exit-code gate and a remaining fatal line after informational exclusions. A match outranks denial: foreground execution throws `SANDBOX_UNAVAILABLE` with that fatal line as detail; a settled `BashProcess` stamps `sandbox.runnerFailed`, and the bash producer renders it through generic `task_output`.
 
 The model sees the current effective file policy in the owner-derived `sandbox:policy` runtime context, while the static tool description explains the denial marker (`[sandbox: file access denied under <mode> mode]`), encourages attempting commands that may be denied, and forbids retrying around a denial; when the escalation fields are advertised, a denied result additionally carries the escalation hint itself, so the sanctioned same-turn retry is prompted at the decision point rather than depending on the model recalling the description (§ Escalation). [The current-policy decision](2026-07-30-current-sandbox-policy-context.md) owns the context's rationale and boundaries.
 
@@ -78,7 +78,7 @@ The model sees the current effective file policy in the owner-derived `sandbox:p
 
 `BashExecRequest.sandboxPolicy` is an optional complete per-call input; resolved specs make the field explicit. `BashExecutor.sandboxMode` remains the capability fact advertising whether the mounted executor can honor that policy, so only a confining composition exposes escalation. The seam accepts any explicit policy; the tool owns session resolution and the wider-only escalation rule. Non-sandboxing executors remain honestly unconfined.
 
-`ctx.sandboxPolicy.resolve()` stamps the complete execution policy — explicit escalation mode > session override > configured default, with `SessionHeader.cwd` > configured fallback root — before the executor runs. `SandboxBashExecutor.resolve()` retains that policy on the spec, or supplies the deployment fallback for a direct agentless caller, so `run()`/`start()` never read mutable session state. Per-process wrap facts are keyed by the returned `BashProcess`; `onProcessDone()` classifies stderr and stamps that handle before `done` resolves, so overlapping processes retain their own modes and runner dialects.
+`ctx.sandboxPolicy.resolve()` stamps the complete execution policy — explicit escalation mode > session override > configured default, with `SessionHeader.cwd` > configured fallback root — before the executor runs. `SandboxBashExecutor.resolve()` retains that policy on the spec, or supplies the deployment fallback for a direct agentless caller, so `run()`/`start()` never read mutable session state. Per-process wrap facts are keyed by the returned `BashProcess`; `onProcessDone()` receives spawn failure out of band from stderr classification and stamps that handle before `done` resolves, so overlapping processes retain their own modes and runner dialects.
 
 When a confining executor is mounted, `bash` advertises paired `sandbox_permissions` and `justification` fields. The schema exposes the full closed escalation vocabulary because effective mode is per-session; execution rejects any target that is not strictly wider than that call's effective mode. Approval resolves before execution. `allowed-once` stamps the granted mode onto only that request, while `rejected`, `cancelled`, `unavailable`, a missing approval service, or a missing agent all fail closed with distinct results. No grant is persisted.
 
@@ -117,10 +117,10 @@ fs/web/todo execute in-process, so their sandbox semantics are policy at their s
 
 ### Testing
 
-- **Unit:** pin platform selection and profiles, fail-closed runner classification, per-call mode/root resolution, per-process facts, escalation validation and outcomes, permission preset folding and write-through, and runtime-context ordering and materialization.
+- **Unit:** pin platform selection and profiles, direct provider-argv handoff, spawn-level failures with invalid-workdir controls, missing/non-executable/missing-interpreter evidence, malformed-runner negative controls, confined `BASH_ENV` ordering, structured runner classification (including partial-Landlock notice-only child outcomes, gated fatal evidence, child exits 126/127, and foreground/background parity), per-call mode/root resolution, per-process facts, escalation validation and outcomes, permission preset folding and write-through, and runtime-context ordering and materialization.
 - **Keyless real-runner:** exercise bwrap, Landlock, and Seatbelt against real filesystem effects at provider and bash-consumer layers; one real Cordis context concurrently drives two project sessions through shipped bash and fs tools, proving own-root success and sibling-root denial. Packed-install coverage proves the registry launcher remains executable. CI rejects a silent all-skip.
 - **With-key:** start the real ACP composition in read-only mode, let a model-driven bash write hit the runner's denial marker, then drive the bridge answerer and disk effect through granted and rejected workspace-write retries; unavailable credentials or runners self-skip.
-- **Snapshot:** pin the atomic current-policy context and both scripted approval branches. A real ACP example scenario places its session under the user home while the deployment fallback points at `/tmp`, then pins both the workspace-write runtime-context message and a successful deployment-selected mutation; this distinguishes session-root resolution from the process fallback without depending on runner-specific denial text. Other snapshots start unconfined so unrelated fixtures remain platform-independent.
+- **Snapshot:** pin the atomic current-policy context and both scripted approval branches. A real ACP example scenario places its session under the user home while the deployment fallback points at `/tmp`, then pins both the workspace-write runtime-context message and a successful deployment-selected mutation; this distinguishes session-root resolution from the process fallback without depending on runner-specific denial text. A POSIX fake partial-Landlock provider pins direct bash `false` as an ordinary child result and a missing provider executable as foreground/background infrastructure failure through the assembled app. Other snapshots start unconfined so unrelated fixtures remain platform-independent.
 
 ## Deferred phases
 
@@ -128,12 +128,12 @@ Each phase gets its full design when picked up, validated against the code at th
 
 - **Second consumer** — `subagent-acp` optionally confines child agents (per-call policy; unconfined default — a child agent must write its own persistence).
 - **More environments** — an environment-coherent capability group example (e.g. bash+fs against one container).
-- **Windows chain** — `PLATFORM_CHAINS.win32` is reserved and empty (fail-closed); filling it means a confinement runner from the AppContainer/restricted-token family, shipped from its own repository on the `node-addon-landlock-run` template, plus its profile dialect and denial/runner-failure signatures. Wrapping the third-party landstrip runner instead was [considered and rejected](../../rejected/feature/2026-07-26-evaluate-landstrip-for-windows-sandbox-rung.md) — not battle-tested enough for a security invariant.
+- **Windows chain** — `PLATFORM_CHAINS.win32` is reserved and empty (fail-closed); filling it means a confinement runner from the AppContainer/restricted-token family, shipped from its own repository on the `node-addon-landlock-run` template, plus its profile dialect, denial signatures, and runner-failure rules. Wrapping the third-party landstrip runner instead was [considered and rejected](../../rejected/feature/2026-07-26-evaluate-landstrip-for-windows-sandbox-rung.md) — not battle-tested enough for a security invariant.
 
 ## Alternatives considered
 
 - **Command-string heuristic preflight** — rejected: cannot understand expansion/subprocesses/symlinks; the strict attempt (run it, let the kernel decide) is the only trustworthy denial signal.
-- **Functionally probe even a platform's sole backend** — rejected: probing arbitrates between candidates; with one there is nothing to decide, and probe cost taxes the first confined command of every session (prohibitive for heavy future backends). The runner's own exec-time fail-closed refusal plus `runnerFailureSignatures` classification carries the safety property instead.
+- **Functionally probe even a platform's sole backend** — rejected: probing arbitrates between candidates; with one there is nothing to decide, and probe cost taxes the first confined command of every session (prohibitive for heavy future backends). The runner's own exec-time fail-closed refusal plus structured `runnerFailureRules` classification carries the safety property instead.
 - **Commit the built launcher binaries** — rejected: a binary in a diff is unreviewable and churns history; reviewed source + native CI builds + the launcher repo's byte-pinned publish rehearsal keep bytes out of every tree.
 - **Compile the launcher on install** — rejected: pushes a C toolchain onto every consumer; a fallback that exists only where a compiler happens to be is not a fallback.
 - **Cross-compile both architectures from one builder** — rejected: requires carrying a pinned cross toolchain (rustup targets, zig, or a container image) solely to rebuild two ~70 KB binaries; per-architecture native runners already exist and each builds its own platform package (the `node-addon-require-builtin` model, the launcher repo's own pipeline).
@@ -172,19 +172,20 @@ Costs and accepted limits:
 - **The one-wrapper illusion is given up knowingly.** A `tools/pre-execute` wrapper plus prompt conventions does not solve sandbox approval — the correct design costs structured denials, native runner probes, per-call policy carriage, and consistent cross-family enforcement, and this design pays it.
 - **`read-only` became a cross-family boundary through a follow-up.** This RFC shipped bash-only enforcement; the [cross-family fs sandbox RFC](2026-07-14-cross-family-fs-sandbox.md) extends the same mode vocabulary to the filesystem tools through a sandboxed `ctx.fs` provider and relocates the mode/root config and the `sandbox/mode` override to `ctx.sandboxPolicy` (§ In-process tools).
 - **Windows has no backend.** Its chain slot is reserved empty — fail-closed, never a fallthrough; filling it is a deferred phase.
-- **The Seatbelt rung leans on Apple's deprecated-but-shipped `sandbox-exec` CLI.** As darwin's sole candidate it is selected without probing, so a future removal surfaces at execution as the runner-failure classification — re-thrown `SANDBOX_UNAVAILABLE`, the command never runs; fail closed, never open.
+- **The Seatbelt rung leans on Apple's deprecated-but-shipped `sandbox-exec` CLI.** As darwin's sole candidate it is selected without probing, so a future removal under a usable workdir surfaces as a runner-attributable spawn failure and an executable refusal through its fatal signature — both become `SANDBOX_UNAVAILABLE`, and the command never runs; fail closed, never open.
 - **Landlock confinement is only as complete as the running kernel's ABI.** Reported as `enforcement: 'partial'` rather than refused — the deliberate trade that keeps the fallback available on older-kernel hosts.
+- **Runner attribution uses an in-band protocol.** Exit status plus stderr cannot cryptographically identify the writer, so a confined child can mimic a fatal runner line and status to cause an availability/diagnostic false attribution. The conjunction and exact notice exclusion reduce accidental matches; this is not a sandbox bypass because the child is already confined.
 - **The launcher arrives as a registry dependency.** Trusted through its own repository's release pipeline (reviewed C source, native CI builders, byte-pinned publish rehearsal) plus this repo's version pin — the real-kernel e2e legs are what vouch for behavior through the installed bytes.
 - **The model may over-ask.** Escalating without denial grounding, or picking `danger-full-access` where `workspace-write` suffices: the description steers and the enum forces the ladder, but the human prompt is the actual gate; the `approval/asked` reasons make over-asking auditable, and a `prepend` policy answerer can auto-reject patterns a deployment never wants.
 - **The advertised target set is static while the effective mode is per-session** (schemas are registry-global) — a session already at the widest mode is still offered the fields. Harmless by construction: the strict-wider check at execution, not the enum, is the safety boundary — a non-widening request fails with its own text and never prompts anyone.
-- **A granted escalation is not a working sandbox.** An unavailable backend still fails closed even for a granted escalation to a confining mode — at `confine()` when the platform has no chain or every probe fails, at execution when an unprobed sole runner refuses (classified as a sandbox failure, not a command failure) — while a granted `danger-full-access` run never touches the provider at all: there the grant, not the probe, is the authority.
+- **A granted escalation is not a working sandbox.** An unavailable backend still fails closed even for a granted escalation to a confining mode — at `confine()` when the platform has no chain or every probe fails, through the spawn channel when the selected executable cannot start, or through a structured rule when a started runner refuses — while a granted `danger-full-access` run never touches the provider at all: there the grant, not the probe, is the authority.
 - **Runtime-context history is append-only.** A policy switch appends a complete superseding snapshot after retained history, preserving the stable system-and-conversation prefix; unchanged state adds no message.
 - **Older policy snapshots remain in history.** Each full snapshot explicitly supersedes earlier runtime-context snapshots, so replay and compaction need only retain the latest materialized message.
 
 ## FAQ
 
 - **A command came back with `[sandbox: file access denied under read-only mode]` — did it fail?** It RAN, and the kernel refused a file effect: the denial is a result fact orthogonal to exit code. The teaching forbids retrying around it; the one sanctioned move is the same command retried once with an escalation request.
-- **How is a BROKEN sandbox told apart from a failing command?** Runner failure outranks denial in classification: a failed run matching the wrap's `runnerFailureSignatures` means the command NEVER ran — foreground re-throws the structured `SANDBOX_UNAVAILABLE` with the runner's stderr line, a background task stamps `sandbox.runnerFailed` and renders its own marker. A broken sandbox can never read as a failing command, and the command never runs unconfined.
+- **How is a BROKEN sandbox told apart from a failing command?** Any provider-argv spawn rejection proves the confined launch never started, but it identifies a broken runner only when the caller-owned workdir is usable and Node reports attributable `ENOENT` or `EACCES` for that argv[0]. A bare `syscall: 'spawn'` without an exact error path and all other rejections remain ordinary command-start errors. After a process starts, runner failure outranks denial only when one `runnerFailureRules` entry matches both its optional exit-code gate and a fatal stderr line after exact informational exclusions. Foreground failures throw structured `SANDBOX_UNAVAILABLE` with spawn or matched-line detail; an asynchronously rejected or settled background task stamps `sandbox.runnerFailed` and renders its own marker. A `SubprocessService` that synchronously throws the same provenanced `ENOENT`/`EACCES` shape makes background start throw the structured error; other synchronous errors propagate unchanged. A Landlock partial-enforcement notice plus an ordinary child failure remains a command result.
 - **What happens on a platform with no backend — Windows today?** `confine()` throws the fail-closed `SANDBOX_UNAVAILABLE` and the command never spawns; `win32` is a reserved EMPTY chain, pinned by test to fail closed identically until a Windows runner fills it (§ Deferred phases).
 - **`bwrap` is installed on my host but unusable (disabled unprivileged userns, an LSM denying `mount`) — what happens?** The chain probe is functional — it builds and enforces a real profile rather than checking `--version` — so a present-but-unusable `bwrap` fails its probe, selection falls to the registry-installed Landlock launcher, and the verdict is cached for the provider's lifetime.
 - **Does the sandbox restrict network or process visibility?** No — `SandboxMode` claims FILE effects only; the bwrap profile deliberately does not unshare pid, and no backend claims network. Whether network restriction becomes its own knob is left open in § The seam.

.agents/notes/implemented/feature/2026-07-06-sandbox.zh.md

@@ -36,9 +36,9 @@ harness 是一个 SDK，因此约束必须是开发者可组合的能力：是
   name: '@deepseek-ai/dsh-permission'      # one product-facing select over both mechanism knobs
 ```
 
-这一替换对 `ctx.bash` 的所有消费方透明：bash 工具、钩子命令和后台任务照常运行，使用提供方返回的包装 argv 启动。删除 `sandbox` 和 `permission` 条目、将 `bash` 替换为 `@deepseek-ai/dsh-bash-local` 即为退出——执行恢复为无约束，升级字段从工具 schema 中消失，因为它们是基于已挂载执行器的能力门控，而非基于配置。仅省略 `approval` 则保留约束，但会以自身错误文本拒绝每次升级；`permission` 还要求 approval seam 和约束执行器同时存在，因此部分组合的 preset 层在加载时即大声失败。
+这一替换对 `ctx.bash` 的所有消费方透明：bash 工具、钩子命令和后台任务照常运行，直接使用提供方返回的已包装 argv 启动。删除 `sandbox` 和 `permission` 条目、将 `bash` 替换为 `@deepseek-ai/dsh-bash-local` 即为退出——执行恢复为无约束，升级字段从工具 schema 中消失，因为它们是基于已挂载执行器的能力门控，而非基于配置。仅省略 `approval` 则保留约束但以自身错误文本关闭每次升级；`permission` 还要求 approval seam 和约束执行器同时存在，因此部分组合的 preset 层在加载时即大声失败。
 
-配置错误会显式导致失败：`mode` 不在封闭词汇中时在插件加载时被拒绝；主机上没有可用后端时在 `confine()` 阶段（命令 spawn 之前）抛出结构化的 `SANDBOX_UNAVAILABLE`，而非降级为无约束执行。`dsh-sandbox-local` 上的 `runnerCommand` 是运维人员对一个 bwrap 兼容 runner 的显式断言（跳过链和探测）；它同时充当 keyless 测试的确定性 fake-runner seam。
+配置错误会显式导致失败：`mode` 不在封闭词汇中时在插件加载时被拒绝；主机上没有可用后端时在 `confine()` 阶段抛出结构化的 `SANDBOX_UNAVAILABLE`，而非降级为无约束执行。如果所选 runner 以可归因的 `ENOENT` 或 `EACCES` 拒绝，消费方会在任何命令开始前通过 spawn 通道报告同一基础设施错误；其他 spawn 错误仍保留本地命令启动语义，同时也不会运行任何内容。`dsh-sandbox-local` 上的 `runnerCommand` 是运维人员对一个 bwrap 兼容 runner 的显式断言（跳过链和探测）；它同时充当 keyless 测试的确定性 fake-runner seam。
 
 被拒绝的文件操作返回 `[sandbox: file access denied under <mode> mode]` 标记，并附带不要绕过拒绝的指令。约束执行器添加配对的 `sandbox_permissions` 和 `justification` 字段，用于一次经批准的重试，该重试必须严格宽于会话的有效模式。授权仅放宽该次重试；拒绝则不执行任何内容，返回 `the user rejected escalating this command to "<mode>"`，且不允许再次请求。由归属方派生的运行时上下文会说明当前文件策略，但不会取代这些强制执行边界。当 `dsh-permission` 与某个 UI 适配器一起组合时，一个 preset 会同时选定两个配置项的值；不匹配的组合折叠为 `custom`。[ACP 自动化组合](../../../../examples/acp-agent/README.md)不挂载该 UI 服务，而是显式选定其部署模式。
 
@@ -50,7 +50,7 @@ OS 子进程约束适用于 bash 执行器（包括钩子命令），后续还
 
 #### seam：`ctx.sandbox`
 
-`dsh-sandbox` 负责定义词汇和 `SandboxProvider` 契约：`confine(argv, policy)` 返回调用方应当 spawn 的替代 argv（经过包装，使进程及其所有子进程在约束下运行），加上所选后端达到的 `enforcement` 完整度、其拒绝方言（`denialSignatures`，该后端内核在拒绝文件操作时打印到 stderr 的子串）、以及其 runner 失败方言（`runnerFailureSignatures`，runner 本身失败——因而命令从未运行——时的自我标识方式）；没有可用后端时抛出失败关闭的 `SANDBOX_UNAVAILABLE` 错误，绝不静默放行。词汇：`SandboxMode`（`read-only` / `workspace-write` / `danger-full-access`，仅限文件操作——不声称覆盖网络和进程可见性）、`SandboxEnforcement`（`full` / `partial`）、`SandboxExecutionPolicy`（每次能力调用的完整 mode + 工作区根目录）以及 `SandboxPolicy`（提供给约束后端的子集）。
+`dsh-sandbox` 负责定义词汇和 `SandboxProvider` 契约：`confine(argv, policy)` 返回调用方应当 spawn 的替代 argv（经过包装，使进程及其所有子进程在约束下运行），加上所选后端达到的 `enforcement` 完整度、其拒绝方言（`denialSignatures`，该后端内核在拒绝文件操作时打印到 stderr 的子串），以及其结构化 runner 失败证据（`runnerFailureRules`，可选的允许退出码加上排除整行精确信息性行后按行匹配的致命签名）；没有可用后端时抛出失败关闭的 `SANDBOX_UNAVAILABLE` 错误，绝不静默放行。词汇：`SandboxMode`（`read-only` / `workspace-write` / `danger-full-access`，仅限文件操作——不声称覆盖网络和进程可见性）、`SandboxEnforcement`（`full` / `partial`）、`SandboxExecutionPolicy`（每次能力调用的完整 mode + 工作区根目录）以及 `SandboxPolicy`（提供给约束后端的子集）。
 
 策略随每次调用而非提供方携带：两个消费方可以在同一时刻以不同策略约束（bash 在 `read-only` 下运行，而一个受约束的子 agent 保持其状态目录可写），且经批准的升级重试是一次带有更宽策略的新调用——在配置固定的提供方模式下无法表达。
 
@@ -60,17 +60,17 @@ OS 子进程约束适用于 bash 执行器（包括钩子命令），后续还
 
 #### 本地后端与随附 launcher
 
-`dsh-sandbox-local` 在提供方生命周期内选择一个平台 runner 并缓存结论。Linux 功能性探测 `bwrap` 然后 Landlock；macOS 使用 Seatbelt。不支持的平台和不可用的 runner 失败关闭。每次包装携带后端特定的拒绝签名和 runner 失败签名，以便 `dsh-bash-sandbox` 区分被拒绝的文件操作与损坏的沙箱。`runnerCommand` 作为运维人员对 bwrap 形式 runner 的断言跳过选择，但缺失或不可执行的命令仍被归类为沙箱失败，绝不无约束地运行负载。
+`dsh-sandbox-local` 在提供方生命周期内选择一个平台 runner 并缓存结论。Linux 功能性探测 `bwrap` 然后 Landlock；macOS 使用 Seatbelt。不支持的平台和不可用的 runner 失败关闭。每次包装携带后端特定的拒绝签名和 runner 失败规则，以便 `dsh-bash-sandbox` 区分被拒绝的文件操作与损坏的沙箱。Landlock runner 失败需要退出码 125，加上一行不是通知的 `landlock-run:` 诊断；即使子进程以 1、2 或 125 退出，精确匹配的部分强制执行通知仍只是信息。Bubblewrap 和 Seatbelt 仍仅依据签名，因为两者的公开契约均未保留 launcher 失败状态。`runnerCommand` 保留面向运维人员的 `runnerFailureSignatures` 配置，要求其中条目均为非空单行，并将它们映射为一条内部致命规则。消费方会直接 spawn 每个返回的 argv，因此缺失的 runner、不可执行的 runner，或 shebang 解释器不可用的可执行脚本会通过可归因的 `ENOENT`／`EACCES` spawn 通道失败，而成功启动的子进程以 126 或 127 退出时仍按普通结果处理。运维人员配置的脚本必然要先启动解释器，再应用自身 profile。
 
-launcher 是一个约 300 行的 C 程序（纯 C11，直接使用 Landlock UAPI——除静态链接的 musl 外无其他库，因此审计面仅为该文件加内核的稳定 syscall 契约）：`--ro <path>` / `--rw <path>` 授权，`--`，被包装的 argv；它为自身安装规则集并执行 `exec`（规则集跨 `execve` 继承，且它在限制前设置 `no_new_privs`）；`--probe` 在一个短生命周期子进程中强制最大规则集，仅当内核确实强制时才以 0 退出；launcher 失败以 125 退出且不 exec。
+launcher 是一个约 300 行的 C 程序（纯 C11，直接使用 Landlock UAPI——除静态链接的 musl 外无其他库，因此审计面仅为该文件加内核的稳定 syscall 契约）：`--ro <path>` / `--rw <path>` 授权，`--`，被包装的 argv；它为自身安装规则集并执行 `exec`（规则集跨 `execve` 继承，且它在限制前设置 `no_new_privs`）；`--probe` 在一个短生命周期子进程中强制最大规则集，仅当内核确实强制时才以 0 退出；所有 launcher 失败都会以 125 退出且不运行子进程，并打印一行致命的 `landlock-run:` 诊断。成功完成 exec 的子进程也可能返回 125，因此仅凭退出状态不能作为 launcher 失败的证据。较旧的 ABI 会在执行子进程之前打印精确的 `landlock-run: partial enforcement (older Landlock ABI)` 通知，因此该行不是致命证据。
 
-Landlock launcher 源码和包工作区位于 `native/landlock-run`，与 harness 消费方同仓。独立的 [`node-addon-landlock-run`](https://github.com/deepseek-harness/node-addon-landlock-run) 仓库是用于打包并发布 npm 包族的发布镜像；导出流程归 `native/README.md` 所有。平台二进制由 npm 选择，入口包拥有路径解析、探测和 CLI 参数，而 harness 将沙箱模式映射为授权。将入口点与其二进制一起版本化，使探测解析和启动语法保持对齐。
+Landlock launcher 源码和包工作区位于 `native/landlock-run`，与 harness 消费方同仓。独立的 [`node-addon-landlock-run`](https://github.com/deepseek-harness/node-addon-landlock-run) 仓库是用于打包并发布 npm 包族的发布镜像；导出流程归 `native/README.md` 所有。平台二进制由 npm 选择，入口包拥有路径解析、探测、CLI 参数、致命前缀和部分强制执行通知，而 harness 将沙箱模式映射为授权。将入口点与其二进制一起版本化，使探测解析和启动语法保持对齐。
 
 后端 profile 共享模式契约但在必要的主机授权上有所不同。Landlock 和 Seatbelt 在 read-only 模式下仅允许 `/dev/null`；workspace-write 还允许各自所需的主机临时目录根。每次包装携带后端特定的拒绝签名。Landlock 在较旧的 ABI 无法管控所有操作时报告 partial enforcement，而成功的 bwrap 和 Seatbelt profile 报告 full enforcement。
 
 #### bash 消费方
 
-`dsh-bash-sandbox` 扩展 `LocalBashExecutor`，并把即将 spawn 的确切 `['bash', '-c', command]` argv 交给 `ctx.sandbox`。拒绝是与其他结果正交的事实，依据当前 runner 的 stderr 方言保守分类。Runner 失败优先于拒绝：前台执行抛出 `SANDBOX_UNAVAILABLE`；结算后的 `BashProcess` 会盖章 `sandbox.runnerFailed`，bash 生产者再通过通用 `task_output` 渲染它。
+`dsh-bash-sandbox` 扩展 `LocalBashExecutor`，把精确的 `['bash', '-c', command]` argv 交给 `ctx.sandbox`，并直接 spawn 提供方返回的 argv。这样，随附的原生 runner 建立约束后，shell 语义与 `BASH_ENV` 仍由内层 Bash 处理。提供方错误原样传播。进程启动前，只有当调用方拥有的 workdir 经独立验证可用，并且 Node 报告 `ENOENT` 或 `EACCES`，且带有明确指向提供方 argv[0] 的来源信息时，拒绝才会归因于 runner；没有精确错误路径的裸 `syscall: 'spawn'`、其他错误码、无效 workdir、资源失败、无关 syscall 与无结构拒绝保留本地命令启动语义。前台执行会将可归因于 runner 的拒绝转为 `SANDBOX_UNAVAILABLE` 并附上原始详细信息；异步后台拒绝则盖章 `runnerFailed: true`、`denied: false`。如果 `SubprocessService` 同步抛出同样带有来源信息的 `ENOENT`／`EACCES` 形态，后台启动会抛出 `SANDBOX_UNAVAILABLE`；其他同步错误原样传播。进程启动后，前台与后台共用一个 runner 失败分类器：先排除信息性行，再要求规则的退出码门控与余下的一行致命诊断同时匹配。匹配结果优先于拒绝：前台执行抛出 `SANDBOX_UNAVAILABLE`，并以该致命行作为详细信息；结算后的 `BashProcess` 会盖章 `sandbox.runnerFailed`，bash 生产者再通过通用 `task_output` 渲染它。
 
 模型会在归属方派生的 `sandbox:policy` 运行时上下文中看到当前有效的文件策略；静态工具描述则解释拒绝标记（`[sandbox: file access denied under <mode> mode]`），鼓励尝试可能被拒绝的命令，并禁止绕过拒绝重试。当升级字段被公布时，被拒绝的结果还会携带升级提示本身，使按规定进行的同轮次重试在决策点得到提示，而非依赖模型回忆描述（§ 升级机制）。[当前策略决策](2026-07-30-current-sandbox-policy-context.md)负责该上下文的理由与边界。
 
@@ -78,7 +78,7 @@ Landlock launcher 源码和包工作区位于 `native/landlock-run`，与 harnes
 
 `BashExecRequest.sandboxPolicy` 是可选的完整按调用输入；解析后的 spec 使该字段显式。`BashExecutor.sandboxMode` 仍是公布已挂载执行器能否兑现该策略的能力事实，因此只有约束组合才暴露升级。seam 接受任何显式策略；工具拥有会话解析和「仅更宽」的升级规则。非沙箱执行器诚实地保持无约束。
 
-`ctx.sandboxPolicy.resolve()` 在执行器运行前盖章完整执行策略——显式升级模式 > 会话覆盖 > 配置默认值，且 `SessionHeader.cwd` > 配置的后备根目录。`SandboxBashExecutor.resolve()` 在 spec 上保留该策略，或为直接的无 agent 调用方提供部署后备值，使 `run()`/`start()` 永不读取可变会话状态。每进程包装事实以返回的 `BashProcess` 为键；`onProcessDone()` 在 `done` 结算前分类 stderr 并给该句柄盖章，因此重叠进程各自保留自己的模式和 runner 方言。
+`ctx.sandboxPolicy.resolve()` 在执行器运行前盖章完整执行策略——显式升级模式 > 会话覆盖 > 配置默认值，且 `SessionHeader.cwd` > 配置的后备根目录。`SandboxBashExecutor.resolve()` 在 spec 上保留该策略，或为直接的无 agent 调用方提供部署后备值，使 `run()`/`start()` 永不读取可变会话状态。每进程包装事实以返回的 `BashProcess` 为键；`onProcessDone()` 会通过 stderr 分类之外的通道接收 spawn 失败，并在 `done` 结算前给该句柄盖章，因此重叠进程各自保留自己的模式和 runner 方言。
 
 当约束执行器被挂载时，`bash` 公布配对的 `sandbox_permissions` 和 `justification` 字段。schema 暴露完整的封闭升级词汇，因为有效模式是按会话的；执行拒绝任何不严格宽于该调用有效模式的目标。批准在执行之前解析。`allowed-once` 仅将授权模式盖章到该请求上，而 `rejected`、`cancelled`、`unavailable`、缺失的 approval 服务或缺失的 agent 都以各自不同的结果文本失败关闭。授权不持久化。
 
@@ -117,10 +117,10 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
 
 ### 测试
 
-- **单元测试：** 固定平台选择和 profile、失败关闭的 runner 分类、按调用的模式/根目录解析、按进程事实、升级验证和结果、权限 preset fold 和写入透传、以及运行时上下文排序与具体化。
+- **单元测试：** 固定平台选择和 profile、直接交接提供方返回的 argv、带有无效 workdir 对照的 spawn 层失败、runner 缺失／不可执行／解释器缺失证据、格式错误 runner 阴性对照、受约束的 `BASH_ENV` 求值顺序、结构化 runner 分类（包括只有部分强制执行通知的子进程结果、带门控的致命证据、子进程退出码 126／127，以及前台／后台一致性）、按调用的模式/根目录解析、按进程事实、升级验证和结果、权限 preset fold 和写入透传，以及运行时上下文排序与具体化。
 - **Keyless 真实 runner：** 在提供方和 bash 消费方层面对 bwrap、Landlock 和 Seatbelt 执行真实文件系统效果测试；一个真实 Cordis 上下文通过已交付的 bash 和 fs 工具并发驱动两个项目会话，证明在自身根目录写入成功、在兄弟根目录写入被拒绝。打包安装测试证明注册表 launcher 保持可执行。CI 拒绝静默全跳过。
 - **With-key：** 以只读模式启动真实 ACP 组合，让模型驱动的 bash 写入命中 runner 的拒绝标记，再通过已授权与被拒绝的 workspace-write 重试驱动 bridge 应答器和磁盘效果；不可用的凭证或 runner 自动跳过。
-- **快照：** 固定原子化的当前策略上下文和两个脚本化的 approval 分支。一个真实 ACP 示例场景把会话放在用户主目录下，同时让部署后备根目录指向 `/tmp`，然后固定 workspace-write 运行时上下文消息与一次成功的、由部署选定的变更；这能区分会话根目录解析与进程后备值，而不依赖 runner 特定的拒绝文本。其他快照以无约束启动，使无关 fixture（测试前置数据）保持平台无关。
+- **快照：** 固定原子化的当前策略上下文和两个脚本化的 approval 分支。一个真实 ACP 示例场景把会话放在用户主目录下，同时让部署后备根目录指向 `/tmp`，然后固定 workspace-write 运行时上下文消息与一次成功的、由部署选定的变更；这能区分会话根目录解析与进程后备值，而不依赖 runner 特定的拒绝文本。一个模拟 Landlock 部分强制执行行为的 POSIX 提供方会在组装后的应用中固定直接执行 bash `false` 时仍得到普通子进程结果，并固定提供方可执行文件缺失时在前台／后台均为基础设施失败。其他快照以无约束启动，使无关 fixture（测试前置数据）保持平台无关。
 
 ## 延迟阶段
 
@@ -128,12 +128,12 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
 
 - **第二个消费方**——`subagent-acp` 可选地约束子 agent（按调用策略；默认无约束——子 agent 必须写入自己的持久化）。
 - **更多环境**——环境一致的能力组示例（如 bash+fs 对一个容器）。
-- **Windows 链**——`PLATFORM_CHAINS.win32` 保留为空（失败关闭）；填充它意味着来自 AppContainer/restricted-token 家族的约束 runner，从其自己的仓库按 `node-addon-landlock-run` 模板交付，加上其 profile 方言和拒绝/runner 失败签名。改为包装第三方 landstrip runner 的方案[经考虑后已驳回](../../rejected/feature/2026-07-26-evaluate-landstrip-for-windows-sandbox-rung.md)——它所经受的实战检验还不足以承载安全不变式。
+- **Windows 链**——`PLATFORM_CHAINS.win32` 保留为空（失败关闭）；填充它意味着来自 AppContainer/restricted-token 家族的约束 runner，从其自己的仓库按 `node-addon-landlock-run` 模板交付，加上其 profile 方言、拒绝签名和 runner 失败规则。改为包装第三方 landstrip runner 的方案[经考虑后已驳回](../../rejected/feature/2026-07-26-evaluate-landstrip-for-windows-sandbox-rung.md)——它所经受的实战检验还不足以承载安全不变式。
 
 ## 曾考虑的替代方案
 
 - **命令字符串启发式预检**：否决。无法理解展开/子进程/符号链接；严格尝试（运行它，让内核决定）是唯一可信的拒绝信号。
-- **即使平台仅有一个后端也功能性探测**：否决。探测用于在候选者之间仲裁；只有一个时无需决策，且探测开销对每个会话的首次约束命令征税（对未来重量级后端而言代价过高）。runner 自身执行时的失败关闭拒绝加 `runnerFailureSignatures` 分类承载了安全属性。
+- **即使平台仅有一个后端也功能性探测**：否决。探测用于在候选者之间仲裁；只有一个时无需决策，且探测开销对每个会话的首次约束命令征税（对未来重量级后端而言代价过高）。runner 自身执行时的失败关闭拒绝加结构化 `runnerFailureRules` 分类承载了安全属性。
 - **提交构建好的 launcher 二进制**：否决。diff 中的二进制不可审查且膨胀历史；经审查的源码 + 原生 CI 构建 + launcher 仓库的字节固定发布演练使二进制远离所有代码树。
 - **安装时编译 launcher**：否决。将 C 工具链强加给每个消费方；仅在碰巧有编译器时才存在的备选不是备选。
 - **从一个构建器交叉编译两种架构**：否决。仅为重建两个约 70 KB 的二进制就需要携带一个固定的交叉工具链（rustup targets、zig 或容器镜像）；每架构的原生 runner 已存在，各自构建自己的平台包（`node-addon-require-builtin` 模式，launcher 仓库自己的流水线）。
@@ -172,19 +172,20 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
 - **单一包装的幻觉被有意放弃。**`tools/pre-execute` 包装加提示词约定无法解决沙箱批准——正确的设计需要结构化拒绝、原生 runner 探测、按调用策略承载和一致的跨工具族强制，本设计为此付出了代价。
 - **`read-only` 通过后续设计成为跨工具族边界。** 本 Agent Note 最初只交付 bash 强制；[跨工具族 fs 沙箱 Agent Note](2026-07-14-cross-family-fs-sandbox.md) 通过沙箱化的 `ctx.fs` 提供方把同一模式词汇扩展到文件系统工具，并将 mode/root 配置和 `sandbox/mode` 覆盖迁移到 `ctx.sandboxPolicy`（§ 进程内工具）。
 - **Windows 没有后端。** 其链槽保留为空——失败关闭，绝不穿透；填充它是延迟阶段。
-- **Seatbelt 层级依赖 Apple 已弃用但仍交付的 `sandbox-exec` CLI。** 作为 darwin 的唯一候选，它无需探测即被选中，因此未来移除会在执行时作为 runner 失败分类浮现——重新抛出 `SANDBOX_UNAVAILABLE`，命令从未运行；失败关闭，绝不开放。
+- **Seatbelt 层级依赖 Apple 已弃用但仍交付的 `sandbox-exec` CLI。** 作为 darwin 的唯一候选，它无需探测即被选中，因此在 workdir 可用时，未来移除会表现为可归因于 runner 的 spawn 失败，可执行文件拒绝则通过其致命签名体现——两者都会变为 `SANDBOX_UNAVAILABLE`，且命令绝不会运行；失败关闭，绝不开放。
 - **Landlock 约束的完整度取决于运行内核的 ABI。** 报告为 `enforcement: 'partial'` 而非拒绝——这是有意的权衡，使备选在旧内核主机上仍可用。
+- **Runner 归因使用带内协议。** 退出状态与 stderr 无法以密码学方式识别写入者，因此受限子进程可以模仿 runner 的致命诊断行和状态，造成可用性或诊断误归因。多项证据的合取与精确通知排除减少了意外匹配；这不是沙箱绕过，因为子进程已经受到限制。
 - **launcher 作为注册表依赖到达。** 通过其自身仓库的发布流水线（经审查的 C 源码、原生 CI 构建器、字节固定的发布演练）加上本仓库的版本固定获得信任——真实内核 e2e 测试环节会验证安装产物的实际行为。
 - **模型可能过度请求。** 在没有拒绝依据的情况下升级，或在 `workspace-write` 足够时选择 `danger-full-access`：描述引导且枚举强制阶梯，但人的提示词是实际门控；`approval/asked` 原因使过度请求可审计，且 `prepend` 策略应答器可以自动拒绝部署永远不想要的模式。
 - **公布的目标集是静态的，而有效模式是按会话的**（schema 是注册表全局的）——已处于最宽模式的会话仍被提供这些字段。构造上无害：执行时的严格放宽检查（而非枚举）是安全边界——非放宽请求以自身文本失败且不提示任何人。
-- **授权的升级不等于可工作的沙箱。** 不可用的后端即使对授权升级到约束模式也仍然失败关闭——在平台没有链或所有探测失败时于 `confine()` 阶段，在未探测的唯一 runner 拒绝时于执行阶段（归类为沙箱失败而非命令失败）——而授权的 `danger-full-access` 运行根本不触及提供方：此时授权（而非探测）是权威。
+- **授权的升级不等于可工作的沙箱。** 不可用的后端即使对授权升级到约束模式也仍然失败关闭——平台没有链或所有探测失败时在 `confine()` 阶段失败，所选可执行文件无法启动时通过 spawn 通道失败，已启动的 runner 拒绝时则通过结构化规则失败——而授权的 `danger-full-access` 运行根本不触及提供方：此时授权（而非探测）是权威。
 - **运行时上下文历史仅追加。** 策略切换会在保留的历史之后追加一份用于取代先前快照的完整快照，从而保留稳定的系统与对话前缀；状态不变时不添加消息。
 - **旧策略快照仍保留在历史中。** 每份完整快照都会明确取代更早的运行时上下文快照，因此回放与压缩（compaction）只需保留最新具体化的消息。
 
 ## FAQ
 
 - **一个命令返回了 `[sandbox: file access denied under read-only mode]`——它失败了吗？** 它运行了，内核拒绝了一个文件操作：拒绝是与退出码正交的结果事实。相关指令禁止通过绕过限制来重试；唯一被认可的动作是以升级请求重试同一命令一次。
-- **如何区分损坏的沙箱与失败的命令？** Runner 失败在分类中优先于拒绝：匹配包装的 `runnerFailureSignatures` 的失败运行意味着命令从未运行——前台重新抛出结构化的 `SANDBOX_UNAVAILABLE` 并附带 runner 的 stderr 行，后台任务盖章 `sandbox.runnerFailed` 并渲染自己的标记。损坏的沙箱永远不会被误判为命令执行失败，且命令永远不会无约束运行。
+- **如何区分损坏的沙箱与失败的命令？** 提供方 argv 的任何 spawn 拒绝都能证明受限启动从未开始，但只有在调用方拥有的 workdir 可用，且 Node 为该 argv[0] 报告可归因的 `ENOENT` 或 `EACCES` 时，才能据此判定 runner 损坏。没有精确错误路径的裸 `syscall: 'spawn'` 和其他所有拒绝仍是普通的命令启动错误。进程启动后，只有当 `runnerFailureRules` 中某一条目同时匹配其可选退出码门控，以及排除整行精确信息性行后的一行致命 stderr 诊断时，runner 失败才会优先于拒绝。前台失败会抛出结构化的 `SANDBOX_UNAVAILABLE`，并附带 spawn 错误或匹配行作为详细信息；遭异步拒绝或已结算的后台任务则盖章 `sandbox.runnerFailed` 并渲染自己的标记。如果 `SubprocessService` 同步抛出同样带有来源信息的 `ENOENT`／`EACCES` 形态，后台启动会抛出该结构化错误；其他同步错误原样传播。Landlock 部分强制执行通知加上普通子进程失败时，仍返回命令结果。
 - **在没有后端的平台上会发生什么——今天的 Windows？** `confine()` 抛出失败关闭的 `SANDBOX_UNAVAILABLE`，命令永不 spawn；`win32` 是保留的空链，由测试固定为同样失败关闭，直到 Windows runner 填充它（§ 延迟阶段）。
 - **`bwrap` 已安装在我的主机上但不可用（禁用了非特权 userns、LSM 拒绝 `mount`）——会发生什么？** 链探测是功能性的——它构建并强制一个真实 profile 而非检查 `--version`——因此存在但不可用的 `bwrap` 探测失败，选择落到注册表安装的 Landlock launcher，结论在提供方生命周期内缓存。
 - **沙箱限制网络或进程可见性吗？** 不——`SandboxMode` 仅声称文件操作；bwrap profile 刻意不 unshare pid，没有后端声称网络。网络限制是否成为自己的旋钮留在 § seam 中开放。

.agents/notes/implemented/feature/2026-07-24-web-session-model-selector.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-24-web-session-model-selector.md
-2026-07-24-web-session-model-selector.md: 003c0b5ac1c6963701e1d93e4c3ff8615fc45dd5
-2026-07-24-web-session-model-selector.zh.md: a8e74fb1539348535f89f49d7c6d56b2bab79278
+2026-07-24-web-session-model-selector.md: 05e923fb3b5df72485ffb8d6ff5aa0104b19a8a0
+2026-07-24-web-session-model-selector.zh.md: bdfa2f856a799cc86600650829a2df62a7dfdc18

.agents/notes/implemented/feature/2026-07-24-web-session-model-selector.md

@@ -12,11 +12,11 @@ The Web conversation displayed and sent through the Host's fixed provider/model
 
 The Web Host reuses `installAgentLlmTarget` for every created or resumed agent. The provider/model/reasoning target starts from the latest `request/header` when the session has used a model, otherwise from the Host default route. `session.selectModel` changes the session-local mutable target, and prompt assembly captures it with request routing; a switch during a running step therefore applies to the next assembled step. The next consumed target persists through the existing full `request/header` snapshot, while a choice that has not reached a request remains process-local.
 
-The session RPC domain exposes a `session.models` directory and `session.selectModel`. The directory is built dynamically from the LLM registry and grouped by provider; each listed model's exact metadata adds adapter-owned reasoning effort ids, names, descriptions, and optional default. Provider catalogs and exact metadata load concurrently by provider and fail independently, so successful groups remain usable alongside retryable failure records. Catalog membership stays advisory: the current model is inserted as an unlisted row when its registered provider omits it, while exact resolution decides whether a route and explicit effort are available. Selection uses `resolveCallConfig` to reject unsupported effort ids and materialize an adapter-configured default before updating the target.
+The session RPC domain exposes a `session.models` directory and `session.selectModel`. The directory is built dynamically from the LLM registry and grouped by provider; each listed model's exact metadata adds adapter-owned reasoning effort ids, names, descriptions, and optional default. Provider catalogs and exact metadata load concurrently by provider and fail independently, so successful groups remain usable alongside retryable failure records. Catalog membership stays advisory: `session.models.current` is returned independently and can remain routable when absent from every group, but the Host does not synthesize an unlisted row after its provider stops advertising it. The two surfaces answer that state differently on purpose: the TUI still renders the unlisted current model as its own row and marks it current, while Web shows the unset trigger label and asks for a replacement. Web is the surface where a catalog is edited, so a target the user just deleted should read as a decision to make rather than a selection to keep; the TUI, which only picks from what exists, has no such edit to reconcile. The cost is real and accepted — a Web composer showing the unset label can still send to the routed target — and the divergence is deliberate, not a missed migration. Exact resolution decides whether a route and explicit effort are available. Selection uses `resolveCallConfig` to reject unsupported effort ids and materialize an adapter-configured default before updating the target.
 
 The browser `ModelService` owns one `ModelDirectory` per live session. Its snapshot contains the current complete target, grouped catalog, provider failures, operation error, and `idle`/`loading`/`ready`/`selecting`/`error` state. Mounting primes the trigger label and each menu open refreshes the directory. Directory and selection calls share an operation generation so older responses cannot replace a newer result; connection reset discards the process-local projection before restoring the Host target. Failures retain the previous current target and usable groups.
 
-`@deepseek-ai/dsh-client-ui-conversation` declares the session-scoped single slot `conversation.input.model` as a child of its composer-bar entry. InputBar renders the seat in its trailing controls immediately before the pending indicator and primary button; the seat receives the bar's `locked` owner prop and session scope. `@deepseek-ai/dsh-client-ui-model` occupies that seat and also contributes `/model` over the same directory. Its compact trigger displays the catalog model name and effective reasoning label, falling back to ids when metadata is absent. The upward menu first offers Model and, when the current exact model supports it, Effort; Model drills into provider groups, while Effort drills into the adapter-ordered levels. The provider-default row appears only when the adapter does not configure a model default.
+`@deepseek-ai/dsh-client-ui-conversation` declares the session-scoped single slot `conversation.input.model` as a child of its composer-bar entry. InputBar renders the seat in its trailing controls immediately before the pending indicator and primary button; the seat receives the bar's `locked` owner prop and session scope. `@deepseek-ai/dsh-client-ui-model` occupies that seat and also contributes `/model` over the same directory. Its compact trigger displays the exact catalog model name and effective reasoning label. When the current target is absent from the groups, the trigger instead displays `Select model`, the model list marks no row active, and the Effort row stays absent; choosing a listed model replaces the complete target through the existing selection path. The upward menu otherwise first offers Model and Effort; Model drills into provider groups, while Effort drills into the adapter-ordered levels. The provider-default row appears only when the adapter does not configure a model default.
 
 The production browser roster is assembled from `apps/cli/config/base.cordis.yml` plus `apps/cli/config/web.cordis.yml`; the model feature is one `dshClient` row rather than a package hardcoded in Web boot code. Its package manifest orders it after the runtime and command feature, while Cordis service injection waits for the conversation slot before registering the composer occupant.
 
@@ -40,4 +40,4 @@ Any Host-backed Web conversation, including a blank session, can switch among dy
 
 ## Testing
 
-Host tests pin grouped discovery, catalog and exact-metadata failure isolation, logged effort restoration, unlisted current targets, unsupported effort rejection, default materialization, and next-assembly switching. Client tests pin the shared directory, reconnect restoration, and complete-target submission. Component tests pin dynamic effort labels, descriptions, provider-default exposure, and effort submission. The keyless built-app fixture loads the production model plugin, selects OpenAI's GPT-5 and its Max effort, sends a turn, and verifies that the next generated response reports both ids.
+Host tests pin grouped discovery, catalog and exact-metadata failure isolation, logged effort restoration without stale-row injection, advisory unlisted selection, unsupported effort rejection, default materialization, and next-assembly switching. Client tests pin the shared directory, reconnect restoration, and complete-target submission. Component tests pin dynamic effort labels, descriptions, provider-default exposure, effort submission, and the `Select model` fallback for a removed row. The keyless built-app fixture loads the production model plugin, selects OpenAI's GPT-5 and its Max effort, sends a turn, and verifies that the next generated response reports both ids; the DeepSeek configuration fixture removes the active catalog row and pins the fallback before choosing a replacement.

.agents/notes/implemented/feature/2026-07-24-web-session-model-selector.zh.md

@@ -12,11 +12,11 @@ Web 对话原本通过 Host 固定的提供方与模型路由显示并发送消
 
 Web Host 为每个新建或恢复的 agent（智能体）复用 `installAgentLlmTarget`。如果会话已经使用过模型，提供方／模型／推理（reasoning）目标从最新的 `request/header` 开始；否则采用 Host 默认路由。`session.selectModel` 会更改会话级可变目标，提示词组装则将该目标与请求路由一并捕获，因此运行中步骤发生的切换会应用于下一个组装步骤。下一个实际采用的目标通过现有的完整 `request/header` 快照持久化；尚未进入请求的选择则仅保存在当前进程中。
 
-会话 RPC 领域公开 `session.models` 模型目录与 `session.selectModel`。该目录从 LLM（大语言模型）注册表动态构建，并按提供方分组；每个已列出模型的精确元数据还会加入由适配器持有的推理强度 ID、名称、说明和可选默认值。各提供方的目录与精确元数据会按提供方并发加载，且彼此独立失败，因此成功加载的分组仍可与可重试的失败记录一同使用。模型是否位于目录仅供参考：如果当前模型的已注册提供方没有列出该模型，系统会将其作为未列出行插入；精确解析则决定路由与显式推理强度是否可用。选择操作通过 `resolveCallConfig` 拒绝不支持的推理强度 ID，并在更新目标前具体化适配器配置的默认值。
+会话 RPC 领域公开 `session.models` 模型目录与 `session.selectModel`。该目录从 LLM（大语言模型）注册表动态构建，并按提供方分组；每个已列出模型的精确元数据还会加入由适配器持有的推理强度 ID、名称、说明和可选默认值。各提供方的目录与精确元数据会按提供方并发加载，且彼此独立失败，因此成功加载的分组仍可与可重试的失败记录一同使用。模型是否位于目录仅供参考：`session.models.current` 独立返回，即使不在任何分组中也仍然可以路由，但提供方停止公布该模型后，Host 不会合成未列出行。两个前门有意对这一状态给出不同回答：TUI 仍把未列出的当前模型渲染为独立一行并标记为当前，Web 则显示未设置状态的触发器标签并要求选择替代模型。Web 正是编辑目录所在的前门，因此用户刚刚删除的目标应当读作一个有待作出的决定，而不是一项可以保留的选择；TUI 只在已存在的模型中挑选，没有这类编辑需要调和。这一代价真实存在且已被接受——显示未设置标签的 Web composer 仍会发送到实际路由的目标——这一分歧是有意为之，而不是一处遗漏的迁移。精确解析决定路由与显式推理强度是否可用。选择操作通过 `resolveCallConfig` 拒绝不支持的推理强度 ID，并在更新目标前具体化适配器配置的默认值。
 
 浏览器中的 `ModelService` 为每个实时会话持有一个 `ModelDirectory`。其快照包含当前完整目标、分组目录、提供方失败记录、操作错误，以及 `idle`、`loading`、`ready`、`selecting`、`error` 状态。挂载时会预先填充触发器标签，此后每次打开菜单都会刷新目录。目录与选择调用共用操作代次，防止较早响应覆盖较新结果；连接重置会先丢弃当前进程中的投影，再恢复 Host 目标。失败时保留先前的当前目标和可用分组。
 
-`@deepseek-ai/dsh-client-ui-conversation` 将会话作用域的单实例 slot `conversation.input.model` 声明为其输入栏 entry 的子 slot。InputBar 在尾部控件区将该 seat 渲染于 pending 指示器与主按钮之前；该 seat 接收输入栏的 `locked` owner prop 与会话作用域。`@deepseek-ai/dsh-client-ui-model` 占用该 seat，并在同一目录上提供 `/model`。其紧凑型触发器显示目录中的模型名称与生效的推理强度标签；元数据缺失时则回退到相应 ID。向上展开的菜单首先提供 Model，并在当前精确模型支持时提供 Effort；Model 可深入提供方分组，Effort 可深入适配器排序的级别。仅当适配器没有配置模型默认值时，才显示提供方默认值行。
+`@deepseek-ai/dsh-client-ui-conversation` 将会话作用域的单实例 slot `conversation.input.model` 声明为其输入栏 entry 的子 slot。InputBar 在尾部控件区将该 seat 渲染于 pending 指示器与主按钮之前；该 seat 接收输入栏的 `locked` owner prop 与会话作用域。`@deepseek-ai/dsh-client-ui-model` 占用该 seat，并在同一目录上提供 `/model`。其紧凑型触发器显示目录中精确模型的名称与生效的推理强度标签。当前目标不在分组中时，触发器改为显示 `Select model`，模型列表不标记任何活动行，Effort 行也保持隐藏；选择一个已列出的模型，会通过现有选择路径替换完整目标。除此情形外，向上展开的菜单会首先提供 Model 与 Effort；Model 可深入提供方分组，Effort 可深入适配器排序的级别。仅当适配器没有配置模型默认值时，才显示提供方默认值行。
 
 生产环境的浏览器名册由 `apps/cli/config/base.cordis.yml` 与 `apps/cli/config/web.cordis.yml` 共同组装；模型功能对应其中一行 `dshClient` 配置项，而不是 Web boot 代码中硬编码的包。其包 manifest（元数据清单）将加载顺序置于运行时与命令功能之后；Cordis 服务注入则等待 conversation slot 可用，再注册 composer 占用方。
 
@@ -40,4 +40,4 @@ Web Host 为每个新建或恢复的 agent（智能体）复用 `installAgentLlm
 
 ## 测试
 
-Host 测试固定分组发现、目录与精确元数据失败隔离、已记录推理强度恢复、当前未列出目标、不支持的推理强度拒绝、默认值具体化，以及切换仅影响下一次组装。客户端测试固定共享目录、重连恢复与完整目标提交。组件测试固定动态推理强度标签、说明、提供方默认值展示与推理强度提交。无密钥 built-app fixture（测试前置数据）加载生产模型插件，选择 OpenAI 的 GPT-5 及其 Max 推理强度，发起一个轮次，并验证下一条生成的响应会报告两个 ID。
+Host 测试固定分组发现、目录与精确元数据失败隔离、已记录推理强度恢复且不注入陈旧行、建议性的未列出模型选择、不支持的推理强度拒绝、默认值具体化，以及切换仅影响下一次组装。客户端测试固定共享目录、重连恢复与完整目标提交。组件测试固定动态推理强度标签、说明、提供方默认值展示、推理强度提交，以及已删除模型行的 `Select model` 回退。无密钥 built-app fixture（测试前置数据）加载生产模型插件，选择 OpenAI 的 GPT-5 及其 Max 推理强度，发起一个轮次，并验证下一条生成的响应会报告两个 ID；DeepSeek 配置 fixture 会删除活动目录行，在选择替代模型之前固定该回退。

.agents/notes/implemented/process/2026-07-04-doc-tiers-and-budgets.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write .agents/notes/implemented/process/2026-07-04-doc-tiers-and-budgets.md
-2026-07-04-doc-tiers-and-budgets.md: a52c40a9a147fd39fdec4c61079822f1b1115227
-2026-07-04-doc-tiers-and-budgets.zh.md: c26efe0e06d73ff0bd5608c7f44d54ebbe373bed
+2026-07-04-doc-tiers-and-budgets.md: e7b3421d09a1ae5ab9a9373e8040832c1b0d4b97
+2026-07-04-doc-tiers-and-budgets.zh.md: 18bf98777c87512e6161a4529db66030ced47bfd

.agents/notes/implemented/process/2026-07-04-doc-tiers-and-budgets.md

@@ -1,4 +1,4 @@
-# Agent Note: Documentation tiers, budgets, and the ceiling gate
+# Agent Note: Documentation structure, tiers, and budgets
 
 Status: implemented
 
@@ -6,13 +6,14 @@ English | [中文](2026-07-04-doc-tiers-and-budgets.zh.md)
 
 ## Problem
 
-Standing docs accumulated repeated rules, retold incidents, duplicated package maps, and stale Agent Note summaries despite existing writing guidance. Because review alone did not prevent that growth, the repository needed a mechanical budget alongside its documentation taxonomy.
+Standing docs accumulated repeated rules, retold incidents, duplicated package maps, and stale Agent Note summaries despite existing writing guidance. That guidance also did not define how a document's place in the hierarchy limits its scope or how ordered teaching differs from lookup-oriented material. Because review alone did not prevent that growth, the repository needed a mechanical budget alongside its documentation taxonomy.
 
 ## Decision
 
-- **A tier taxonomy with one home per fact.** [docs/AGENTS.md](../../../../docs/AGENTS.md) is the documentation standard: it assigns every Markdown tier a single job (standing orders, system map, type catalog, decision records, incident stories, how-tos, per-package contracts, generated catalogs, workflows), forbids restating a fact outside its home tier (link instead), and carries the slop checklist used when writing or reviewing any doc.
+- **Structure follows the documentation tree.** [docs/AGENTS.md](../../../../docs/AGENTS.md) is the documentation standard: a document owns detail about its subject, summarizes only the purpose, responsibility, and high-level behavior of direct children, and links to deeper owners. [Agent Notes](../../README.md) remain outside this structural contract. Every human-facing document is a tutorial with an ordered outcome or a reference with an explicit lookup scope; a [postmortem](../../../../docs/postmortem/README.md) is an incident-scoped reference whose chronology records evidence. Tutorials introduce concepts in prerequisite order for the reader's starting knowledge.
+- **A tier taxonomy with one home per fact.** The standard assigns every Markdown tier one job, forbids restating a fact outside its home tier, and carries the slop checklist used when writing or reviewing any doc.
 - **A narrow, hard budget gate.** [scripts/verify-doc-budgets.ts](../../../../scripts/verify-doc-budgets.ts) joins `doc-sync`: every doc listed in [scripts/doc-budgets.manifest.json](../../../../scripts/doc-budgets.manifest.json) must stay under its word ceiling (`wc -w` semantics, whole file), and a budgeted file that is missing fails the gate so a rename cannot silently orphan its budget. Scope is deliberately only the accretion-prone standing docs — the root and subtree `AGENTS.md` files, `architecture.md`, `packages/README.md`, and the standing policy docs they evict content into (`docs/testing.md`, `docs/defensive-patterns.md`). Reference docs, Agent Notes, and package READMEs are unbudgeted: length is legitimate there when every row is a fact, and review plus the slop checklist govern them.
-- **Ceilings are an enforcement frontier that ratchets.** A ceiling sits at least 5% above the doc's current size — working headroom, so routine wording edits pass while real growth still trips the gate — and ratchets down, keeping that margin, as the doc is brought to its target budget (root `AGENTS.md` ≤ 1,500 words; `architecture.md` ≤ 1,800; subtree `AGENTS.md` ≤ 600; `packages/README.md` ≤ 600). When the gate goes red the fix is to relocate or condense per the taxonomy; raising a ceiling is permitted only with explicit justification in the PR description, the manifest diff being the reviewable act.
+- **Ceilings are an enforcement frontier that ratchets.** A doc at or below its target keeps at least 5% headroom as its ceiling ratchets down; a doc above target keeps a frozen ceiling that prevents growth until it reaches the target (root `AGENTS.md` ≤ 1,600 words; `architecture.md` ≤ 1,800; subtree `AGENTS.md` ≤ 600 except `packages/AGENTS.md` ≤ 650 and `docs/AGENTS.md` ≤ 1,250; `packages/README.md` ≤ 600). When the gate goes red, relocate or condense; raise a ceiling only with explicit PR justification.
 - **A thin workflow skill, contracts in docs.** [.agents/skills/dsh-doc-standards](../../../skills/dsh-doc-standards/SKILL.md) carries the placement/audit/red-gate workflow and defers to the standard as its source of truth, the same split as [dsh-translate-docs](../../../skills/dsh-translate-docs/SKILL.md) over the i18n contract.
 
 ## Alternatives considered
@@ -23,6 +24,7 @@ Standing docs accumulated repeated rules, retold incidents, duplicated package m
 
 ## Consequences
 
-- Adding to a budgeted doc now requires displacement: relocate the addition to its taxonomy home with a pointer, or condense existing prose to pay for it. Growth without pruning fails CI.
-- The bring-under-target rewrites land as stacked follow-ups that ratchet the manifest down as they merge; until each lands, its doc's frozen ceiling only prevents further growth.
+- Adding to a budgeted doc requires displacement: relocate the addition to its taxonomy home with a pointer, or condense existing prose to pay for it. Growth without pruning fails CI.
+- Structural review starts with ownership and document form before sentence-level editing, so lower-level detail moves to its owner instead of being polished in the wrong place.
+- Budgeted docs that remain above target cannot grow; reaching the target restores the 5% working headroom.
 - Word count is a crude proxy accepted deliberately: it cannot judge quality, but it forces the relocation decision at exactly the moment content is being added, which is when the author has the context to place it correctly.

.agents/notes/implemented/process/2026-07-04-doc-tiers-and-budgets.zh.md

@@ -1,4 +1,4 @@
-# Agent Note: 文档分层、预算与上限门禁
+# Agent Note: 文档结构、层级与预算
 
 Status: implemented
 
@@ -6,14 +6,15 @@ Status: implemented
 
 ## 问题
 
-尽管已有写作指导，常设文档仍不断累积重复规则、反复讲述的事故、重复的包映射，以及陈旧的 Agent Note 摘要。仅靠评审无法阻止这种增长，因此仓库需要在文档分类体系之外再配一套可自动执行的预算。
+尽管已有写作指导，常设文档仍不断累积重复规则、反复讲述的事件、重复的包（package）映射，以及陈旧的 Agent Note（agent 决策记录）摘要。该指导也未明确文档在层级中的位置如何限定其内容范围，以及按顺序引导读者学习的内容与面向查阅的材料有何不同。仅靠评审无法阻止这种增长，因此仓库需要在文档分类之外再配一套机械预算。
 
 ## 决策
 
-- **每项事实只归属一处的层级分类体系。**[docs/AGENTS.md](../../../../docs/AGENTS.md) 是文档标准：它为每种 Markdown 层级分配单一职责（常设指令、系统图、类型目录、决策记录、事故叙事、操作指南、各包契约、生成的目录、工作流），禁止在事实归属层级之外重复陈述（应改为链接），并包含编写或评审任何文档时使用的赘余检查清单。
-- **范围窄且严格的预算门禁。**[scripts/verify-doc-budgets.ts](../../../../scripts/verify-doc-budgets.ts) 接入 `doc-sync`：[scripts/doc-budgets.manifest.json](../../../../scripts/doc-budgets.manifest.json) 列出的每份文档都不得超过其词数上限（采用 `wc -w` 语义，统计整个文件）；预算内文件缺失也会使门禁失败，使文件重命名无法悄然摆脱预算约束。范围刻意只涵盖容易膨胀的常设文档——根目录和子树中的 `AGENTS.md` 文件、`architecture.md`、`packages/README.md`，以及它们将内容移入的常设策略文档（`docs/testing.md`、`docs/defensive-patterns.md`）。参考文档、Agent Note 和包 README 不设预算：只要每一行都是事实，长度在这些位置就是合理的；评审和赘余检查清单负责约束它们。
-- **上限是会逐步收紧的执行红线。** 上限设定为文档当前词数的至少 105%（留出工作余量，使日常措辞调整能通过，而真正的膨胀仍会触发门禁），并随着文档被精简到目标预算而同步下调、保持该余量（根 `AGENTS.md` ≤ 1,500 词；`architecture.md` ≤ 1,800；子树 `AGENTS.md` ≤ 600；`packages/README.md` ≤ 600）。门禁变红时，修复方式是按分类体系迁移或压缩内容；只有在 PR（Pull Request）描述中给出明确理由时才允许提高上限，manifest（元数据清单）的 diff 本身就是可供评审的变更。
-- **精简的工作流 skill（技能），契约归文档。**[.agents/skills/dsh-doc-standards](../../../skills/dsh-doc-standards/SKILL.md) 承载文档放置、审计和门禁失败处理工作流，并以文档标准为真源，与 [dsh-translate-docs](../../../skills/dsh-translate-docs/SKILL.md) 和 i18n 契约之间的分工相同。
+- **结构遵循文档树。**[docs/AGENTS.md](../../../../docs/AGENTS.md) 是文档标准：文档负责承载其主题的详细内容，仅概述直接子项的目的、职责和高层行为，并链接到更深层内容的归属文档。[Agent Note](../../README.md) 仍不受这一结构契约约束。每份面向人的文档要么是按顺序引导读者达成结果的教程（tutorial），要么是查阅范围明确的参考文档（reference）；[事故复盘（postmortem）](../../../../docs/postmortem/README.md) 是范围限定于单个事件的参考文档，其时间线记录证据。教程结合读者的起始知识，按前置依赖顺序介绍概念。
+- **每项事实只归属一处的层级分类。**文档标准为每种 Markdown 层级分配单一职责，禁止在事实归属层级之外重复陈述，并包含编写或评审任何文档时使用的赘余检查清单。
+- **范围窄且严格的预算门禁。**[scripts/verify-doc-budgets.ts](../../../../scripts/verify-doc-budgets.ts) 接入 `doc-sync`：[scripts/doc-budgets.manifest.json](../../../../scripts/doc-budgets.manifest.json) 列出的每份文档都必须低于其字数上限（采用 `wc -w` 语义，统计整个文件）；预算内文件缺失也会使门禁失败，使重命名无法悄然遗落其预算。范围刻意只涵盖容易膨胀的常设文档——根目录和子树中的 `AGENTS.md` 文件、`architecture.md`、`packages/README.md`，以及它们将内容移入的常设策略文档（`docs/testing.md`、`docs/defensive-patterns.md`）。参考文档、Agent Note 和包 README 不设预算：只要每一行都是事实，长度在这些位置就是合理的；评审和赘余检查清单负责约束它们。
+- **上限是只进不退的执行红线。** 达到或低于目标的文档在上限逐步下调时保留至少 5% 的余量；高于目标的文档则维持冻结的上限，在达到目标之前不得增长（根 `AGENTS.md` ≤ 1,600 词；`architecture.md` ≤ 1,800；子树 `AGENTS.md` ≤ 600，但 `packages/AGENTS.md` ≤ 650、`docs/AGENTS.md` ≤ 1,250；`packages/README.md` ≤ 600）。门禁变红时，迁移或压缩内容；只有在 PR（Pull Request）描述中给出明确理由时才提高上限。
+- **精简的工作流 skill（技能），契约归文档。**[.agents/skills/dsh-doc-standards](../../../skills/dsh-doc-standards/SKILL.md) 承载放置/审计/红灯门禁工作流，并以文档标准为真源，与 [dsh-translate-docs](../../../skills/dsh-translate-docs/SKILL.md) 和 i18n 契约之间的分工相同。
 
 ## 曾考虑的替代方案
 
@@ -23,6 +24,7 @@ Status: implemented
 
 ## 后果
 
-- 向受预算约束的文档添加内容现在需要腾挪空间：将新增内容迁移到其分类体系归属地并留下链接，或压缩现有行文来腾出空间。只增不减会导致 CI 失败。
-- 精简到目标预算的重写以堆叠的后续 PR 落地，每次合并时同步下调 manifest 中的上限；在各自落地之前，文档冻结的上限仅阻止进一步膨胀。
-- 词数是一个粗糙的代理指标，这是有意接受的：它无法判断质量，但它在内容被添加的那一刻强制触发迁移决策，而那正是作者拥有足够上下文来正确放置内容的时刻。
+- 向受预算约束的文档添加内容需要置换：将新增内容迁移到其分类体系归属地并留下指针，或压缩现有行文来腾出空间。只增不减会导致 CI 失败。
+- 结构评审先检查归属关系和文档形式，再进行句子层面的编辑，使较低层级的细节迁移到其归属文档，而不是在错误的位置加以润色。
+- 仍高于目标的受预算约束文档不得增长；达到目标后，将恢复 5% 的工作余量。
+- 字数是一个粗糙的代理指标，这是有意接受的：它无法判断质量，但它在内容被添加的那一刻强制触发迁移决策，而那正是作者拥有足够上下文来正确放置内容的时刻。

.agents/skills/dsh-doc-standards/SKILL.md

@@ -1,6 +1,6 @@
 ---
 name: dsh-doc-standards
-description: 'Use when writing, moving, reviewing, or auditing documentation in the deepseek-harness repo — choosing where content belongs, trimming doc slop, responding to a verify-doc-budgets gate failure, or requests like "improve the docs", "audit the docs for slop", "where should this be documented", "this doc is too long".'
+description: 'Use when writing, moving, reviewing, or auditing documentation in the deepseek-harness repo — choosing hierarchy and detail, separating tutorials from references, checking tutorial progression, trimming doc slop, responding to a verify-doc-budgets failure, or requests like "improve the docs", "audit the docs", "where should this be documented", or "this doc is too long".'
 ---
 
 # Applying the DeepSeek Harness Documentation Standard
@@ -9,24 +9,32 @@ The contract lives in [docs/AGENTS.md](../../../docs/AGENTS.md). This workflow c
 
 ## Sources of truth (read, don't re-summarize)
 
-- [docs/AGENTS.md](../../../docs/AGENTS.md) — the taxonomy ("one home per fact"), budgets, slop checklist.
+- [docs/AGENTS.md](../../../docs/AGENTS.md) — hierarchy, tutorial/reference forms, taxonomy, budgets, and slop checklist.
 - [.agents/notes/README.md](../../notes/README.md) — when a decision earns an Agent Note, how to file it, and what goes inside one (the header block, per-lifecycle skeleton, and Alternatives-considered mandate, gated by `verify-agent-note-format`); [docs/postmortem/README.md](../../../docs/postmortem/README.md) — when an incident earns a postmortem.
 - [docs/i18n/README.md](../../../docs/i18n/README.md) — the bilingual pairing contract; editing either side of a pair obligates the counterpart in the same change.
 - Root [AGENTS.md](../../../AGENTS.md) — the standing orders whose budget discipline this skill protects.
 - [Archived Agent Notes](../../notes/archived/AGENTS.md) — frozen historical snapshots excluded from editorial maintenance and evolving documentation gates.
 
-## Placing content
+## Review structure before prose
 
-Run the placement test in the standard's taxonomy table, then check the constraints that make a placement expensive or wrong:
+Apply the standard's authoring order to every human-facing document in scope. Do not apply this structural pass to Agent Notes. Classify a postmortem as a reference scoped to one incident; preserve the chronological evidence required by its contract without treating chronology as a teaching sequence.
+
+1. Locate the document in the repository and navigation trees. State its own subject and identify its direct children.
+2. Set the detail boundary. Keep full detail about the document's subject, summarize direct children by purpose, responsibility, and high-level behavior, and move deeper explanations to their owning descendants with links. Treat test infrastructure as descendant-owned unless it is the document's subject.
+3. Classify the document from its intended use, not its path or title. A tutorial must lead through ordered work to an observable outcome; a reference must support lookup within an explicit scope without requiring sequential reading.
+4. For a tutorial, privately classify the starting reader and concepts as beginner, intermediate, or advanced. Trace each concept to its prerequisites, reorder premature material, and move optional advanced detail to a later tutorial or reference.
+5. Split substantial mixed forms. Keep a small secondary form only behind a clear structural boundary.
+
+Then check constraints that make placement expensive or wrong:
 
 - Paired docs (`pnpm run verify-translation-pairing --list`) cost a zh counterpart update and a `--write` re-record on every edit — prefer an unpaired home for content that will churn.
 - Generated catalogs are never hand-edited; if the fact belongs there, change the generator's source.
 - Before renaming or moving any doc, grep for inbound references: `verify-md-links` catches Markdown links, `verify-doc-refs` catches `docs/*.md` citations in TypeScript comments, but nothing catches heading-anchor fragments — grep `#the-heading` across the repo yourself (one anchor is hardcoded in `scripts/gen-cordis-catalog.ts`).
 - A move is atomic: remove from the old home, add to the new home, and fix every inbound link in the same change.
 
-## Auditing the corpus
+## Audit the corpus
 
-The audit is a hunt for the standard's slop checklist, cheapest probes first. Verify and fetch the PR's live base, then run `pnpm --silent run change-scope --base <verified-base-ref>` to identify committed and dirty paths before applying semantic judgment. After a retarget or base merge, rerun the report and repeat the audit for prose introduced by the new base rather than relying on the earlier result.
+After the structural pass, hunt the standard's slop checklist with the cheapest probes first. Verify and fetch the PR's live base, then run `pnpm --silent run change-scope --base <verified-base-ref>` to identify committed and dirty paths before applying semantic judgment. After a retarget or base merge, rerun the report and audit prose introduced by the new base.
 
 1. Measure: `pnpm run verify-doc-budgets --list`, then `git ls-files '*.md' ':(exclude)vendor/**' | xargs wc -w | sort -rn | head -30` to spot unbudgeted outliers.
 2. Hunt narrated history: `rg -n "no longer|used to|previously|was moved|renamed" --glob '*.md' --glob '*.ts' --glob '!vendor/**'` and keep only contrasts against a live alternative. Keep the vendor exclusion last so include globs cannot override it.

README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write README.md
-README.md: 7777c35a2785856e553cb963920288ff8021dc27
-README.zh.md: 85ee977063e4d822d71e1e780ab4a461260aaf53
+README.md: b8e46044fb8857730b32d9fbbb9ed4de964d6017
+README.zh.md: e289d523bf61a577f1dd2335b3b4567736d9100d

README.md

@@ -8,11 +8,9 @@ It uses an architecture where **everything is a plugin**.
 
 ## Internal testing notice
 
-感谢您愿意拨冗试用 DeepSeek Harness。当前版本仍处于内部测试阶段，功能仍待完善，体验难免有些粗糙。
+DeepSeek Harness is under internal testing. Features and interfaces may change.
 
-“如切如磋，如琢如磨。” 产品的成长，离不开一次次真实的碰撞与坦诚的反馈。您在真实使用中发现的问题，也可能促使我们重新审视，甚至推翻已有的设计。
-
-为了帮助我们更准确地还原您真实使用中的问题，内测版本默认会上传所有 Session Log；如需关闭，可以设置环境变量 `DSH_TELEMETRY_DISABLED=1`。另外，如果您有任何反馈与建议，请在企业微信群中留言告诉我们。每一条反馈，都会帮助我们把它打磨得更好。
+The internal build uploads all Session Logs by default to help diagnose reported problems. Set `DSH_TELEMETRY_DISABLED=1` to disable telemetry. Send feedback through the internal WeChat group.
 
 ## Install
 
@@ -26,7 +24,7 @@ scripts/install.sh
 
 The installer requires `git` and Node `^22.19 || >=24`, offers to install `pnpm` when it is missing, prompts for a DeepSeek API key, builds the required repository artifacts, and launches the Web UI.
 
-The installer keeps every checkout under `~/.dsh/source`: the master clone at `~/.dsh/source/master` and each install's staging checkout as a git worktree `~/.dsh/source/staging-<timestamp>`. The stable symlink `~/.dsh/source/current` points at the active staging worktree, and `dsh` in `~/.local/bin` links to `current/bin/dsh`, so an upgrade repoints one symlink and the `dsh` on PATH never moves. Re-running the command adds a fresh staging worktree from an updated master and repoints `current` at it. See [`scripts/install.sh`](scripts/install.sh) for alternate install locations and other options.
+The default active checkout is `~/.dsh/source/current`, and the launcher is linked into `~/.local/bin`. Re-run the installer to update. [`scripts/install.sh`](scripts/install.sh) owns alternate locations, update mechanics, and recovery options.
 
 ## Use DeepSeek Harness
 
@@ -84,11 +82,6 @@ Follow <a href="https://x.com/Deepseekharness">DeepSeek Harness on Twitter</a> f
 
 ## Development
 
-```sh
-pnpm install
-pnpm run test:coverage
-```
-
 Start with the [development guide](docs/development.md) and read the [architecture](docs/architecture.md) before changing packages.
 
 For agents, follow [AGENTS.md](AGENTS.md).

README.zh.md

@@ -8,11 +8,9 @@ DeepSeek Harness（`dsh`）是一款基于 DeepSeek Harness SDK 构建的开源
 
 ## 内测声明
 
-感谢您愿意拨冗试用 DeepSeek Harness。当前版本仍处于内部测试阶段，功能仍待完善，体验难免有些粗糙。
+DeepSeek Harness 正处于内部测试阶段，功能和接口可能发生变化。
 
-“如切如磋，如琢如磨。” 产品的成长，离不开一次次真实的碰撞与坦诚的反馈。您在真实使用中发现的问题，也可能促使我们重新审视，甚至推翻已有的设计。
-
-为了帮助我们更准确地还原您真实使用中的问题，内测版本默认会上传所有 Session Log；如需关闭，可以设置环境变量 `DSH_TELEMETRY_DISABLED=1`。另外，如果您有任何反馈与建议，请在企业微信群中留言告诉我们。每一条反馈，都会帮助我们把它打磨得更好。
+为帮助诊断上报的问题，内测版本默认上传所有会话日志。设置 `DSH_TELEMETRY_DISABLED=1` 可关闭遥测。请通过内部企业微信群反馈问题和建议。
 
 ## 安装
 
@@ -26,7 +24,7 @@ scripts/install.sh
 
 安装器要求系统已安装 `git` 和 Node `^22.19 || >=24`，缺少 `pnpm` 时可代为安装，并会提示输入 DeepSeek API 密钥，然后构建所需的仓库产物并启动 Web UI。
 
-安装器会把所有检出都放在 `~/.dsh/source` 下：master 克隆位于 `~/.dsh/source/master`，每次安装的 staging 检出是一个 git worktree `~/.dsh/source/staging-<时间戳>`。稳定符号链接 `~/.dsh/source/current` 指向当前生效的 staging worktree，`~/.local/bin` 中的 `dsh` 链接到 `current/bin/dsh`，因此升级只需重指一个符号链接，PATH 上的 `dsh` 从不移动。再次运行该命令会基于更新后的 master 新增一个 staging worktree，并把 `current` 重指到它。其他安装位置和选项见 [`scripts/install.sh`](scripts/install.sh)。
+默认生效的检出位于 `~/.dsh/source/current`，启动器链接到 `~/.local/bin`。再次运行安装器即可更新。其他位置、更新机制和恢复选项由 [`scripts/install.sh`](scripts/install.sh) 负责。
 
 ## 使用 DeepSeek Harness
 
@@ -88,11 +86,6 @@ pnpm run demo:acp
 
 ## 开发
 
-```sh
-pnpm install
-pnpm run test:coverage
-```
-
 请先阅读[开发指南](docs/development.md)；修改包之前，请阅读[架构文档](docs/architecture.md)。
 
 面向 agent：遵循 [AGENTS.md](AGENTS.md)。

THIRD_PARTY_NOTICES.md

@@ -27,7 +27,7 @@ The Cordis framework and its foundation libraries are source-vendored into this
 
 ## Runtime npm dependencies
 
-External packages that a workspace package resolves at runtime. `scripts/install.sh` installs this repository itself, so the tier covers every plugin a user can mount from `cordis.yml` — not only what the `dsh` CLI/TUI, the Web UI, and the Python SDK runtime load by default.
+External packages that a workspace package resolves at runtime. `scripts/install.sh` installs this repository itself, so the tier covers every plugin a user can mount from `cordis.yml` — not only what the `dsh` CLI, Web UI, and Python SDK runtime load by default.
 
 | Package | License |
 | --- | --- |

apps/cli/README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write apps/cli/README.md
-README.md: 6fdca68eed11dffe46bf2fbde9a7899359690dca
-README.zh.md: d8d7122729df1dd8aaed8207ddfb0a0470778b01
+README.md: ce7af5a299e45d6f107686aff043246914dce8ed
+README.zh.md: e97fec9d6bb726cb1e419a1ca2fa1871d4d203ca

apps/cli/README.md

@@ -2,75 +2,24 @@
 
 English | [中文](README.zh.md)
 
-The `dsh` command has three entry modes: a required raw config overlay, a one-shot headless prompt, and the Web UI. [`src/args.ts`](src/args.ts) owns the Commander grammar, and [`src/bin.ts`](src/bin.ts) dynamically imports only the selected runner. Unknown commands and leaked options fail with a nonzero exit code.
+The `dsh` command is the product launcher for raw Cordis configurations, the Web UI, and one-shot headless tasks. [`src/args.ts`](src/args.ts) owns the command grammar, and [`src/bin.ts`](src/bin.ts) loads only the selected runner. Invalid commands, options from another mode, configuration errors, and boot failures exit nonzero.
 
-## Raw config
-
-Raw `dsh` requires an explicit patch-list config:
-
-```sh
-dsh --config ./app.cordis.yml
-```
-
-The named file is applied directly over [`config/base.cordis.yml`](config/base.cordis.yml) through the Include plugin's patch algorithm. It is not a complete replacement tree, and neither the personal `$DSH_HOME/config.yaml` nor another surface overlay is added. The base deliberately contains no startup agent or interaction front door; the required overlay selects those deployment details. Relative config paths resolve from the invoking directory. A parse, schema, resolution, or plugin boot failure is reported and exits nonzero. SIGINT and SIGTERM dispose the mounted root before exit.
-
-A patch targets a base row by `id` and replaces that row's complete `config` value rather than deep-merging keys. Patch lists may also insert new rows whose plugin modules the shipped Loader can resolve:
-
-```yaml
-- id: agent-loop
-  config:
-    agents:
-      - id: main
-        provider: deepseek-official
-        model: deepseek-v4-flash
-```
-
-Inspect the effective tree without booting it:
-
-```sh
-dsh --dump-default-config
-dsh --config ./app.cordis.yml --dump-config
-```
-
-`--dump-default-config` prints only the shipped base. `--dump-config` requires `--config` and prints base plus overlay with provenance comments. Composition uses `applyEntryPatches` and `entryListSchema` from `@cordisjs/plugin-include`; `!!js` expressions remain unevaluated, and unmatched patch targets are reported on stderr.
-
-## Web and headless
+## Entry modes
 
-`dsh web` boots `base.cordis.yml` plus [`config/web.cordis.yml`](config/web.cordis.yml), followed by `$DSH_HOME/config.yaml` when present. `dsh web --config <path>` replaces that personal layer with the explicit patch list. `--host`, `--port`, `--workspace-root`, and repeatable `--trusted-host` values become Web host patches; their owning plugin schemas validate them at boot. `--dev` mounts the client-plugin HMR receiver and expects a separate `pnpm run dev:web` watcher for no-refresh client bundle updates.
+| Command | Purpose |
+|---|---|
+| `dsh --config ./app.cordis.yml` | Run an explicit patch-list configuration over the shipped base. |
+| `dsh web` | Start the browser UI with the shipped Web composition and optional personal configuration. |
+| `dsh -p "task"` | Run one fresh persisted session, print the final answer, and exit. |
 
-```sh
-dsh web
-dsh web --config ./web-profile.cordis.yml
-dsh web --dump-default-config
-dsh web --dump-config
-```
+The invoking directory is the default workspace root. Web and headless share the shipped provider, persistence, policy, tool, repository Plugin, and telemetry composition; raw config selects its own deployment-specific front door.
 
-The production Web runner needs built package and frontend artifacts (`pnpm run build`). It serves `http://127.0.0.1:3080` by default. Binding all interfaces also trusts the machine's discovered LAN IP literals; `--trusted-host` adds named authorities accepted by the `/api` browser-trust fence.
-
-`dsh -p "task"` uses the same base and Web composition with the startup personal config, starts its Web host on an OS-assigned port, runs one fresh persisted session, prints the final answer, and exits. It accepts neither `--config` nor raw config-dump flags.
-
-Web and headless process shutdown gives the plugin tree up to five seconds to dispose. The first `SIGINT`/`SIGTERM` starts that graceful drain; a second signal forces immediate exit. If headless normal completion is already stuck in disposal, the first `Ctrl+C` is the escalation and exits immediately instead of being swallowed.
-
-Both modes treat the invoking directory as the default workspace root, load applicable `AGENTS.md` or `CLAUDE.md` instructions with a 65,536-byte render budget, and use an in-memory SQLite session content index. Web watches valid personal config edits; headless reads the file once at startup. The [app-boot personal-config contract](../../packages/ui/app-boot/README.md#personal-config) owns layer precedence, credential storage, live-update failure behavior, and `$DSH_HOME` resolution.
-
-New sessions default to the `workspace-write` permission preset. Bash and filesystem mutations are restricted to the session workspace and platform temporary roots; reads, network access, and process visibility are not confined. `DSH_PERMISSION_MODE` changes the process fallback. Stored General-settings permissions affect later Web sessions, not an already-open one.
-
-`DSH_TOOLS_MODE` selects `native`, `code`, or `both` for the Web/headless process; another value fails at boot. [`config/core-web.cordis.yml`](config/core-web.cordis.yml) is an optional Web overlay that reduces the native model surface to persistent `bash` and `str_replace_editor` while retaining the shipped host, browser, workspace, persistence, and permission composition.
-
-## Shared deployment behavior
-
-The base mounts the native DeepSeek adapter, settings and credential providers, stable `web_search`, repository Plugin support, and session telemetry. Provider credentials live in `$DSH_HOME/.env` or the ambient environment and remain rotatable because the launcher never hoists the credential file into `process.env`. Search uses `DEEPSEEK_API_KEY` and accepts `DEEPSEEK_SEARCH_BASE_URL`; `web_fetch` is disabled unless an overlay inserts a provider and enables it.
-
-Session events stream as OTLP/HTTP logs by default. `DSH_TELEMETRY_OTLP_URL` selects another collector. Any non-empty `DSH_TELEMETRY_DISABLED` disables the telemetry row before boot. The shipped base has no telemetry redaction rule, so exported records can contain message text, tool arguments and results, and workspace paths; the [telemetry Agent Note](../../.agents/notes/implemented/feature/2026-07-31-web-telemetry-default-mount.md) owns that deployment decision.
-
-The empty `repository-plugins` row lets Web/headless personal config and raw overlays mount prepared immutable repository Plugin generations. See the [repository Plugin contract](../../packages/cordis/repository-plugin/README.md#standalone-app-configuration). The CLI also ships `@deepseek-ai/dsh-mcp-client` as a dependency for overlays, but no MCP server is enabled by default because each server command is trusted executable code outside the agent sandbox.
+## Raw config
 
-## Source launcher
+Raw `dsh` requires `--config`. The named patch list is applied directly over [`config/base.cordis.yml`](config/base.cordis.yml); it is not a complete replacement tree and does not add a surface overlay or personal `$DSH_HOME/config.yaml`. Use `--dump-default-config` and `--dump-config` to inspect the resulting tree without booting it.
 
-Link the source-running launcher onto PATH:
+The [CLI behavior reference](reference/README.md) owns exact overlay precedence, flags, shutdown behavior, deployment defaults, and the source launcher.
 
-```sh
-ln -sf "$(pwd)/bin/dsh" ~/.local/bin/dsh
-```
+## Development
 
-It resolves the checkout through its real path and launches `apps/cli/src/bin.ts` with `node --import tsx/esm`. `TSX_TSCONFIG_PATH` is pinned to the checkout root, so workspace package resolution is independent of the invoking directory. `pnpm run dsh` uses the same entry and forwards arguments. The built form is `apps/cli/lib/bin.js` after `pnpm run build`.
+Production Web and headless runs require built package and frontend artifacts. From a checkout, `pnpm run dsh` runs the TypeScript entry and forwards arguments; the [source-launcher reference](reference/README.md#source-launcher) describes the PATH symlink and module-resolution contract.

apps/cli/README.zh.md

@@ -2,75 +2,24 @@
 
 [English](README.md) | 中文
 
-`dsh` 命令有三种入口模式：必需的原始配置 overlay、一次性 headless 提示词，以及 Web UI。[`src/args.ts`](src/args.ts) 拥有 Commander 命令行语法，[`src/bin.ts`](src/bin.ts) 只会动态导入选中模式的运行器。未知命令和误传入其他模式的选项都会以非零代码退出。
+`dsh` 命令是原始 Cordis 配置、Web UI 和一次性无头任务的产品启动器。[`src/args.ts`](src/args.ts) 负责命令语法，[`src/bin.ts`](src/bin.ts) 只加载选中的运行器。无效命令、来自其他模式的选项、配置错误和启动失败都会以非零状态退出。
 
-## 原始配置
-
-原始 `dsh` 要求显式传入一份 patch 列表配置：
-
-```sh
-dsh --config ./app.cordis.yml
-```
-
-指定文件会通过 Include 插件的 patch 算法，直接应用在 [`config/base.cordis.yml`](config/base.cordis.yml) 之上。它不是完整替换树，系统也不会添加个人 `$DSH_HOME/config.yaml` 或其他 surface overlay。base 有意不包含启动 agent（智能体）或交互入口；必需的 overlay 负责选择这些部署细节。相对配置路径以调用目录为基准解析。配置解析、schema 校验、模块解析或插件启动失败都会被报告，并以非零代码退出。SIGINT 和 SIGTERM 会在退出前 dispose（资源释放）已挂载的根上下文。
-
-patch 通过 `id` 定位 base 配置项，并替换该配置项的完整 `config` 值，而不是深度合并各个键。它也可以插入新配置项：
-
-```yaml
-- id: agent-loop
-  config:
-    agents:
-      - id: main
-        provider: deepseek-official
-        model: deepseek-v4-flash
-```
-
-可以在不启动应用的情况下检查有效配置树：
-
-```sh
-dsh --dump-default-config
-dsh --config ./app.cordis.yml --dump-config
-```
-
-`--dump-default-config` 只打印随附 base。`--dump-config` 要求提供 `--config`，并打印带来源注释的 base 与 overlay。组合过程使用 `@cordisjs/plugin-include` 的 `applyEntryPatches` 和 `entryListSchema`；`!!js` 表达式保持未求值状态，未匹配的 patch 目标会报告到 stderr。
-
-## Web 与 headless
+## 入口模式
 
-`dsh web` 会启动 `base.cordis.yml` 加 [`config/web.cordis.yml`](config/web.cordis.yml)，并在 `$DSH_HOME/config.yaml` 存在时继续应用该文件。`dsh web --config <path>` 会以显式 patch 列表替换个人层。`--host`、`--port`、`--workspace-root` 和可重复的 `--trusted-host` 值会转为 Web 宿主 patch；各自所属插件的 schema 会在启动时校验它们。`--dev` 会挂载客户端插件 HMR（热模块替换）接收器，要实现无需刷新的客户端 bundle 更新，还需单独运行 `pnpm run dev:web` watcher。
+| 命令 | 用途 |
+|---|---|
+| `dsh --config ./app.cordis.yml` | 在随附基础配置之上运行显式 patch 列表配置。 |
+| `dsh web` | 使用随附 Web 组合和可选个人配置启动浏览器 UI。 |
+| `dsh -p "task"` | 运行一个新的持久化会话，打印最终答案并退出。 |
 
-```sh
-dsh web
-dsh web --config ./web-profile.cordis.yml
-dsh web --dump-default-config
-dsh web --dump-config
-```
+调用目录是默认 workspace 根目录。Web 与无头模式共享随附的提供方、持久化、策略、工具、repository Plugin 和遥测组合；原始配置自行选择部署专用前端入口。
 
-生产 Web 运行器需要已构建的包（package）与前端产物（`pnpm run build`）。它默认通过 `http://127.0.0.1:3080` 提供服务。绑定所有网络接口时，系统也会信任本机探测到的 LAN IP 字面量；`--trusted-host` 可添加 `/api` 浏览器信任边界所接受的具名权威。
-
-`dsh -p "task"` 使用相同的 base 与 Web 组合及启动时个人配置，在由操作系统分配的端口上启动 Web 宿主，运行一个全新的持久会话，打印最终答案后退出。它不接受 `--config` 或原始配置输出标志。
-
-Web 与 headless 的进程关闭流程最多给插件树 5 秒执行 dispose。第一次 `SIGINT`/`SIGTERM` 会启动这次优雅排空；第二次信号会立即强制退出。如果 headless 的正常完成流程已经卡在 dispose 中，第一次 `Ctrl+C` 就会触发强制退出：进程立即结束，该信号不再被吞掉。
-
-两种模式都以调用目录作为默认 workspace 根目录，加载适用的 `AGENTS.md` 或 `CLAUDE.md` 指令，渲染预算为 65,536 字节，并使用内存 SQLite 会话内容索引。Web 会持续应用有效的个人配置编辑；headless 只在启动时读取该文件一次。层次优先级、凭据存储、实时更新失败行为与 `$DSH_HOME` 解析均由 [app-boot 个人配置契约](../../packages/ui/app-boot/README.md#personal-config) 统一定义。
-
-新会话默认使用 `workspace-write` 权限 preset。Bash 和文件系统写操作受限于会话 workspace 与平台临时根目录；读取、网络访问与进程可见性不受限制。`DSH_PERMISSION_MODE` 会改变进程回退值。已存储的常规设置权限会影响之后的 Web 会话，不会更改已打开的会话。
-
-`DSH_TOOLS_MODE` 为 Web/headless 进程选择 `native`、`code` 或 `both`；其他值会在启动时失败。[`config/core-web.cordis.yml`](config/core-web.cordis.yml) 是可选的 Web overlay，它在保留随附宿主、浏览器、workspace、持久化与权限组合的同时，将面向原生模型的工具缩减为持久 `bash` 和 `str_replace_editor`。
-
-## 共享部署行为
-
-base 会挂载原生 DeepSeek 适配器、设置与凭据提供方、稳定的 `web_search`、仓库插件支持与会话遥测。提供方凭据位于 `$DSH_HOME/.env` 或环境中，且仍可轮换，因为启动器绝不会把凭据文件提升进 `process.env`。搜索使用 `DEEPSEEK_API_KEY` 并接受 `DEEPSEEK_SEARCH_BASE_URL`；除非 overlay 插入提供方并启用 `web_fetch`，否则后者处于禁用状态。
-
-会话事件默认以 OTLP/HTTP 日志的形式流式发送。`DSH_TELEMETRY_OTLP_URL` 用于选择其他 collector。`DSH_TELEMETRY_DISABLED` 的任何非空值都会在启动前禁用遥测配置项。随附 base 没有遥测脱敏规则，因此导出记录可能包含消息文本、工具参数与结果，以及 workspace 路径；该部署决策由[遥测 Agent Note](../../.agents/notes/implemented/feature/2026-07-31-web-telemetry-default-mount.md) 统一定义。
-
-空的 `repository-plugins` 配置项允许 Web/headless 个人配置与原始 overlay 挂载已准备的不可变仓库插件 generation。详见[仓库插件契约](../../packages/cordis/repository-plugin/README.md#standalone-app-configuration)。CLI（命令行界面）还将 `@deepseek-ai/dsh-mcp-client` 作为 overlay 依赖发布，但默认不启用任何 MCP 服务器，因为每条服务器命令都是 agent 沙箱之外的受信任可执行代码。
+## 原始配置
 
-## 源码启动器
+原始 `dsh` 必须提供 `--config`。指定的 patch 列表直接应用到 [`config/base.cordis.yml`](config/base.cordis.yml) 之上；它不是完整替代树，也不会添加 surface overlay 或个人 `$DSH_HOME/config.yaml`。使用 `--dump-default-config` 和 `--dump-config` 可在不启动的情况下检查生成的配置树。
 
-将以源码运行的启动器链接到 PATH：
+[CLI（命令行界面）行为参考](reference/README.md)负责确切的 overlay 优先级、flag、关闭行为、部署默认值和源码启动器。
 
-```sh
-ln -sf "$(pwd)/bin/dsh" ~/.local/bin/dsh
-```
+## 开发
 
-它会通过自身实际路径解析该检出，并使用 `node --import tsx/esm` 启动 `apps/cli/src/bin.ts`。`TSX_TSCONFIG_PATH` 固定指向检出根目录，因此 workspace 包解析不受调用目录影响。`pnpm run dsh` 使用同一入口并转发参数。构建后的形式是执行 `pnpm run build` 后的 `apps/cli/lib/bin.js`。
+生产环境的 Web 和无头运行需要已构建的包与前端产物。在 checkout 中，`pnpm run dsh` 会运行 TypeScript 入口并转发参数；[源码启动器参考](reference/README.md#source-launcher)说明 PATH 符号链接和模块解析契约。

apps/cli/reference/README.i18n.yaml

@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write apps/cli/reference/README.md
+README.md: b37ec9ed61ea4e9899a51316065d4188f30997ad
+README.zh.md: ca29808a6c8e670f0d0b82c59b1a2c1fa0e13565

apps/cli/reference/README.md

@@ -0,0 +1,76 @@
+# `dsh` CLI behavior reference
+
+English | [中文](README.zh.md)
+
+This reference defines the raw-config, Web, and headless command modes. Argv is parsed once through [`src/args.ts`](../src/args.ts), and [`src/bin.ts`](../src/bin.ts) dynamically imports only the selected runner.
+
+## Raw config
+
+Raw `dsh` requires an explicit patch-list config:
+
+```sh
+dsh --config ./app.cordis.yml
+```
+
+The named file is applied directly over [`config/base.cordis.yml`](../config/base.cordis.yml) through the Include plugin's patch algorithm. It is not a complete replacement tree, and neither the personal `$DSH_HOME/config.yaml` nor another surface overlay is added. The base deliberately contains no startup agent or interaction front door; the required overlay selects those deployment details. Relative config paths resolve from the invoking directory. A parse, schema, resolution, or plugin boot failure is reported and exits nonzero. SIGINT and SIGTERM dispose the mounted root before exit.
+
+A patch targets a base row by `id` and replaces that row's complete `config` value rather than deep-merging keys. Patch lists may also insert new rows whose plugin modules the shipped Loader can resolve:
+
+```yaml
+- id: agent-loop
+  config:
+    agents:
+      - id: main
+        provider: deepseek-official
+        model: deepseek-v4-flash
+```
+
+Inspect the effective tree without booting it:
+
+```sh
+dsh --dump-default-config
+dsh --config ./app.cordis.yml --dump-config
+```
+
+`--dump-default-config` prints only the shipped base. `--dump-config` requires `--config` and prints base plus overlay with provenance comments. Composition uses `applyEntryPatches` and `entryListSchema` from `@cordisjs/plugin-include`; `!!js` expressions remain unevaluated, and unmatched patch targets are reported on stderr.
+
+## Web and headless
+
+`dsh web` boots `base.cordis.yml` plus [`config/web.cordis.yml`](../config/web.cordis.yml), followed by `$DSH_HOME/config.yaml` when present. `dsh web --config <path>` replaces that personal layer with the explicit patch list. `--host`, `--port`, `--workspace-root`, and repeatable `--trusted-host` values become Web host patches; their owning plugin schemas validate them at boot. `--dev` mounts the client-plugin HMR receiver and expects a separate `pnpm run dev:web` watcher for no-refresh client bundle updates.
+
+```sh
+dsh web
+dsh web --config ./web-profile.cordis.yml
+dsh web --dump-default-config
+dsh web --dump-config
+```
+
+The production Web runner needs built package and frontend artifacts (`pnpm run build`). It serves `http://127.0.0.1:3080` by default. Binding all interfaces also trusts the machine's discovered LAN IP literals; `--trusted-host` adds named authorities accepted by the `/api` browser-trust fence.
+
+`dsh -p "task"` uses the same base and Web composition with the startup personal config, starts its Web host on an OS-assigned port, runs one fresh persisted session, prints the final answer, and exits. It accepts neither `--config` nor raw config-dump flags.
+
+Web and headless process shutdown gives the plugin tree up to five seconds to dispose. The first `SIGINT`/`SIGTERM` starts that graceful drain; a second signal forces immediate exit. If headless normal completion is already stuck in disposal, the first `Ctrl+C` is the escalation and exits immediately instead of being swallowed.
+
+Both modes treat the invoking directory as the default workspace root, load applicable `AGENTS.md` or `CLAUDE.md` instructions with a 65,536-byte render budget, and use an in-memory SQLite session content index. Web watches valid personal config edits; headless reads the file once at startup. The [app-boot personal-config contract](../../../packages/ui/app-boot/README.md#personal-config) owns layer precedence, credential storage, live-update failure behavior, and `$DSH_HOME` resolution.
+
+New sessions default to the `workspace-write` permission preset. Bash and filesystem mutations are restricted to the session workspace and platform temporary roots; reads, network access, and process visibility are not confined. `DSH_PERMISSION_MODE` changes the process fallback. Stored General-settings permissions affect later Web sessions, not an already-open one.
+
+`DSH_TOOLS_MODE` selects `native`, `code`, or `both` for the Web/headless process; another value fails at boot. [`config/core-web.cordis.yml`](../config/core-web.cordis.yml) is an optional Web overlay that reduces the native model surface to persistent `bash` and `str_replace_editor` while retaining the shipped host, browser, workspace, persistence, and permission composition.
+
+## Shared deployment behavior
+
+The base mounts the native DeepSeek adapter, settings and credential providers, stable `web_search`, repository Plugin support, and session telemetry. Provider credentials live in `$DSH_HOME/.env` or the ambient environment and remain rotatable because the launcher never hoists the credential file into `process.env`. Search uses `DEEPSEEK_API_KEY` and accepts `DEEPSEEK_SEARCH_BASE_URL`; `web_fetch` is disabled unless an overlay inserts a provider and enables it.
+
+Session events stream as OTLP/HTTP logs by default. `DSH_TELEMETRY_OTLP_URL` selects another collector. Any non-empty `DSH_TELEMETRY_DISABLED` disables the telemetry row before boot. The shipped base has no telemetry redaction rule, so exported records can contain message text, tool arguments and results, and workspace paths; the [telemetry Agent Note](../../../.agents/notes/implemented/feature/2026-07-31-web-telemetry-default-mount.md) owns that deployment decision.
+
+The empty `repository-plugins` row lets Web/headless personal config and raw overlays mount prepared immutable repository Plugin generations. See the [repository Plugin contract](../../../packages/cordis/repository-plugin/README.md#standalone-app-configuration). The CLI also ships `@deepseek-ai/dsh-mcp-client` as a dependency for overlays, but no MCP server is enabled by default because each server command is trusted executable code outside the agent sandbox.
+
+## Source launcher
+
+Link the source-running launcher onto PATH:
+
+```sh
+ln -sf "$(pwd)/bin/dsh" ~/.local/bin/dsh
+```
+
+It resolves the checkout through its real path and launches `apps/cli/src/bin.ts` with `node --import tsx/esm`. `TSX_TSCONFIG_PATH` is pinned to the checkout root, so workspace package resolution is independent of the invoking directory. `pnpm run dsh` uses the same entry and forwards arguments. The built form is `apps/cli/lib/bin.js` after `pnpm run build`.

apps/cli/reference/README.zh.md

@@ -0,0 +1,76 @@
+# `dsh` CLI（命令行界面）行为参考
+
+[English](README.md) | 中文
+
+本参考定义原始配置、Web 和无头命令模式。参数由 [`src/args.ts`](../src/args.ts) 统一解析，[`src/bin.ts`](../src/bin.ts) 只动态导入选中的运行器。
+
+## 原始配置
+
+原始 `dsh` 必须提供显式 patch 列表配置：
+
+```sh
+dsh --config ./app.cordis.yml
+```
+
+指定文件通过 Include 插件的 patch 算法直接应用到 [`config/base.cordis.yml`](../config/base.cordis.yml) 之上。它不是完整替代树，也不会添加个人 `$DSH_HOME/config.yaml` 或其他 surface overlay。基础配置刻意不包含启动 agent（智能体）或交互前端入口；必填 overlay 负责选择这些部署细节。相对配置路径从调用目录解析。配置解析、schema 校验、模块解析或插件启动失败会得到报告并以非零状态退出。收到 SIGINT 或 SIGTERM 时，挂载的根节点会先 dispose（资源释放）再退出。
+
+patch 通过 `id` 定位基础配置行，并替换该行完整的 `config` 值，而不是深度合并各键。patch 列表也可插入新行，只要随附 Loader 能解析其插件模块：
+
+```yaml
+- id: agent-loop
+  config:
+    agents:
+      - id: main
+        provider: deepseek-official
+        model: deepseek-v4-flash
+```
+
+可在不启动的情况下检查生效的配置树：
+
+```sh
+dsh --dump-default-config
+dsh --config ./app.cordis.yml --dump-config
+```
+
+`--dump-default-config` 只打印随附基础配置。`--dump-config` 必须与 `--config` 同时使用，并打印基础配置和带来源注释的 overlay。组合使用 `@cordisjs/plugin-include` 的 `applyEntryPatches` 与 `entryListSchema`；`!!js` 表达式保持未求值，找不到目标的 patch 会报告到 stderr。
+
+## Web 与无头模式
+
+`dsh web` 启动 `base.cordis.yml` 加 [`config/web.cordis.yml`](../config/web.cordis.yml)，并在 `$DSH_HOME/config.yaml` 存在时继续加载它。`dsh web --config <path>` 用显式 patch 列表替代该个人层。`--host`、`--port`、`--workspace-root` 和可重复的 `--trusted-host` 值会成为 Web 宿主 patch；负责这些值的插件 schema 会在启动时验证它们。`--dev` 挂载客户端插件 HMR（热模块替换）接收器；若要无刷新更新客户端 bundle，还需单独运行 `pnpm run dev:web` watcher。
+
+```sh
+dsh web
+dsh web --config ./web-profile.cordis.yml
+dsh web --dump-default-config
+dsh web --dump-config
+```
+
+生产 Web 运行器需要已构建的包和前端产物（`pnpm run build`）。默认服务地址是 `http://127.0.0.1:3080`。绑定所有接口时，还会信任机器自动发现的 LAN IP 字面量；`--trusted-host` 可添加 `/api` 浏览器信任围栏接受的具名 authority。
+
+`dsh -p "task"` 使用同一基础配置和 Web 组合，并加载启动时的个人配置；它在 OS 分配的端口上启动 Web 宿主，运行一个新的持久化会话，打印最终答案并退出。它不接受 `--config` 或原始配置 dump flag。
+
+Web 和无头进程关闭时会给插件树最多 5 秒完成 dispose。第一次 `SIGINT`/`SIGTERM` 启动该优雅排空；第二次信号强制立即退出。如果无头模式正常结束时已经卡在 dispose 中，第一次 `Ctrl+C` 就会升格并立即退出，而不会被吞掉。
+
+两种模式都将调用目录作为默认 workspace 根目录，以 65,536 字节渲染预算加载适用的 `AGENTS.md` 或 `CLAUDE.md` 指令，并使用内存 SQLite 会话内容索引。Web 监视有效的个人配置编辑；无头模式只在启动时读取该文件。[app-boot 个人配置契约](../../../packages/ui/app-boot/README.md#personal-config)负责配置层优先级、凭据存储、实时更新失败行为和 `$DSH_HOME` 解析。
+
+新会话默认使用 `workspace-write` 权限预设。Bash 和文件系统修改仅限于会话 workspace 与平台临时根目录；读取、网络访问和进程可见性不受限制。`DSH_PERMISSION_MODE` 更改进程后备值。General settings 中存储的权限影响后续 Web 会话，不改变已打开的会话。
+
+`DSH_TOOLS_MODE` 为 Web／无头进程选择 `native`、`code` 或 `both`；其他值会导致启动失败。[`config/core-web.cordis.yml`](../config/core-web.cordis.yml) 是可选 Web overlay：它在保留随附宿主、浏览器、workspace、持久化和权限组合的同时，把原生模型 surface 缩减为持久 `bash` 和 `str_replace_editor`。
+
+## 共享部署行为
+
+基础配置挂载原生 DeepSeek 适配器、settings 与凭据提供方、稳定的 `web_search`、repository Plugin 支持和会话遥测。提供方凭据存放在 `$DSH_HOME/.env` 或环境中；启动器从不把凭据文件提升到 `process.env`，因此凭据可以轮换。搜索使用 `DEEPSEEK_API_KEY` 并接受 `DEEPSEEK_SEARCH_BASE_URL`；只有 overlay 插入提供方并启用 `web_fetch` 后，该工具才可用。
+
+会话事件默认作为 OTLP/HTTP 日志流式发送。`DSH_TELEMETRY_OTLP_URL` 选择其他 collector。任何非空 `DSH_TELEMETRY_DISABLED` 都会在启动前禁用遥测配置行。随附基础配置没有遥测脱敏规则，因此导出的记录可能包含消息文本、工具参数与结果以及 workspace 路径；该部署决策由[遥测 Agent Note](../../../.agents/notes/implemented/feature/2026-07-31-web-telemetry-default-mount.md)负责。
+
+空 `repository-plugins` 行让 Web／无头个人配置和原始 overlay 能够挂载已准备的不可变 repository Plugin generation。参见 [repository Plugin 契约](../../../packages/cordis/repository-plugin/README.md#standalone-app-configuration)。CLI 还随附 `@deepseek-ai/dsh-mcp-client` 作为 overlay 的依赖，但默认不启用 MCP 服务器，因为每条服务器命令都是 agent 沙箱之外的受信任可执行代码。
+
+## 源码启动器
+
+把源码运行启动器链接到 PATH：
+
+```sh
+ln -sf "$(pwd)/bin/dsh" ~/.local/bin/dsh
+```
+
+它通过 real path 解析 checkout，并使用 `node --import tsx/esm` 启动 `apps/cli/src/bin.ts`。`TSX_TSCONFIG_PATH` 固定到 checkout 根目录，因此 workspace 包解析不依赖调用目录。`pnpm run dsh` 使用同一入口并转发参数。运行 `pnpm run build` 后，构建形式为 `apps/cli/lib/bin.js`。

apps/web/tests/access-confirmation.e2e.ts

@@ -2,7 +2,6 @@
 // the same locale-aware, in-page risk confirmation. Zero model calls: the
 // scenario boots the shipped Web composition and exercises the real
 // permission projection, client command path, HTTP RPC, and pushed update.
-import { mkdirSync } from 'node:fs'
 import { fileURLToPath } from 'node:url'
 import { join } from 'node:path'
 import type { Browser, Page } from 'playwright'
@@ -12,27 +11,7 @@ import {
   assertFixtureInventory, captureStableAria, compareOrRefreshGolden,
   launchWebScaffold, watchConsole, webSnapshotMode, type WebScaffold,
 } from './scaffold.ts'
-import { ZH_BROWSER_LOCALE, saveFailureShot } from './support.ts'
-
-/**
- * connectFreshWorkspace twin over the product default Chinese locale (the
- * shared helper's anchors assume the English page every other scenario
- * boots; this scenario deliberately keeps zh, so the localized picker
- * copy is the anchor set).
- */
-async function connectFreshWorkspaceZh(page: Page, root: string, name = 'workspace'): Promise<void> {
-  mkdirSync(join(root, name), { recursive: true })
-  await page.getByRole('button', { name: '选择工作区' }).click()
-  const dialog = page.getByRole('dialog', { name: '选择工作区目录' })
-  await dialog.waitFor({ timeout: 10_000 })
-  await dialog.getByRole('button', { name: '编辑路径' }).click()
-  const pathInput = dialog.getByRole('textbox', { name: '编辑路径' })
-  await pathInput.fill(join(root, name))
-  await pathInput.press('Enter')
-  await dialog.getByRole('button', { name: '打开', exact: true }).click()
-  await page.locator('textarea:enabled[placeholder="描述你想要构建的内容"]')
-    .waitFor({ timeout: 15_000 })
-}
+import { ZH_BROWSER_LOCALE, connectFreshWorkspaceZh, saveFailureShot } from './support.ts'
 
 const SNAPSHOT_DIR = fileURLToPath(new URL('./snapshots/access-confirmation', import.meta.url))
 const UI_EXPECTED = join(SNAPSHOT_DIR, 'ui.expected.md')

apps/web/tests/chat-scroll-fixture.ts

@@ -179,7 +179,7 @@ function fixtureLog(session: Session): string {
 export function createChatScrollFixture(options: ChatScrollFixtureOptions): ChatScrollFixture {
   const turns = options.turns ?? DEFAULT_TURNS
   const markers = markerHelpers(options.markerPrefix)
-  const session = new Session(SessionId(`chat-scroll-${options.markerPrefix.toLowerCase()}-template`))
+  const session = Session.create(SessionId(`chat-scroll-${options.markerPrefix.toLowerCase()}-template`))
 
   for (let turn = 1; turn <= turns; turn += 1) {
     session.append('turn/start', {

apps/web/tests/complex-history.perf.ts

@@ -319,7 +319,7 @@ function fixtureLog(session: Session): string {
 }
 
 function smallSidebarFixture(): string {
-  const session = new Session(SessionId('perf-small-template'))
+  const session = Session.create(SessionId('perf-small-template'))
   session.append('turn/start', {
     turn: 1,
     trigger: { kind: 'message', source: { kind: 'user' } },
@@ -342,7 +342,7 @@ function smallSidebarFixture(): string {
 }
 
 function longHistoryFixture(): string {
-  const session = new Session(SessionId(LONG_SESSION_ID))
+  const session = Session.create(SessionId(LONG_SESSION_ID))
   for (let turn = 1; turn <= LONG_HISTORY_TURNS; turn += 1) {
     session.append('turn/start', {
       turn,

apps/web/tests/lifecycle-chrome.e2e.ts

@@ -104,6 +104,10 @@ describe('web e2e: lifecycle & chrome (workspace flow / reload / dark mode)', ()
       await input.press('Enter')
       const planButton = activePage.getByRole('button', { name: 'Plan mode on, press to turn off' })
       await planButton.waitFor({ timeout: 10_000 })
+      // The golden encodes an empty composer, and the button arriving does not
+      // mean the submitted text is gone yet: under load the capture caught a
+      // textbox still holding `/plan`.
+      await expect.poll(() => input.inputValue(), { timeout: 10_000 }).toBe('')
       const planSnapshot = await captureStableAria(activePage, '[class*="frame"]', activeScaffold.workspaceCwd)
       await compareOrRefreshGolden(PLAN_ACTIVE_EXPECTED, planSnapshot, MODE)
       const planStyle = await planButton.evaluate((element) => {

apps/web/tests/markdown-images.e2e.ts

@@ -82,7 +82,7 @@ async function stopServer(server: Server): Promise<void> {
 
 /** Build one closed, invariant-checked session fixture with remote and local image Markdown. */
 function markdownImageFixture(remoteUrl: string): string {
-  const session = new Session(SessionId('markdown-image-source'))
+  const session = Session.create(SessionId('markdown-image-source'))
   session.append('turn/start', {
     turn: 1,
     trigger: { kind: 'message', source: { kind: 'user' } },

apps/web/tests/message-actions.e2e.ts

@@ -126,9 +126,8 @@ describe('web e2e: message IconActions and clocks on settled history', () => {
 
   it.skipIf(MODE === 'record')('matches the conversation aria golden with IconActions and clocks', async () => {
     onTestFailed(() => saveFailureShot(page, 'web-e2e-message-actions-aria'))
-    await page.getByRole('button', {
-      name: 'Select model, current deepseek-v4-flash',
-    }).waitFor({ timeout: 10_000 })
+    await page.getByRole('button', { name: 'Select model', exact: true })
+      .waitFor({ timeout: 10_000 })
     // Keep a footer focused so opacity-hidden actions stay in the a11y tree
     // as an active/focused control during the capture.
     await page.getByRole('button', { name: 'Copy' }).first().focus()

apps/web/tests/onboarding-deepseek-config.e2e.ts

@@ -12,7 +12,7 @@ import {
   acknowledgeReloadConnectionLoss, assertFixtureInventory, captureStableAria, compareOrRefreshGolden,
   launchWebScaffold, watchConsole, webSnapshotMode, type WebScaffold,
 } from './scaffold.ts'
-import { ZH_BROWSER_LOCALE, saveFailureShot } from './support.ts'
+import { ZH_BROWSER_LOCALE, connectFreshWorkspaceZh, saveFailureShot } from './support.ts'
 import { settingsNamespace } from '@deepseek-ai/dsh-settings'
 import {
   WELCOME_NOTICE_ACK_FIELD, WELCOME_NOTICE_COPY, WELCOME_NOTICE_SETTINGS_NAMESPACE,
@@ -22,6 +22,7 @@ import {
 const SNAPSHOT_DIR = fileURLToPath(new URL('./snapshots/onboarding-deepseek-config', import.meta.url))
 const WELCOME_EXPECTED = join(SNAPSHOT_DIR, 'welcome.expected.md')
 const MISSING_EXPECTED = join(SNAPSHOT_DIR, 'missing.expected.md')
+const MODELS_EXPECTED = join(SNAPSHOT_DIR, 'models.expected.md')
 const MODE = webSnapshotMode()
 
 describe.skipIf(MODE === 'record')('web e2e: first-run DeepSeek credential setup', () => {
@@ -160,7 +161,60 @@ describe.skipIf(MODE === 'record')('web e2e: first-run DeepSeek credential setup
     expect(tripwire.pageErrors).toEqual([])
   }, 60_000)
 
+  it('configures arbitrary DeepSeek models and prompts after the selected model is removed', async () => {
+    onTestFailed(() => saveFailureShot(page, 'web-e2e-onboarding-deepseek-models'))
+    // Opened here rather than inherited: the credential test reloads the page
+    // to exercise the welcome step, so nothing carries an open dialog across.
+    await page.getByRole('button', { name: '设置', exact: true }).click()
+    const settings = page.getByRole('dialog', { name: '设置' })
+    await settings.waitFor({ timeout: 10_000 })
+    await settings.getByRole('button', { name: '模型' }).click()
+    const deepSeek = settings.getByText('DeepSeek', { exact: true }).first()
+    await deepSeek.waitFor({ timeout: 10_000 })
+    await deepSeek.locator('xpath=ancestor::li').getByRole('button', { name: '编辑' }).click()
+    await settings.getByText('自定义设置').click()
+    await settings.getByRole('button', { name: /删除模型/ }).first().click()
+    await settings.getByRole('button', { name: '添加模型' }).click()
+    const customModelId = settings.getByLabel('模型 ID 2')
+    await customModelId.fill('private-preview')
+    await settings.getByLabel('显示名称 2').fill('Private Preview')
+    // Capacities live behind the row's own disclosure, as in the pi-ai form.
+    await settings.getByRole('button', { name: '容量 2' }).click()
+    await settings.getByLabel('上下文窗口 2').fill('131072')
+    await settings.getByLabel('最大输出 token 数 2').fill('64K')
+
+    const modelEditor = await captureStableAria(page, '[role="dialog"]', scaffold.workspaceCwd)
+    await compareOrRefreshGolden(MODELS_EXPECTED, modelEditor, MODE)
+    await settings.getByRole('button', { name: '保存', exact: true }).click()
+    await customModelId.waitFor({ state: 'detached', timeout: 15_000 })
+
+    const document = await readFile(join(scaffold.harnessHome, 'settings.yaml'), 'utf8')
+    expect(document).toContain('id: deepseek-v4-pro')
+    expect(document).toContain('id: private-preview')
+    expect(document).toContain('name: Private Preview')
+    expect(document).toContain('contextWindow: 131072')
+    expect(document).toContain('maxTokens: 64000')
+    expect(document).not.toContain('id: deepseek-v4-flash')
+
+    await page.keyboard.press('Escape')
+    // A connected Workspace is what puts a live composer — and its model
+    // trigger — on the page; the scaffold boots without one.
+    await connectFreshWorkspaceZh(page, scaffold.workspaceCwd, 'model-fallback-e2e')
+
+    const modelTrigger = page.getByRole('button', { name: '选择模型', exact: true })
+    await modelTrigger.waitFor({ timeout: 10_000 })
+    await modelTrigger.click()
+    await page.getByRole('menuitem', { name: /模型/ }).click()
+    expect(await page.getByText('deepseek-v4-flash', { exact: true }).count()).toBe(0)
+    await page.getByRole('menuitemradio', { name: 'Private Preview' }).waitFor({ timeout: 10_000 })
+    expect(tripwire.warnings).toEqual([])
+    expect(tripwire.pageErrors).toEqual([])
+  }, 60_000)
+
   it('keeps the fixture inventory closed', async () => {
-    await assertFixtureInventory(SNAPSHOT_DIR, ['missing.expected.md', 'welcome.expected.md'])
+    await assertFixtureInventory(
+      SNAPSHOT_DIR,
+      ['missing.expected.md', 'models.expected.md', 'welcome.expected.md'],
+    )
   })
 })

apps/web/tests/scaffold.ts

@@ -56,7 +56,7 @@ import type {} from '@deepseek-ai/dsh-agent'
 import { prepareWebRuntimeContext } from '../../cli/src/web.ts'
 import { DIST_INDEX, REPO_ROOT, requireDist } from './support.ts'
 
-/** Snapshot mode for the lane, from $DSH_SNAPSHOT (same vocabulary as the ACP/TUI suites). */
+/** Snapshot mode for the lane, from $DSH_SNAPSHOT (same vocabulary as the other snapshot suites). */
 export type WebSnapshotMode = 'replay' | 'record' | 'refresh'
 
 /**
@@ -592,9 +592,9 @@ export async function compareOrRefreshGolden(goldenPath: string, actual: string,
 }
 
 /**
- * Fixture-inventory guard (the TUI afterAll shape): the scenario directory
- * holds exactly the expected files and every committed JSONL is a scrub
- * fixed-point without a run-local browser RPC id.
+ * Fixture-inventory guard: the scenario directory holds exactly the expected
+ * files and every committed JSONL is a scrub fixed-point without a run-local
+ * browser RPC id.
  * @param dir - the scenario snapshot directory.
  * @param expected - the exact expected file inventory.
  */

apps/web/tests/seeded-history.e2e.ts

@@ -241,12 +241,11 @@ describe('web e2e: seeded history renders through cold resume', () => {
 
   it.skipIf(MODE === 'record')('matches the historical conversation aria golden', async () => {
     onTestFailed(() => saveFailureShot(page, 'web-e2e-seeded-aria'))
-    await page.getByRole('button', {
-      // This scenario deliberately leaves the LLM seam open to prove zero
-      // model calls. History still restores the selected id, but no catalog
-      // adapter exists to provide its presentation name.
-      name: 'Select model, current deepseek-v4-flash',
-    }).waitFor({ timeout: 10_000 })
+    // This scenario deliberately leaves the LLM seam open to prove zero
+    // model calls. History still restores the routed id, but without an
+    // advertised catalog row the selector prompts for a listed replacement.
+    await page.getByRole('button', { name: 'Select model', exact: true })
+      .waitFor({ timeout: 10_000 })
     const snapshot = (await captureStableAria(page, '[class*="centerCol"]', scaffold.workspaceCwd))
       .split(SEED_ID).join('{{seededId}}')
     await compareOrRefreshGolden(UI_EXPECTED, snapshot, MODE)

apps/web/tests/settings-chrome.e2e.ts

@@ -61,6 +61,30 @@ describe('web e2e: settings modal and General preferences', () => {
     await dialog.getByRole('button', { name: 'Workspace Write' }).waitFor({ timeout: 10_000 })
     await expect.poll(() => dialog.getByText('语言', { exact: true }).count(), { timeout: 5_000 }).toBe(1)
     await expect.poll(() => dialog.getByText('外观', { exact: true }).count(), { timeout: 5_000 }).toBe(1)
+    const openDocument = dialog.getByRole('button', { name: '打开配置文件' })
+    await openDocument.waitFor({ timeout: 10_000 })
+    let openRequests = 0
+    await page.route('**/api/settings.openDocument', async (route) => {
+      const envelope = route.request().postDataJSON() as {
+        rpcId: string
+        payload: Record<string, never>
+      }
+      expect(envelope.payload).toEqual({})
+      openRequests += 1
+      await route.fulfill({
+        status: 200,
+        contentType: 'application/json',
+        body: JSON.stringify({
+          type: 'server-response',
+          rpcId: envelope.rpcId,
+          result: { ok: true, value: { opened: true } },
+        }),
+      })
+    })
+    await openDocument.click()
+    await expect.poll(() => openRequests, { timeout: 5_000 }).toBe(1)
+    await expect.poll(() => openDocument.isEnabled(), { timeout: 5_000 }).toBe(true)
+    await page.unroute('**/api/settings.openDocument')
     // Golden of the freshly opened dialog (default zh, General active).
     const snapshot = await captureStableAria(page, '[role="dialog"]', scaffold.workspaceCwd)
     await compareOrRefreshGolden(DIALOG_EXPECTED, snapshot, MODE)

apps/web/tests/snapshots/bash-abort-row/ui.expected.md

@@ -26,8 +26,8 @@
 - button "Commands":
   - img
 - 'button "Access mode, current: Workspace Write"': Workspace Write
-- button "Select model, current deepseek-v4-flash":
-  - text: deepseek-v4-flash
+- button "Select model":
+  - text: Select model
   - img
 - button "Send message" [disabled]
 - text: 1 turns · 1 steps Tool call {{duration}} Cache hit 0% Input 10 tok · Output 10 tok

apps/web/tests/snapshots/lifecycle-chrome/plan-active.expected.md

@@ -30,8 +30,8 @@
   - img
 - 'button "Access mode, current: Workspace Write"': Workspace Write
 - button "Plan mode on, press to turn off": Plan
-- button "Select model, current deepseek-v4-flash":
-  - text: deepseek-v4-flash
+- button "Select model":
+  - text: Select model
   - img
 - button "Send message" [disabled]
 - text: Details

apps/web/tests/snapshots/markdown-images/ui.expected.md

@@ -24,8 +24,8 @@
 - button "Commands":
   - img
 - 'button "Access mode, current: Workspace Write"': Workspace Write
-- button "Select model, current deepseek-v4-flash":
-  - text: deepseek-v4-flash
+- button "Select model":
+  - text: Select model
   - img
 - button "Send message" [disabled]
 - text: 1 turns · 1 steps Input 0 tok · Output 0 tok

apps/web/tests/snapshots/message-actions/ui.expected.md

@@ -51,8 +51,8 @@
 - button "Commands":
   - img
 - 'button "Access mode, current: Workspace Write"': Workspace Write
-- button "Select model, current deepseek-v4-flash":
-  - text: deepseek-v4-flash
+- button "Select model":
+  - text: Select model
   - img
 - button "Send message" [disabled]
 - text: 2 turns · 3 steps Tool call {{duration}} Cache hit 98% Input 7.8K tok · Output 103 tok

apps/web/tests/snapshots/models-settings/configured.expected.md

@@ -7,6 +7,7 @@
     - button "模型":
       - img
       - text: 模型
+  - button "打开配置文件"
   - button "关闭":
     - img
     - text: 关闭

apps/web/tests/snapshots/models-settings/empty.expected.md

@@ -7,6 +7,7 @@
     - button "模型":
       - img
       - text: 模型
+  - button "打开配置文件"
   - button "关闭":
     - img
     - text: 关闭

apps/web/tests/snapshots/onboarding-deepseek-config/models.expected.md

@@ -0,0 +1,71 @@
+- dialog "设置":
+  - navigation:
+    - text: 设置
+    - button "通用设置":
+      - img
+      - text: 通用设置
+    - button "模型":
+      - img
+      - text: 模型
+  - button "打开配置文件"
+  - button "关闭":
+    - img
+    - text: 关闭
+  - heading "模型" [level=2]
+  - paragraph: 填入各提供方的 API 密钥即可使用其模型。
+  - list:
+    - listitem:
+      - text: DeepSeek
+      - button "编辑"
+      - text: DeepSeek deepseek-official API 密钥
+      - textbox "API 密钥":
+        - /placeholder: 已配置——输入新值可替换
+      - group:
+        - text: 自定义设置 API 地址
+        - textbox "API 地址":
+          - /placeholder: https://api.deepseek.com
+        - text: 推理强度
+        - combobox "推理强度":
+          - option "默认" [selected]
+          - option "off"
+          - option "high"
+          - option "max"
+        - region "模型目录":
+          - text: 模型目录 已自定义模型目录
+          - button "恢复默认模型"
+          - textbox "模型 ID 1":
+            - /placeholder: 模型 ID
+            - text: deepseek-v4-pro
+          - textbox "显示名称 1":
+            - /placeholder: 显示名称
+            - text: DeepSeek-V4-Pro
+          - button "容量 1":
+            - img
+          - button "删除模型 1":
+            - img
+          - textbox "模型 ID 2":
+            - /placeholder: 模型 ID
+            - text: private-preview
+          - textbox "显示名称 2":
+            - /placeholder: 显示名称
+            - text: Private Preview
+          - button "容量 2" [expanded]:
+            - img
+          - button "删除模型 2":
+            - img
+          - text: 上下文窗口
+          - textbox "上下文窗口 2":
+            - /placeholder: 1M
+            - text: "131072"
+          - text: 最大输出 token 数
+          - textbox "最大输出 token 数 2":
+            - /placeholder: 256K
+            - text: 64K
+          - button "添加模型":
+            - img
+            - text: 添加模型
+      - button "取消"
+      - button "保存"
+  - button "添加提供方":
+    - img
+    - text: 添加提供方

apps/web/tests/snapshots/seeded-history/command-row.expected.md

@@ -47,8 +47,8 @@
 - button "Commands":
   - img
 - 'button "Access mode, current: Read Only"': Read Only
-- button "Select model, current deepseek-v4-flash":
-  - text: deepseek-v4-flash
+- button "Select model":
+  - text: Select model
   - img
 - button "Send message" [disabled]
 - text: 1 turns · 2 steps Tool call {{duration}} Cache hit 98% Input 15.8K tok · Output 135 tok

apps/web/tests/snapshots/seeded-history/ui.expected.md

@@ -45,8 +45,8 @@
 - button "Commands":
   - img
 - 'button "Access mode, current: Workspace Write"': Workspace Write
-- button "Select model, current deepseek-v4-flash":
-  - text: deepseek-v4-flash
+- button "Select model":
+  - text: Select model
   - img
 - button "Send message" [disabled]
 - text: 1 turns · 2 steps Tool call {{duration}} Cache hit 98% Input 15.8K tok · Output 135 tok

apps/web/tests/snapshots/settings-chrome/dialog.expected.md

@@ -7,6 +7,7 @@
     - button "模型":
       - img
       - text: 模型
+  - button "打开配置文件"
   - button "关闭":
     - img
     - text: 关闭

apps/web/tests/support.ts

@@ -88,6 +88,28 @@ export async function connectFreshWorkspace(page: Page, root: string, name = 'wo
     .waitFor({ timeout: 15_000 })
 }
 
+/**
+ * {@link connectFreshWorkspace} over the product default Chinese locale: the
+ * English helper's anchors assume the locale every other scenario boots, so a
+ * scenario that deliberately keeps zh needs the localized picker copy.
+ * @param page - the browser page under test.
+ * @param root - workspace parent directory.
+ * @param name - directory created under `root` and connected.
+ */
+export async function connectFreshWorkspaceZh(page: Page, root: string, name = 'workspace'): Promise<void> {
+  mkdirSync(join(root, name), { recursive: true })
+  await page.getByRole('button', { name: '选择工作区' }).click()
+  const dialog = page.getByRole('dialog', { name: '选择工作区目录' })
+  await dialog.waitFor({ timeout: 10_000 })
+  await dialog.getByRole('button', { name: '编辑路径' }).click()
+  const pathInput = dialog.getByRole('textbox', { name: '编辑路径' })
+  await pathInput.fill(join(root, name))
+  await pathInput.press('Enter')
+  await dialog.getByRole('button', { name: '打开', exact: true }).click()
+  await page.locator('textarea:enabled[placeholder="描述你想要构建的内容"]')
+    .waitFor({ timeout: 15_000 })
+}
+
 /** Failure evidence goes to the gitignored .artifacts/ (repo convention). */
 export async function saveFailureShot(page: Page, name: string): Promise<void> {
   const dir = fileURLToPath(new URL('../../../.artifacts', import.meta.url))

docs/AGENTS.md

@@ -1,10 +1,20 @@
 # AGENTS.md — The documentation standard
 
-This file defines Markdown tiers, writing rules, and `verify-doc-budgets` ceilings. Use [dsh-doc-standards](../.agents/skills/dsh-doc-standards/SKILL.md) for placement and validation, and [dsh-prose-standard](../.agents/skills/dsh-prose-standard/SKILL.md) for required coverage and editorial judgment; the [doc-tiers Agent Note](../.agents/notes/implemented/process/2026-07-04-doc-tiers-and-budgets.md) owns rationale.
+This file defines document structure, Markdown tiers, writing rules, and `verify-doc-budgets` ceilings. Use [dsh-doc-standards](../.agents/skills/dsh-doc-standards/SKILL.md) for placement and validation, and [dsh-prose-standard](../.agents/skills/dsh-prose-standard/SKILL.md) for required coverage and editorial judgment; the [doc-tiers Agent Note](../.agents/notes/implemented/process/2026-07-04-doc-tiers-and-budgets.md) owns rationale.
+
+## Document structure
+
+These rules apply to human-facing documentation; [Agent Notes](../.agents/notes/README.md) remain outside their scope. A [postmortem](postmortem/README.md) is an incident-scoped reference; chronology records evidence, not a teaching sequence. A document's subject and tree position fix its scope: describe its own subject at appropriate detail, describe direct children only by purpose, responsibility, and high-level behavior, and link to the owning descendant for lower-level detail. Document type does not widen that scope. A reference may be exhaustive only about its own subject. Testing mechanisms, fixtures, and harnesses belong at the lowest owning level; higher documents link there.
+
+Classify every in-scope document as a tutorial or reference. A tutorial follows an ordered path to an outcome and introduces only what each step needs. A reference defines a lookup scope and describes current behavior without depending on a teaching sequence. Separate substantial tutorial and reference content; use a clear structural boundary when either part is small.
+
+Before writing a tutorial, privately classify the reader's starting knowledge and each concept as beginner, intermediate, or advanced. Establish prerequisites before dependent concepts, increase difficulty gradually, and move unnecessary advanced material to a later tutorial or reference.
+
+Author in this order: locate the document in the tree; set its permitted detail; choose tutorial or reference; for a tutorial, order concepts by prerequisite and difficulty; relocate descendant-owned detail; replace lower-level explanations with links to their owners.
 
 ## The tier taxonomy: one home per fact
 
-Each fact has one home: the tier whose job it is. Elsewhere, link to that home; `verify-md-links` keeps links resolving while duplicated prose drifts.
+Each fact has one home: the tier whose job it is. Elsewhere, link to that home.
 
 | Tier | Job | Does NOT belong there |
 |---|---|---|
@@ -12,17 +22,15 @@ Each fact has one home: the tier whose job it is. Elsewhere, link to that home;
 | Subtree `AGENTS.md` (`packages/`, `examples/`, `docs/`, `.agents/notes/`) | Orders specific to that subtree | Repo-wide rules the root file already carries |
 | [architecture.md](architecture.md) | The system map: services, the loop, extension seams — read before changing `packages/` | Type shapes (→ core-data-structures), per-package detail (→ package READMEs), decision rationale (→ Agent Notes), implementation-status annotations |
 | [core-data-structures/](core-data-structures/core.md) | The type catalog: literal shapes and semantics of the spine and seam vocabulary | Behavior narration (→ architecture.md) |
-| [Agent Notes](../.agents/notes/README.md) | Active decision records: the why, what-was-given-up, and concise verification contract; `implemented/` notes describe shipped reality in present tense | Migration plans, acceptance-task checklists, fixture walkthroughs, and spec-speak ("should…") once the decision has shipped; archived notes are frozen history, never current authority |
+| [Agent Notes](../.agents/notes/README.md) | Decision records under their own lifecycle contract | Migration plans, checklists, and spec-speak once implemented; archived notes are frozen history |
 | [postmortem/](postmortem/README.md) | Incident stories — the only tier where war-story narrative belongs | — |
 | [cookbook/](cookbook/adding-a-package.md) | Step-by-step how-tos with numbered verify steps | Design rationale (→ the Agent Note each guide links) |
 | [user/](user/index.md) | Product-facing guides published by the documentation website | Generated reference tables, contributor procedures, decision history |
-| Package README | The per-package contract: config, semantics, limitations, extension points, and [Model Experience](cookbook/adding-a-package.md#4-write-the-package-readme) | JSDoc restatement, generated-catalog restatement (event/tool tables), other packages' concerns |
-| [development.md](development.md) | First-stop contributor onboarding: local setup, daily workflow, and CI shape at summary level; a bilingual pair under the [i18n contract](i18n/README.md) | Runtime/version rationale (→ Agent Notes), gate-by-gate enumerations that drift from `package.json` scripts |
+| Package README | Per-package config, semantics, limitations, extension points, and [Model Experience](cookbook/adding-a-package.md#4-write-the-package-readme) | JSDoc or catalog restatement, other packages' concerns |
+| [development.md](development.md) | Contributor onboarding: setup, daily workflow, and CI shape at summary level | Runtime rationale (→ Agent Notes), drifting gate inventories |
 | Generated catalogs: [cordis events](cordis-catalog/events.md), [cordis services](cordis-catalog/services.md), [Cordis core API](cordis-catalog/core/context.md), [tool-catalog](tool-catalog.md), [config-catalog](config-catalog.md), [persistence-catalog](persistence-catalog.md), [module-graph.md](module-graph.md) | Exhaustive enumerations regenerated from source, freshness-gated | Hand edits of any kind |
 | Skills (`.agents/skills/`) | Reusable workflows and specialized decision standards | Product and runtime contracts (→ docs or source) |
 
-Placement: bugs → postmortems; rationale → Agent Notes; procedures → cookbooks; type shapes → core data; package contracts → READMEs; standing orders → root `AGENTS.md` with a rationale link.
-
 ## Writing rules
 
 - **Document current state, not change history.** Avoid "previously/now/no longer", PRs, commits, and stack positions in durable prose; name the live mechanism. Put change stories in commits, PRs, Agent Notes, or postmortems.
@@ -31,7 +39,7 @@ Placement: bugs → postmortems; rationale → Agent Notes; procedures → cookb
 - **Fenced `ts` blocks must compile** (`doc-typecheck`); a pasted type declaration and its original JSDoc use ` ```ts type-equiv `, while a body-stripped public class declaration uses ` ```ts public-api `; register either in the manifest so neither can drift ([mechanics](development.md#documenting-types-verbatim-ts-type-equiv)).
 - **The [core-data-structures catalog](core-data-structures/core.md) updates in the same change** that reshapes a documented type. `verify-type-equiv` catches drifted pastes, not never-documented new types ([what counts as core](core-data-structures/core.md#what-counts-as-core)).
 - **Bilingual pairs update together**: editing either side obligates the counterpart and a re-record in the same change ([i18n contract](i18n/README.md)).
-- **Comments and JSDoc state complete contracts, not reasoning transcripts.** Preserve behavior, conditions, timing, modality, exceptions, consequences, and non-obvious orientation; delete implementation narration, test walkthroughs, review analysis, and code restatement. Keep the local contract and link to its owning rationale. Use [dsh-prose-standard](../.agents/skills/dsh-prose-standard/SKILL.md) for required coverage, decision rules, and examples.
+- **Comments and JSDoc state complete contracts, not reasoning transcripts.** Preserve behavior, timing, modality, exceptions, consequences, and non-obvious orientation; delete narration, test walkthroughs, review analysis, and code restatement. Keep the local contract and link its rationale. Use [dsh-prose-standard](../.agents/skills/dsh-prose-standard/SKILL.md) for details.
 - Your audience is professional programmers. Prefer concise and straight-forward English over metaphor. Do not overuse words like "gate", "vocabulary", "surface", "seams".
 
 ## Wordcount Budgets
@@ -44,20 +52,18 @@ When the gate goes red:
 2. **Condense** content that belongs here but can be shorter.
 3. **Raise** the ceiling only when the words truly need the space; justify the manifest diff in the PR. A too-low ceiling is a budget bug.
 
-Ceilings are guardrails, not reduction targets. Retain at least 5% headroom; lower a ceiling only when the document's durable contract still has room, and raise it when necessary content would otherwise be deleted. Targets: root `AGENTS.md` ≤ 1,600 words; `architecture.md` ≤ 1,800; each subtree `AGENTS.md` ≤ 600, except `packages/AGENTS.md` ≤ 650 and this file ≤ 1,250; `packages/README.md` ≤ 600. Review and the slop checklist govern unbudgeted tiers.
+Ceilings are guardrails, not reduction targets. At or below target, retain at least 5% headroom; above target, freeze the ceiling until relocation or condensation brings the document under target. Lower a ceiling only when the contract still has room, and raise it when content would otherwise be deleted. Targets: root `AGENTS.md` ≤ 1,600 words; `architecture.md` ≤ 1,800; subtree `AGENTS.md` ≤ 600, except `packages/AGENTS.md` ≤ 650 and this file ≤ 1,250; `packages/README.md` ≤ 600. Review governs unbudgeted tiers.
 
 ## The slop checklist
 
 Hunt these in any doc; the [dsh-doc-standards](../.agents/skills/dsh-doc-standards/SKILL.md) skill runs this list as an audit:
 
 - The same rule stated in more than one home. Grep a distinctive phrase; keep one home, convert the rest to links.
-- Narrated history: "previously", "now", "no longer", "used to", "renamed", "was moved", references to PRs or commits. State the current fact; the why belongs in an Agent Note, the story in a postmortem or git.
-- A war story told inline where a one-line rule plus a postmortem/Agent Note link would do.
+- Narrated history or war stories: "previously", "now", "no longer", "used to", "renamed", "was moved", PRs, or commits. State the current fact; link an Agent Note or postmortem when needed.
 - Implementation-status annotations in prose or diagrams ("implemented!", "future: …"). Status rots; the repo layout and package manifests carry it.
-- Hand-restating a generated catalog or JSDoc: event tables, tool arg tables, method signatures. Link instead.
-- Hand-maintained inventories of tests, packages, or implementation status when the tree or a generator is authoritative.
+- Hand-restated catalogs, JSDoc, or inventories of tests, packages, and status when source or a generator is authoritative.
 - Reasoning transcripts: step-by-step implementation narration, proof of obvious branches, test walkthroughs, or rejected local alternatives. Keep the resulting contract or durable rationale; delete the path used to derive it.
-- The same rationale repeated beside sibling methods. State it once at the owning seam or shared helper.
+- Rationale repeated beside sibling methods instead of once at the owning seam or helper.
 - Paragraph walls: one paragraph carrying several rules and parenthetical asides. Split it, or demote the detail to the linked home.
 - Emphasis inflation: bold, CAPS, or "critically" everywhere means nothing stands out. Reserve emphasis for the clause that changes behavior.
 - Spec-speak in `implemented/` Agent Notes: "should", migration plans, acceptance checklists. An implemented Agent Note describes what is, per the [implemented-note instructions](../.agents/notes/implemented/AGENTS.md).

docs/architecture.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/architecture.md
-architecture.md: ee50e249f8e588a3f92f57db929c6bbfb1c853dd
-architecture.zh.md: c2e596cd10c21be2bcaad11323277d04ef9e74a1
+architecture.md: 296aea2056294d08693845e1780a9ac268b21bfc
+architecture.zh.md: c5f87c967d95c8934e919c4c01bc4acd4d3a1a70

docs/architecture.md

@@ -68,9 +68,7 @@ Waterfalls are around-middleware: listeners delegate with `next()`; returning wi
 
 ## Default Loop Lifecycle
 
-A **session** is append-only. An ordinary **turn** claims one queued `send()` item; injection claims none. A successor awaits its predecessor's checkpoint but may share its `running` interval ([decision](../.agents/notes/implemented/simplification/2026-07-17-one-send-one-turn.md)). A turn ends when model or plugins stop it; a **step** is one model request plus tools. Quotes in the [sequence below](agent-lifecycle.md) mark durable events.
-
-Creation without an id mints `<config-id>-session-<uuid>`; `sessionId` resumes or creates, while `resumeSessionId` requires history. Resume restores lineage and delegation depth before publication. Setup failures emit `agent-loop/config-start-failed`; teardown is silent.
+A **session** is append-only. An ordinary **turn** claims one queued `send()` item; injection claims none. A turn ends when the model or plugins stop it; a **step** is one model request plus its tool calls. Agent and session publication happen only after private setup and resume state are ready. Quotes in the [sequence below](agent-lifecycle.md) mark durable events.
 
 ### Turn Flow
 
@@ -119,29 +117,19 @@ idle inject:
   do not open a turn or run the model
 ```
 
-Each step assembles ordered stable system sections, cache-safe dynamic contexts, tool schemas, and variables; unknown references fail the turn. `dsh-system-prompt` owns identity and persona; the loop supplies `provider`, `model`, and `cwd` ([prompt ownership](../.agents/notes/implemented/architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.md)).
-
-Admission-time and active-turn `inject()` stage for the next step; tool-time injection and post-tool `additionalContexts` settle after results. Steering shares the outbox but remains provisional until a request admits it. `steer()` returns a message-owned receipt: after `agent/step` and asynchronous prompt assembly succeed, the loop commits the stable batch, snapshots request history, opens `step/start`, then resolves its receipts as admitted with the turn and step; later arrivals wait. A turn-concluding tool result, broad cancellation, disposal, or a claimed idle-steering turn that never opens a step rejects affected receipts, while `cancel(..., { keepInbox: true })` and non-terminal routing preserve pending delivery. Idle `inject()` appends immediately without changing turn numbers; persistence drains eagerly.
-
-Before driver claim, `updateInbox()` may edit or remove a queued occurrence, or strictly transfer its immutable message into an open next-step window. That transfer ends the queued occurrence and accepts a new steering occurrence; a closed window leaves Queue unchanged. Direct `steer()` remains best-effort for newly submitted input and falls back to a waking follow-up outside the window ([decision](../.agents/notes/implemented/feature/2026-07-30-web-queue-steer-action.md)).
-
-Pruning precedes summaries; overflow retries require durable progress. `agent/request-error` may authorize one retry turn between failed-step and turn close; cancellation wins. Adapter-owned `retryPolicy` makes normal mode bounded; always mode delegates specialized recovery before retrying until success or cancellation ([compaction](../.agents/notes/implemented/architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md), [retry foundation](../.agents/notes/implemented/architecture/2026-06-21-bounded-llm-request-recovery.md), [provider policy](../.agents/notes/implemented/feature/2026-07-24-provider-retry-policies.md)).
+Each step assembles the prompt, tools, runtime context, adapter settings, and model history before recording its reconstruction boundary. Tool calls then run through the shared execution pipeline. `inject()` adds context without opening an idle turn; `steer()` targets a next-step admission window; queued input remains the source of ordinary turns. The generated [agent lifecycle](agent-lifecycle.md) owns exact event order, and the [agent-loop README](../packages/core/agent-loop/README.md) owns queue, steering, retry, and cancellation mechanics.
 
 ### Failure Boundaries
 
-Adapter failures close their step before `agent/request-error` receives the exact `Error`, normalized `LlmFailure`, and signal. A handled failure closes its turn and opens a retry turn from durable history without an idle notification; exhaustion leaves terminal `turn/end`. Failed chunks commit neither messages nor tool calls.
-
-Other failures use `agent/error`. Cancellation and disposal beat recovery. Before request-header commit, the turn signal cancels asynchronous model-capability preparation; undispatched tools get synthetic `tool/call`/`ABORTED_BEFORE_DISPATCH` pairs. Effective `cancel(cause)` emits its cause before queue clearing and abort; observers cannot veto; idle calls emit nothing. Durability records user or parent cancellation as `aborted`, teardown as `disposed`; teardown awaits quiescence. The cause affects reporting, not late result-context handling ([decision](../.agents/notes/implemented/architecture/2026-07-16-explicit-turn-cancellation.md)).
-
-Turn and step events are turn-enclosed. Idle `user/message` and standalone `compact/* { turn: null }` consume no turn; their lock-time markers may interleave with injection. Reload synthesizes interrupted turn ends; `session/end-seed` distinguishes stale compaction orphans from live locks. After close, only `agent/error` reports failures. Each turn has one [TurnEndReason](core-data-structures/session.md#why-a-turn-ended-turnendreasonmap).
+Adapter failures close their step before `agent/request-error` can authorize recovery from durable history. Other failures use `agent/error`; cancellation and disposal take precedence over recovery. Failed model attempts commit no assistant message or tool side effect. Turn closure is represented by one [TurnEndReason](core-data-structures/session.md#why-a-turn-ended-turnendreasonmap); the exact retry contract belongs to [LLM streaming](core-data-structures/llm-streaming.md).
 
 ### Agent Handles
 
-`ctx.agents` owns agents, returning `AgentHandle { agent, dispose() }`. Plugins use `send()` or `followup()`, receipt-bearing `steer()`, and `inject()` presets; [`reserveTurnAdmission()`](../packages/core/agent/README.md#agent-interface-typests) synchronously reserves idle for durable work without changing queued prompt identity. Await a steering receipt when request admission matters; best-effort UI steering may ignore it. `cancel()` and `whenIdle()` control lifecycle. Caller, factory, and consumer co-own teardown through one awaited disposer.
+`ctx.agents` owns agents and returns `AgentHandle { agent, dispose() }`. Plugins submit queued work, steering, or injected context through the [agent interface](../packages/core/agent/README.md#agent-interface-typests); cancellation, idleness, and teardown stay behind the same handle.
 
 ### Agent Scope
 
-Each agent owns scoped `agent.ctx`; shared storage overlays its tool, prompt, and command entries on globals while preserving domain views ([decision](../.agents/notes/implemented/architecture/2026-07-12-scoped-layers-store.md)). Scoped listeners filter dispatch; contributions unwind with awaited cleanup. `CreateAgentOptions.setup(agentCtx)` composes before publication and may return a synchronous commit that the factory invokes immediately before registry entry, after every setup await. Typed resolvers derive carrier checks from merged `Events` and `scopeTarget` ([semantic gates](../.agents/notes/implemented/process/2026-07-14-typescript-program-backed-semantic-gates.md)). Details: [agent scope](../.agents/notes/implemented/architecture/2026-07-08-agent-scope-contexts.md), [subagent composition](../.agents/notes/implemented/feature/2026-07-12-subagent-persona-tool-filter-and-depth.md). `AgentLoop` runs under `ctx.agents.withInitiator()`; private orchestration derives `agent.session`, but turn, step, signal, cwd, and authority stay explicit ([decision](../.agents/notes/implemented/architecture/2026-07-15-agent-initiator-scope.md)).
+Each agent owns scoped `agent.ctx`; shared storage overlays its tools, prompts, and commands on global contributions while scoped listeners filter dispatch. Setup composes before publication and cleanup unwinds contributions. The [agent-scope decision](../.agents/notes/implemented/architecture/2026-07-08-agent-scope-contexts.md) owns the detailed lifecycle.
 
 ## State
 
@@ -198,4 +186,4 @@ New behavior attaches to a documented extension point; a loop change updates thi
 | Fork a live session | call `ctx.sessions.fork(source, boundary?, childSessionId?)` |
 | Scope a registration to one agent | use its `agent.ctx` (see Agent Scope) |
 
-The [extension cookbook](cookbook/extension-cookbook.md) has plugin skeletons and the feature-to-seam map; guides cover [packages](cookbook/adding-a-package.md), [tools](cookbook/adding-a-tool.md), [LLM adapters](cookbook/adding-an-llm-adapter.md), and [vendored packages](cookbook/adding-a-vendored-package.md).
+The [extension cookbook](cookbook/extension-cookbook.md) has plugin skeletons; guides cover [packages](cookbook/adding-a-package.md), [tools](cookbook/adding-a-tool.md), [LLM adapters](cookbook/adding-an-llm-adapter.md), and [vendored packages](cookbook/adding-a-vendored-package.md).

docs/architecture.zh.md

@@ -68,9 +68,7 @@ waterfall（瀑布式事件）是环绕中间件：监听器通过 `next()` 委
 
 ## 默认循环生命周期
 
-**会话**采用仅追加方式。普通**轮次**领取一项已排队的 `send()` 输入；注入不领取输入。后续轮次会等待前一轮次的检查点，但可以与其共用同一个 `running` 区间（[决策](../.agents/notes/implemented/simplification/2026-07-17-one-send-one-turn.md)）。模型或插件停止轮次时，该轮次结束；一个**步骤**包含一次模型请求及其工具。[下文时序](agent-lifecycle.md)中的引号标记持久事件。
-
-创建时若未提供 id，流程会生成 `<config-id>-session-<uuid>`；`sessionId` 用于恢复或创建会话，而 `resumeSessionId` 要求已有历史。恢复流程在发布前还原沿袭关系和委托深度。初始化失败会发出 `agent-loop/config-start-failed`；拆卸过程保持静默。
+**会话**仅追加。普通**轮次**认领一个排队的 `send()` 项；注入不会认领。轮次在模型或插件停止时结束；一个**步骤**由一次模型请求及其工具调用组成。只有私有设置和恢复状态准备完毕后，系统才会发布 agent 与会话。[下文时序](agent-lifecycle.md)中的引号标记持久事件。
 
 ### 轮次流程
 
@@ -119,29 +117,19 @@ idle inject:
   do not open a turn or run the model
 ```
 
-每个步骤都会组装有序的稳定系统提示词片段、缓存安全的动态上下文、工具 schema 和变量；未知引用会使该轮次失败。`dsh-system-prompt` 负责身份和角色设定；循环提供 `provider`、`model` 和 `cwd`（[提示词归属](../.agents/notes/implemented/architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.md)）。
-
-接纳期间和活跃轮次内的 `inject()` 会为下一步骤暂存；工具执行期间的注入和工具执行后的 `additionalContexts` 会在结果记录完毕后落定。steering 与其共用 outbox，但在请求接纳前始终处于待准入状态。`steer()` 会返回归属于该消息的回执：`agent/step` 和异步提示词组装成功后，循环提交稳定批次、捕获请求历史并开启 `step/start`，再将其回执解析为已准入并附带轮次与步骤；后续消息继续等待。结束轮次的工具结果、广义取消、dispose（资源释放），以及已领取 idle-steering 消息却从未开启步骤的轮次，都会拒绝受影响的回执；`cancel(..., { keepInbox: true })` 和非终止型路由则保留待处理投递。空闲状态下的 `inject()` 会立即追加，且不改变轮次编号；持久化层会尽快排空。
-
-驱动器认领之前，`updateInbox()` 可以编辑或移除 queued 单次入队项，也可以严格地把其不可变消息转移到开放的 next-step 窗口。该转移会结束 queued 单次入队项，并接受一个新的 steering 单次入队项；窗口关闭时 Queue 保持不变。直接调用 `steer()` 时，对新提交的输入仍采用尽力而为的语义，并在窗口之外回退为会唤醒 agent 的后续轮次（[决策](../.agents/notes/implemented/feature/2026-07-30-web-queue-steer-action.md)）。
-
-裁剪先于摘要；溢出重试必须取得持久进展。`agent/request-error` 可以在失败步骤与轮次关闭之间授权一个重试轮次；取消优先。适配器拥有的 `retryPolicy` 使 normal mode 保持有界；always mode 先委托专门恢复，再持续重试直至成功或取消（[压缩](../.agents/notes/implemented/architecture/2026-07-10-after-call-compaction-pressure-and-overflow-recovery.md)、[重试基础](../.agents/notes/implemented/architecture/2026-06-21-bounded-llm-request-recovery.md)、[提供方策略](../.agents/notes/implemented/feature/2026-07-24-provider-retry-policies.md)）。
+每个步骤都会组装提示词、工具、运行时上下文、适配器设置和模型历史，随后记录其重建边界。之后，工具调用通过共享执行流水线运行。`inject()` 添加上下文但不打开空闲轮次；`steer()` 针对下一步骤的准入窗口；排队输入仍是普通轮次的来源。精确事件顺序由生成的 [agent 生命周期](agent-lifecycle.md)定义；队列、steering、重试与取消机制由 [agent-loop README](../packages/core/agent-loop/README.md)定义。
 
 ### 失败边界
 
-适配器故障会先关闭自身步骤，再由 `agent/request-error` 接收准确的 `Error`、标准化的 `LlmFailure` 和信号。已处理的失败会关闭所在轮次，并从持久历史开启重试轮次，不发出空闲通知；重试耗尽则留下终态 `turn/end`。失败分片既不提交消息，也不提交工具调用。
-
-其他故障使用 `agent/error`。取消和资源释放优先于恢复。在提交请求头之前，轮次信号会取消异步模型能力准备；尚未分派的工具会得到合成的 `tool/call`/`ABORTED_BEFORE_DISPATCH` 对。实际生效的 `cancel(cause)` 在清空队列和中止前发出原因；观察方不能否决；空闲调用不发事件。持久化层将用户或父级取消记录为 `aborted`，拆卸记录为 `disposed`；拆卸会等待完全停稳。原因只影响报告方式，不影响延迟完成的结果上下文处理（[决策](../.agents/notes/implemented/architecture/2026-07-16-explicit-turn-cancellation.md)）。
-
-轮次和步骤事件均位于轮次边界内。空闲 `user/message` 与独立的 `compact/* { turn: null }` 不占用轮次；其锁定时刻标记可以与注入交错。重新加载会为中断的轮次合成结束事件；`session/end-seed` 区分陈旧的压缩遗留项与活跃锁。关闭后仅由 `agent/error` 报告故障。每个轮次有一个 [TurnEndReason](core-data-structures/session.md#why-a-turn-ended-turnendreasonmap)。
+适配器故障会先关闭步骤，再由 `agent/request-error` 授权从持久历史恢复。其他故障使用 `agent/error`；取消和资源释放优先于恢复。失败的模型尝试不会提交 assistant 消息或工具副作用。轮次关闭由一个 [TurnEndReason](core-data-structures/session.md#why-a-turn-ended-turnendreasonmap)表示；准确的重试契约由 [LLM 流式输出](core-data-structures/llm-streaming.md)定义。
 
 ### Agent 句柄
 
-`ctx.agents` 拥有 agent，返回 `AgentHandle { agent, dispose() }`。插件使用 `send()` 或 `followup()`、带回执的 `steer()` 和 `inject()` 预设；[`reserveTurnAdmission()`](../packages/core/agent/README.md#agent-interface-typests) 为持久工作同步预留空闲状态，同时不改变排队提示词身份。需要确认请求准入时应等待 steering 回执；尽力执行的 UI steering 可以忽略它。`cancel()` 与 `whenIdle()` 控制生命周期。调用方、工厂和消费方通过同一个需等待完成的 disposer 共同拥有拆卸过程。
+`ctx.agents` 管理 agent，并返回 `AgentHandle { agent, dispose() }`。插件通过 [agent 接口](../packages/core/agent/README.md#agent-interface-typests)提交排队工作、steering 或注入上下文；取消、空闲状态和拆卸也都由同一个句柄封装。
 
 ### Agent 作用域
 
-每个 agent 都拥有作用域化的 `agent.ctx`；共享存储会将其工具、提示词和命令条目叠加到全局条目之上，同时保留各领域视图（[决策](../.agents/notes/implemented/architecture/2026-07-12-scoped-layers-store.md)）。作用域监听器会过滤分派；贡献都会在撤销时等待清理完成。`CreateAgentOptions.setup(agentCtx)` 在发布前完成组合，并可返回一个同步提交操作；所有 setup 的 await 均完成后，工厂会在进入注册表前立即调用该操作。类型化解析器从合并后的 `Events` 和 `scopeTarget` 推导载体检查（[语义门禁](../.agents/notes/implemented/process/2026-07-14-typescript-program-backed-semantic-gates.md)）。详情见 [agent 作用域](../.agents/notes/implemented/architecture/2026-07-08-agent-scope-contexts.md)和 [subagent 组合](../.agents/notes/implemented/feature/2026-07-12-subagent-persona-tool-filter-and-depth.md)。`AgentLoop` 在 `ctx.agents.withInitiator()` 内运行；私有编排会派生 `agent.session`，但轮次、步骤、信号、cwd 和权限仍保持显式（[决策](../.agents/notes/implemented/architecture/2026-07-15-agent-initiator-scope.md)）。
+每个 agent 都拥有作用域化的 `agent.ctx`；共享存储会将其工具、提示词和命令叠加到全局贡献之上，作用域监听器则过滤分派。设置过程在发布前完成组合，清理过程会撤销贡献。详细生命周期由 [agent 作用域决策](../.agents/notes/implemented/architecture/2026-07-08-agent-scope-contexts.md)定义。
 
 ## 状态
 
@@ -198,4 +186,4 @@ idle inject:
 | fork 活跃会话 | 调用 `ctx.sessions.fork(source, boundary?, childSessionId?)` |
 | 将注册项限定到单个 agent | 使用其 `agent.ctx`（参见 Agent 作用域） |
 
-[扩展实操手册（cookbook）](cookbook/extension-cookbook.md)提供插件骨架和功能到服务边界的映射；指南涵盖[包](cookbook/adding-a-package.md)、[工具](cookbook/adding-a-tool.md)、[LLM 适配器](cookbook/adding-an-llm-adapter.md)和 [vendored 包](cookbook/adding-a-vendored-package.md)。
+[扩展实操手册（cookbook）](cookbook/extension-cookbook.md)提供插件骨架；指南涵盖[包](cookbook/adding-a-package.md)、[工具](cookbook/adding-a-tool.md)、[LLM 适配器](cookbook/adding-an-llm-adapter.md)和 [vendored 包](cookbook/adding-a-vendored-package.md)。

docs/config-catalog.md

@@ -108,7 +108,7 @@ export interface Config {
 
 Depends on: [`AgentOptions`](core-data-structures/core.md) · [`SessionId`](core-data-structures/core.md)
 
-Source: [`packages/core/agent-loop/src/index.ts:211`](../packages/core/agent-loop/src/index.ts)
+Source: [`packages/core/agent-loop/src/index.ts:212`](../packages/core/agent-loop/src/index.ts)
 
 ## `@deepseek-ai/dsh-agent-spine-demo`
 
@@ -228,7 +228,7 @@ export interface Config {
 }
 ```
 
-Source: [`packages/bash/bash-local/src/index.ts:39`](../packages/bash/bash-local/src/index.ts)
+Source: [`packages/bash/bash-local/src/index.ts:40`](../packages/bash/bash-local/src/index.ts)
 
 ## `@deepseek-ai/dsh-bash-sandbox`
 
@@ -247,7 +247,7 @@ export type Config = LocalConfig
 
 Depends on: [`LocalConfig`](#deepseek-aidsh-bash-local)
 
-Source: [`packages/bash/bash-sandbox/src/index.ts:27`](../packages/bash/bash-sandbox/src/index.ts)
+Source: [`packages/bash/bash-sandbox/src/index.ts:35`](../packages/bash/bash-sandbox/src/index.ts)
 
 ## `@deepseek-ai/dsh-cli-demo`
 
@@ -650,7 +650,7 @@ export interface Config {
   thinking?: 'enabled' | 'disabled'
   /** Default thinking effort (default `high`); `off` disables thinking per request. */
   reasoningEffort?: 'off' | 'high' | 'max'
-  /** Default per-request output cap (default 256,000); explicit request values win. */
+  /** Default per-request output cap (default 256,000); a model's own cap and explicit request values win. */
   maxTokens?: number
   /** Positive context capacity used when the selected model has no exact value (default 1,000,000). */
   defaultContextWindow?: number
@@ -672,6 +672,8 @@ export interface DeepSeekCatalogModel {
   description?: string
   /** Known combined request/response context capacity; omitted when deployment metadata is unavailable. */
   contextWindow?: number
+  /** Per-request output cap for this model; omission falls back to the profile's {@link DeepSeekConnectionOptions.maxTokens}. */
+  maxTokens?: number
 }
 ```
 
@@ -1062,17 +1064,18 @@ export interface Config {
   /**
    * Override the runner argv; bwrap-shaped profile arguments are appended. A
    * non-empty override asserts full enforcement and skips built-in selection and
-   * probing; a broken runner then fails at execution and must be identifiable by
-   * {@link runnerFailureSignatures}.
+   * probing. A runner that starts but refuses its profile must be identifiable by
+   * {@link runnerFailureSignatures}. Consumers classify spawn rejection; only
+   * attributable `ENOENT` or `EACCES` with runner argv[0] provenance becomes an
+   * infrastructure failure.
    */
   runnerCommand?: string[]
   /**
    * Case-insensitive stderr substrings emitted when a configured
    * {@link runnerCommand} refuses its profile before executing the wrapped
    * command. Required and non-empty with `runnerCommand`; rejected without
-   * it. Missing/unexecutable runner errors are added automatically from
-   * `runnerCommand[0]`, while these signatures cover an executable runner's
-   * own failure dialect.
+   * it. Each entry is a non-empty, single-line, case-insensitive substring
+   * covering the executable runner's own failure dialect.
    */
   runnerFailureSignatures?: string[]
   /** Positive timeout for each functional probe; zero would mean unbounded to Node. */
@@ -1080,7 +1083,7 @@ export interface Config {
 }
 ```
 
-Source: [`packages/sandbox/sandbox-local/src/index.ts:19`](../packages/sandbox/sandbox-local/src/index.ts)
+Source: [`packages/sandbox/sandbox-local/src/index.ts:24`](../packages/sandbox/sandbox-local/src/index.ts)
 
 ## `@deepseek-ai/dsh-sandbox-policy`
 

docs/cookbook/adding-a-package.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/cookbook/adding-a-package.md
-adding-a-package.md: 5df22b23a69e2a33a5d963970c771088ed7f3b5d
-adding-a-package.zh.md: e573657988bd67f5ff9b9191550c10bc654ef116
+adding-a-package.md: a45b222f6aed905a18ef9b480c989e6045029afe
+adding-a-package.zh.md: af0e4d0779fa99ce43ebccba00c33eab16c4d362

docs/cookbook/adding-a-package.md

@@ -14,7 +14,6 @@ packages/<group>/<pkg>/
                    # ../../../vendor/cordis (+ ../../../vendor/schemastery if
                    # you use Config, + ../../<group>/<dep> for each dsh dep)
   src/index.ts     # service default export or plugin (name/inject/apply/Config)
-  tests/<x>.spec.ts
   README.md        # service API, events, extension points, design notes,
                    # + gated Model Experience context blocks or short form
                    # + the gated "Known Limitations and Deferred Work" section
@@ -33,11 +32,11 @@ In-package relative imports use explicit `.ts` specifiers in source (for example
 |---|---|
 | `tsconfig.base.json` | no edit for an existing group; for a new group, add a `./packages/<group>/*/src` candidate to the `@deepseek-ai/dsh-*` wildcard |
 | `tsconfig.host.json` (host-side package) or `tsconfig.client.json` (client-side package) | add `{ "path": "./packages/<group>/<pkg>" }` to `references` — exactly one aggregate, never both ([layout](../development.md#typescript-project-layout)) |
-| `knip.json` | only if the package has non-`*.spec.ts` entries (e.g. `*.e2e.ts` → add a per-workspace override like `packages/llm/llm-deepseek`) |
+| `knip.json` | only if the package has entrypoints that repository discovery does not already cover |
 
 A `packages/client/*` package additionally extends `tsconfig.base.client.json` instead of `tsconfig.base.json`, and a client plugin package declares `dshClient` in package.json, exports `./client`, and calls the shared tsdown preset (`packages/client/tsdown.client.ts`) — see [packages/client/AGENTS.md](../../packages/client/AGENTS.md) for the client-side contract.
 
-Covered automatically by globs or package-manifest discovery — no edits needed: root `package.json` workspaces, `scripts/publint-all.ts`, `tsdown.config.ts`, `vitest.config.ts`, `.oxlintrc.json`, `scripts/check-workspace-constraints.ts`.
+Covered automatically by globs or package-manifest discovery — no edits needed: root `package.json` workspaces, `scripts/publint-all.ts`, `tsdown.config.ts`, `.oxlintrc.json`, `scripts/check-workspace-constraints.ts`.
 
 ## 3. Decide the package topology
 
@@ -85,8 +84,7 @@ A package with no context effect or one consumer-owned path uses the audited `No
 pnpm install        # registers the workspace
 pnpm run doc-sync
 pnpm run constraints && pnpm run typecheck && pnpm run lint
-pnpm run test:coverage  # 100% per-file over src (types.ts exempt)
 pnpm run build && pnpm run hygiene
 ```
 
-Test expectations: every registry/registration needs an HMR-safety test (register from a child fiber, dispose it, assert cleanup). Excessive tests are welcome — see [docs/testing.md](../testing.md).
+Follow the [repository testing policy](../testing.md) for the behavior-specific checks and coverage required by the new package.

docs/cookbook/adding-a-package.zh.md

@@ -14,7 +14,6 @@ packages/<group>/<pkg>/
                    # ../../../vendor/cordis (+ ../../../vendor/schemastery if
                    # you use Config, + ../../<group>/<dep> for each dsh dep)
   src/index.ts     # service default export or plugin (name/inject/apply/Config)
-  tests/<x>.spec.ts
   README.md        # service API, events, extension points, design notes,
                    # + gated Model Experience context blocks or short form
                    # + the gated "Known Limitations and Deferred Work" section
@@ -33,11 +32,11 @@ package.json 不变式（由 `pnpm run constraints` / `scripts/check-workspace-c
 |---|---|
 | `tsconfig.base.json` | 已有分组无需编辑；新分组需为 `@deepseek-ai/dsh-*` 通配符添加 `./packages/<group>/*/src` 候选路径 |
 | `tsconfig.host.json`（host 侧包）或 `tsconfig.client.json`（client 侧包） | 在 `references` 中添加 `{ "path": "./packages/<group>/<pkg>" }`——恰好一个聚合，绝不两个都加（[布局](../development.md#typescript-project-layout)） |
-| `knip.json` | 仅当包有非 `*.spec.ts` 入口时需要（如 `*.e2e.ts` → 添加 per-workspace override，参照 `packages/llm/llm-deepseek`） |
+| `knip.json` | 仅当包有仓库发现机制尚未覆盖的入口时需要 |
 
 `packages/client/*` 包改为 extends `tsconfig.base.client.json`（而非 `tsconfig.base.json`）；client 插件包还需在 package.json 声明 `dshClient`、导出 `./client`、调用共享 tsdown preset（`packages/client/tsdown.client.ts`）——client 侧见 [packages/client/AGENTS.md](../../packages/client/AGENTS.md)。
 
-以下内容由 glob 或包 manifest 发现机制自动覆盖，无需手动编辑：根 `package.json` workspaces、`scripts/publint-all.ts`、`tsdown.config.ts`、`vitest.config.ts`、`.oxlintrc.json`、`scripts/check-workspace-constraints.ts`。
+以下内容由 glob 或包 manifest 发现机制自动覆盖，无需手动编辑：根 `package.json` workspaces、`scripts/publint-all.ts`、`tsdown.config.ts`、`.oxlintrc.json`、`scripts/check-workspace-constraints.ts`。
 
 ## 3. 确定包拓扑
 
@@ -85,8 +84,7 @@ Append-only, prefix-stable, replacing, or independent behavior, including the ex
 pnpm install        # registers the workspace
 pnpm run doc-sync
 pnpm run constraints && pnpm run typecheck && pnpm run lint
-pnpm run test:coverage  # 100% per-file over src (types.ts exempt)
 pnpm run build && pnpm run hygiene
 ```
 
-测试要求：每个注册表/注册操作都需要一个 HMR（热模块替换）安全测试（从子 fiber 注册，dispose（资源释放）它，断言清理完成）。鼓励编写充分的测试——见 [docs/testing.md](../testing.md)。
+请遵循[仓库测试政策](../testing.md)，为新包运行行为所需的专项检查并达到相应覆盖率。

docs/cookbook/adding-a-tool.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/cookbook/adding-a-tool.md
-adding-a-tool.md: 0688e4a46b1eb2ca282a7ee970546a34b3c1cc68
-adding-a-tool.zh.md: e12d5e23a93d08f15fb0d1ac40229c5c4ab26e38
+adding-a-tool.md: cb418a9118901cc6572fb17125351bdda922434b
+adding-a-tool.zh.md: d8b73a51d1dde8647e7a032648f8a6344fcc83eb

docs/cookbook/adding-a-tool.md

@@ -1,8 +1,8 @@
-# Cookbook: adding a tool
+# Tool authoring reference
 
 English | [中文](adding-a-tool.zh.md)
 
-How to give the model a new capability. The minimal shape below shows the contract; `packages/bash/tool-bash` is the production-grade three-package seam.
+Reference for the contracts a model-facing tool must satisfy. For an ordered first tool, follow [Build a tool](../user/develop/basic/tool.md). `packages/bash/tool-bash` is the production-grade three-package example.
 
 ## The minimal shape
 
@@ -35,7 +35,7 @@ export function apply(ctx: Context) {
 }
 ```
 
-Registration is effect-based: disposing the plugin fiber unregisters the tool (write the HMR test). Schemas flow into the system-prompt assembly automatically.
+Registration is effect-based: disposing the plugin fiber unregisters the tool. Schemas flow into the system-prompt assembly automatically.
 
 ## Rules of the execute() contract
 
@@ -89,6 +89,6 @@ Hard rules (they bite if broken):
 
 The neutral vocabulary lives in `dsh-tools`; tools never import a UI or transport type. Host/client runtimes map each `card` into their own view. The design and the why are in [the render-intent-union Agent Note](../../.agents/notes/implemented/architecture/2026-07-02-tool-render-intent-union.md); `dsh-tool-fs` (generic/diff) and `dsh-tool-bash` (terminal) are the reference implementations.
 
-## Tests every tool needs
+## Verification
 
-Cover argument rejection, every canonical value and Native rendering shape, output-schema rejection, and HMR disposal. For a side-effecting tool, drive the real tool through the agent loop with a scripted `MockAdapter` and assert its `tool/call` and projected `tool/result` session events; prove the canonical value itself is not persisted. For a UI card, assert the exact `presentCall` and `presentResult` views and exercise the owning host/client projection. Add an assembled snapshot for the shipped model or UI behavior the tool changes.
+Follow the [repository testing policy](../testing.md) and the owning package's test documentation. A shipped model- or UI-visible change requires the assembled coverage specified there.

docs/cookbook/adding-a-tool.zh.md

@@ -1,8 +1,8 @@
-# 实操手册：添加工具
+# 工具编写参考
 
 [English](adding-a-tool.md) | 中文
 
-如何为模型赋予一项新能力。下文的最小形态展示这项契约；`packages/bash/tool-bash` 是生产级、由三个包（package）构成的 seam。
+面向模型的工具必须满足哪些契约，均以本文为准。如需按步骤构建第一个工具，请阅读[构建工具](../user/develop/basic/tool.md)。`packages/bash/tool-bash` 是生产级的三包示例。
 
 ## 最小形态
 
@@ -35,7 +35,7 @@ export function apply(ctx: Context) {
 }
 ```
 
-注册基于副作用：dispose（资源释放）插件 fiber 即注销该工具（请编写 HMR（热模块替换）测试）。schema 会自动流入系统提示词的组装过程。
+注册基于副作用：dispose（资源释放）插件 fiber 即注销该工具。schema 会自动流入系统提示词的组装过程。
 
 ## execute() 契约的规则
 
@@ -89,6 +89,6 @@ producer 提供同步的 `cancel`、在资源清理后 settle 且不 reject 的
 
 中性词汇定义在 `dsh-tools` 中；工具绝不导入 UI 或传输类型。host/client 运行时将每个 `card` 映射到各自的视图。设计与原因见[渲染意图联合体 Agent Note](../../.agents/notes/implemented/architecture/2026-07-02-tool-render-intent-union.md)；`dsh-tool-fs`（generic/diff）和 `dsh-tool-bash`（terminal）是参考实现。
 
-## 每个工具必须的测试
+## 验证
 
-覆盖参数拒绝、每种规范值和 Native 渲染形态、输出 schema 拒绝以及 HMR dispose。对于有副作用的工具，使用脚本化的 `MockAdapter` 驱动真实工具通过 agent loop（智能体循环），并断言其 `tool/call` 和投影后的 `tool/result` 会话事件；同时证明规范值本身未被持久化。对于 UI 卡片，断言 `presentCall` 和 `presentResult` 的精确视图，并实际运行所属 host/client 投影。如果工具改变了已交付的模型或 UI 行为，请添加组装应用快照。
+遵循[仓库测试策略](../testing.md)和所属包的测试文档。已交付且面向模型或 UI 的变更必须提供其中规定的组装覆盖。

docs/cookbook/adding-a-vendored-package.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/cookbook/adding-a-vendored-package.md
-adding-a-vendored-package.md: a951a96f62d2ea3aa693a24d83bf46a1a12070cd
-adding-a-vendored-package.zh.md: 66b71d9ac7901d40074a21213c4962d1db96db8f
+adding-a-vendored-package.md: b85d74a3a09b27254883b88cb8e6587e32ed811c
+adding-a-vendored-package.zh.md: 2927837a28d1e7b593090d581522f69f84504806

docs/cookbook/adding-a-vendored-package.md

@@ -53,7 +53,7 @@ Covered automatically by globs — no edits needed: root `package.json` workspac
 ```sh
 pnpm install        # registers the workspace
 pnpm run typecheck
-pnpm run build && pnpm run test && pnpm run constraints
+pnpm run build && pnpm run constraints
 ```
 
-The source `paths` map lives once in `tsconfig.base.json` and serves every graph. The important isolation boundary is the project-reference graph: vendored source must be referenced through its own `vendor/<dir>/tsconfig.json`, not pulled into an aggregate's strict program ([layout](../development.md#typescript-project-layout)).
+Run the behavior checks selected by the [testing policy](../testing.md). The source `paths` map lives once in `tsconfig.base.json` and serves every graph. The important isolation boundary is the project-reference graph: vendored source must be referenced through its own `vendor/<dir>/tsconfig.json`, not pulled into an aggregate's strict program ([layout](../development.md#typescript-project-layout)).

docs/cookbook/adding-a-vendored-package.zh.md

@@ -53,7 +53,7 @@ vendored TypeScript 源码中的本地相对导入/导出在复制后使用显
 ```sh
 pnpm install        # registers the workspace
 pnpm run typecheck
-pnpm run build && pnpm run test && pnpm run constraints
+pnpm run build && pnpm run constraints
 ```
 
-源码 `paths` 映射只在 `tsconfig.base.json` 存在一份，服务所有图。重要的隔离边界是 project-reference 图：vendored 源码必须通过其自身的 `vendor/<dir>/tsconfig.json` 被引用，而非被拉入某个聚合的严格程序中（[布局](../development.md#typescript-project-layout)）。
+请运行[测试政策](../testing.md)所选择的行为检查。源码 `paths` 映射只在 `tsconfig.base.json` 存在一份，服务所有图。重要的隔离边界是 project-reference 图：vendored 源码必须通过其自身的 `vendor/<dir>/tsconfig.json` 被引用，而非被拉入某个聚合的严格程序中（[布局](../development.md#typescript-project-layout)）。

docs/cookbook/adding-an-llm-adapter.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/cookbook/adding-an-llm-adapter.md
-adding-an-llm-adapter.md: a7f9dced70041653a0cb815147a07b6386d79e3e
-adding-an-llm-adapter.zh.md: 494115bdd5cc2365feb0964f2bef4d6dc070d3f6
+adding-an-llm-adapter.md: 4fcc646ed2eea8a6170027d01761887aa28b0045
+adding-an-llm-adapter.zh.md: 35a671416f8160a6187a06f3dbd614dfe4faa778

docs/cookbook/adding-an-llm-adapter.md

@@ -34,13 +34,10 @@ Registration is effect-based (HMR-safe); one adapter per provider route — dupl
 
 Provider-specific thinking-mode toggles remain in the adapter's Config. Exact model metadata uses one provider-neutral capability seam: implement `resolveModel()` with provider/model identity and optional `context` and `reasoning` fields, declare a configured `defaultEffort` only when one exists, and honor the resolver's optional `AbortSignal`. Reasoning efforts are ordered opaque ids mapped to provider requests by the adapter. Preserve the adapter's authoritative selectable list, including an adapter-defined `off` when supported, without exposing final wire spellings or clamping unsupported values; an id need not equal its wire representation.
 
-## Structure that worked
+## Implementation structure
 
-Split the adapter into testable stages (llm-deepseek's layout): wire types (`types.ts`, coverage-exempt) → request serializer → SSE/transport parser → chunk-translation state machine → a thin adapter class wiring them. Each stage gets its own unit suite.
+Keep wire types, request serialization, transport parsing, chunk translation, and the adapter class as separate responsibilities; [`llm-deepseek`](../../packages/llm/llm-deepseek/README.md) is the reference layout.
 
-## Testing
+## Verification
 
-- **Unit: mock the provider, not the harness.** A scripted `node:http` server speaking the provider's wire format covers happy paths, every error status, malformed payloads, premature closes, and aborts — no network, and it drives the 100% per-file coverage gate. Works for SDK-backed adapters too (point the SDK's baseURL at the mock).
-- **Hostile framing tests.** Split stream payloads at arbitrary byte positions (including mid-UTF-8) — real networks do.
-- **E2E: `tests/*.e2e.ts`** under `pnpm run test:e2e`, gated with `describe.skipIf(!process.env.MY_KEY)` so CI (no secrets) stays green. Cover representative model/provider/API families and every provider mode you map, a tool-call round trip INCLUDING the follow-up turn with results in history, and loose assertions only (substring/structure, bounded maxTokens — real models are nondeterministic).
-- Register the e2e file pattern in `knip.json` (per-workspace `entry` override) or knip flags it unused.
+Follow the [repository testing policy](../testing.md), which owns adapter coverage, real-provider checks, and published-entry requirements.

docs/cookbook/adding-an-llm-adapter.zh.md

@@ -34,13 +34,10 @@ export function apply(ctx: Context, config: Config) {
 
 提供方特有的思考模式开关仍放在适配器的 Config 中。确切模型元数据使用一处提供方无关的能力 seam：实现 `resolveModel()`，返回提供方/模型身份以及可选的 `context` 和 `reasoning` 字段；仅当存在配置指定的默认值时才声明 `defaultEffort`；遵守解析模型时传入的可选 `AbortSignal`。推理（reasoning）强度是由适配器映射到提供方请求的有序不透明 ID。请保留适配器给出的权威可选列表，包括适配器在支持时定义的 `off`；不得暴露最终协议值的具体拼写，也不得自动调整不支持的值。ID 无需与其协议表示相同。
 
-## 经验证有效的结构
+## 实现结构
 
-将适配器拆分为可测试的阶段（llm-deepseek 的布局）：协议格式（wire format）类型（`types.ts`，豁免覆盖率）→ 请求序列化器 → SSE/传输解析器 → 分片转换状态机 → 一个将它们串联的薄适配器类。每个阶段配备独立的单元测试套件。
+让协议类型、请求序列化、传输解析、分片转换和适配器类分别承担独立职责；[`llm-deepseek`](../../packages/llm/llm-deepseek/README.md) 是参考布局。
 
-## 测试
+## 验证
 
-- **单元测试：mock 提供方，而非 harness。** 用脚本化的 `node:http` 服务器模拟提供方的协议格式，覆盖正常路径、所有错误状态码、畸形载荷、连接提前关闭和中止——无需网络，且能满足 100% 逐文件覆盖率门禁。对基于 SDK 的适配器同样适用（将 SDK 的 baseURL 指向 mock 服务器）。
-- **恶意分帧测试。** 在任意字节位置（包括 UTF-8 字符中间）切割流载荷——真实网络环境正是如此。
-- **e2e：`tests/*.e2e.ts`**，通过 `pnpm run test:e2e` 运行，以 `describe.skipIf(!process.env.MY_KEY)` 守卫，确保无密钥的 CI 保持绿色。覆盖具有代表性的模型/提供方/API 系列以及你映射的每种提供方模式、一次务必包含后续轮次（历史中带工具结果）的工具调用往返，以及仅做宽松断言（子串/结构匹配、有界的 maxTokens——真实模型是非确定性的）。
-- 在 `knip.json` 中注册 e2e 文件模式（per-workspace `entry` 覆盖），否则 knip 会将其标记为未使用。
+遵循[仓库测试策略](../testing.md)，该策略负责适配器覆盖、真实提供方检查和已发布入口要求。

docs/cookbook/extension-cookbook.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/cookbook/extension-cookbook.md
-extension-cookbook.md: 07073c39f8a9b998b09b0815257d995b174c8be7
-extension-cookbook.zh.md: 10664af9a39f0a1663c316869ba8ef02f67c29fe
+extension-cookbook.md: 66cdee676936029e4db03bf0c1a0436d58f29bcd
+extension-cookbook.zh.md: 7e29d34b5777a4a4b19c8d3da2fc078df519c6d4

docs/cookbook/extension-cookbook.md

@@ -2,9 +2,7 @@
 
 English | [中文](extension-cookbook.zh.md)
 
-> FIXME: This important guide has not received sufficient human design review; complete that review before the first release.
-
-The three plugin shapes you write against the harness extension surface, as illustrative snippets (elided imports and helper stubs — not copy-paste-complete). For the full step-by-step guides see [adding a package](adding-a-package.md), [adding a tool](adding-a-tool.md), and [adding an LLM adapter](adding-an-llm-adapter.md); for the seams these hook into see [docs/architecture.md](../architecture.md).
+Reference shapes for the harness extension surface. The snippets omit imports and helper implementations and are not copy-paste-complete. For concrete authoring paths, see the [package checklist](adding-a-package.md), [first-tool tutorial](../user/develop/basic/tool.md), [tool reference](adding-a-tool.md), and [LLM adapter guide](adding-an-llm-adapter.md); the [architecture](../architecture.md) owns the system and extension-seam map.
 
 ## A tool plugin
 
@@ -92,38 +90,3 @@ export function apply(ctx: Context) {
 ## Runnable wirings
 
 Runnable leaves load their plugin trees from `examples/*/cordis.yml`; the root `demo:*` scripts and those leaf directories are the authoritative inventory. Non-interactive leaves use [`@deepseek-ai/dsh-cli-demo`](../../packages/examples/cli-demo), ACP leaves use [`@deepseek-ai/dsh-acp-demo`](../../packages/examples/acp-demo), JSON-RPC leaves use [`@deepseek-ai/dsh-jsonrpc-demo`](../../packages/examples/jsonrpc-demo), and the app packages share [`@deepseek-ai/dsh-agent-spine-demo`](../../packages/examples/agent-spine-demo).
-
-## The feature → mechanism map
-
-Every product feature maps to a listener on a documented extension seam — the microkernel claim made checkable ([microkernel Agent Note](../../.agents/notes/implemented/architecture/2026-06-11-microkernel-event-taxonomy.md)). No row modifies the loop.
-
-`system-prompt/assemble` is an expert cooperative whole-assembly transform: its returned assembly is authoritative, so listener authors own preserving active Code Mode and structured-output protocol contributions. Prefer `ctx.tools.restrict()` for tool filtering that must stay aligned across presentation, lookup, and execution.
-
-| Product feature | Plugin mechanism |
-|---|---|
-| Hook system (user + project level) | listeners on `agent/session-start`, `agent/prompt-submit`, `agent/request`, `tools/pre-execute`, `tools/post-execute`, and `agent/turn-stopping`; the waterfall seams return typed decisions, while `agent/turn-stopping` may steer another step; the `dsh-hooks-claude` / `dsh-hooks-codex` bridges map hook config files onto these seams |
-| `/goal` | `ctx.goals` owns durable state, `dsh-goal-session` schedules same-session rounds through the public `Agent`, and separate command/tool producers expose human/model control |
-| `/loop` | on the `turn/end` session event, `followup()` the next iteration; or force-continue |
-| Dynamic workflow | `ctx.workflows` + the worker-thread engine + the `workflow` tool; structured in-process children enforce output with scoped prompt/tool registrations, a monotonic tool guard, final `tools/result` commit (including enclosing `run_code`), and the structured-output execution's monotonic `concludeTurn()` marker |
-| Queued + steering messages | core `Agent.followup()` / `Agent.steer()` |
-| Context compaction (auto + manual) | the `ctx.compact` seam + `dsh-compact-basic`; automatic pressure runs on serial `agent/step`, canonical overflow recovery runs on `agent/request-error`, and manual callers use the same compact service ([compaction Agent Note](../../.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md) — the model-facing `/compact` consumer tool is deferred) |
-| System prompt configurability | `ctx.systemPrompt.section()` with ordering and scope-local shadowing |
-| AGENTS.md (root) | a section provider reading the file |
-| AGENTS.md (subdir, on-touch) + file-change notices | `agent.inject()` from a watcher / tool-result listener |
-| Built-in tools | `ctx.tools.register()`; schemas flow into the assembly automatically — the `dsh-tool-*` families (bash, fs, web, subagent, todo) are the shipped examples |
-| ToolSearch / progressive disclosure | replace a scoped `ctx.tools.restrict()` registration as the visible set changes; the registry keeps presentation, lookup, and execution aligned |
-| Tool deadline / retry / metrics | wrap core dispatch with `tools/execute`; a wrapper may replace `exec.signal`, delegate, and inspect the normalized result in one lexical lifetime |
-| Final tool-result metrics / audit / capture | observe immutable authoritative outcomes with `tools/result`; use `tools/post-execute` instead only when the plugin must transform the result or attach context |
-| Monotonic terminal turn policy | call `ToolExecution.concludeTurn()` from the successful terminal tool; later tool calls in the same response remain guardable, and the loop stops after the step |
-| Subprocess sandbox (landlock / sandbox-exec) | use a `ctx.sandbox` backend through `dsh-bash-sandbox`; use `tools/pre-execute` for capability-level denial |
-| Permission system / AskUserQuestion | return `ask` from `tools/pre-execute` and answer through `ctx.approval`; register a separate model-facing ask tool for ordinary user questions |
-| Plan mode | Shipped: [`@deepseek-ai/dsh-plan-mode`](../../packages/plan/plan-mode/README.md) — logged `plan/mode` state, the `plan:policy` guidance section, `/plan [message]` entry, `/plan off` direct exit, and the user-reviewed `exit_plan_mode` exit; enforcement stays on the independent sandbox/approval axes |
-| Sub-agent delegation | the `ctx.subagents` provider registry (`dsh-subagent-spawn`/`-fork`/`-acp`) + `dsh-tool-subagent` exposing one configured provider to the model |
-| MCP | one plugin per server: discover tools → `ctx.tools.register()` |
-| Skills | section + tool registration; `inject()` skill content on invocation |
-| Memory | section provider + tool |
-| Scheduled tasks (cron) | a plugin registers model-callable scheduling tools; timer fires → `followup(…, {source: {kind: 'cron', …}})` when idle / `inject()` notification when busy |
-| UI (GUI; CLI emits JSONL) | listen `session/event` (assistant chunks, boundaries, tool activity); input → `followup()` |
-| Telemetry / replayable trace | `session/event` → JSONL; replay = `sessions.create(id, { seed })` |
-| Model adapters | `LlmAdapter` subclass via `registerAdapter` (`dsh-llm-deepseek`, `dsh-llm-pi-ai`) |
-| Plugin hot-reload | every registration is a `ctx.effect` → vendored HMR just works |

docs/cookbook/extension-cookbook.zh.md

@@ -2,9 +2,7 @@
 
 [English](extension-cookbook.md) | 中文
 
-> FIXME：这篇重要指南尚未经过充分的人工设计审查；请在首次发布前完成审查。
-
-针对 harness 扩展表面编写的三种插件形态，以示意性代码片段呈现（省略了 import 和辅助桩——不可直接复制运行）。完整的分步指南见[添加包（package）](adding-a-package.md)、[添加工具](adding-a-tool.md)和[添加 LLM（大语言模型）适配器](adding-an-llm-adapter.md)；这些插件所挂接的 seam 见 [docs/architecture.md](../architecture.md)。
+harness 扩展表面的参考形态。代码片段省略了 import 和辅助实现，无法直接复制运行。具体编写路径见[包检查清单](adding-a-package.md)、[第一个工具教程](../user/develop/basic/tool.md)、[工具参考](adding-a-tool.md)和 [LLM（大语言模型）适配器指南](adding-an-llm-adapter.md)；系统与扩展 seam 映射由[架构文档](../architecture.md)负责。
 
 ## 工具插件
 
@@ -92,38 +90,3 @@ export function apply(ctx: Context) {
 ## 可运行的组装示例
 
 可运行叶子从 `examples/*/cordis.yml` 加载各自的插件树；根目录的 `demo:*` 脚本和这些叶子目录是权威清单。非交互式叶子使用 [`@deepseek-ai/dsh-cli-demo`](../../packages/examples/cli-demo)，ACP 叶子使用 [`@deepseek-ai/dsh-acp-demo`](../../packages/examples/acp-demo)，JSON-RPC 叶子使用 [`@deepseek-ai/dsh-jsonrpc-demo`](../../packages/examples/jsonrpc-demo)，应用包共享 [`@deepseek-ai/dsh-agent-spine-demo`](../../packages/examples/agent-spine-demo)。
-
-## 功能→机制映射
-
-每个产品功能都映射到一个文档化扩展 seam 上的监听器——微内核声明由此可验证（[微内核 Agent Note](../../.agents/notes/implemented/architecture/2026-06-11-microkernel-event-taxonomy.md)）。没有任何一行修改循环本身。
-
-`system-prompt/assemble` 是一个专家协作式的整体装配变换：其返回的装配结果具有权威性，因此监听器作者有责任保留活跃的 Code Mode 和结构化输出协议的贡献。对于需要在展示、查找和执行之间保持对齐的工具过滤，优先使用 `ctx.tools.restrict()`。
-
-| 产品功能 | 插件机制 |
-|---|---|
-| 钩子系统（用户级 + 项目级） | `agent/session-start`、`agent/prompt-submit`、`agent/request`、`tools/pre-execute`、`tools/post-execute` 和 `agent/turn-stopping` 上的监听器；waterfall seam 返回类型化决策，`agent/turn-stopping` 则可通过 steering 触发下一步；`dsh-hooks-claude` / `dsh-hooks-codex` 桥接器将钩子配置文件映射到这些 seam 上 |
-| `/goal` | `ctx.goals` 管理持久状态，`dsh-goal-session` 通过公共 `Agent` 调度同会话回合，独立的命令/工具生产方分别提供人类/模型控制 |
-| `/loop` | 在 `turn/end` 会话事件上 `followup()` 下一次迭代；或强制继续 |
-| 动态工作流 | `ctx.workflows` + worker-thread 引擎 + `workflow` 工具；结构化的进程内子任务通过作用域化的 prompt/工具注册、单调工具守卫、最终 `tools/result` 提交（包括外层 `run_code`）和结构化输出执行的单调 `concludeTurn()` 标记来强制输出 |
-| 排队消息 + steering（中途引导） | 核心 `Agent.followup()` / `Agent.steer()` |
-| 上下文压缩（context compaction）（自动 + 手动） | `ctx.compact` seam + `dsh-compact-basic`；自动压力检查运行在串行 `agent/step`，规范化溢出恢复运行在 `agent/request-error`，手动调用方使用同一个压缩服务（[压缩 Agent Note](../../.agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md)——面向模型的 `/compact` 消费方工具已推迟） |
-| 系统提示词可配置性 | `ctx.systemPrompt.section()`，支持排序与作用域局部覆盖 |
-| AGENTS.md（根目录） | 一个读取该文件的 section provider |
-| AGENTS.md（子目录，按需触发）+ 文件变更通知 | 从 watcher / tool-result 监听器调用 `agent.inject()` |
-| 内置工具 | `ctx.tools.register()`；schema 自动流入装配——`dsh-tool-*` 系列（bash、fs、web、subagent、todo）是已交付的示例 |
-| ToolSearch / 渐进式披露 | 当可见集变化时替换一个作用域化的 `ctx.tools.restrict()` 注册；注册表保持展示、查找和执行三者对齐 |
-| 工具截止时间 / 重试 / 指标 | 用 `tools/execute` 包裹核心分发；包装器可替换 `exec.signal`、委托执行，并在同一词法生命周期内检视规范化结果 |
-| 最终工具结果指标 / 审计 / 捕获 | 用 `tools/result` 观察不可变的权威结果；仅当插件需要变换结果或附加上下文时才使用 `tools/post-execute` |
-| 单调终端轮次策略 | 从成功的终端工具调用 `ToolExecution.concludeTurn()`；同一响应中后续工具调用仍可由守卫阻止，循环在该步骤后停止 |
-| 子进程沙箱（landlock / sandbox-exec） | 通过 `dsh-bash-sandbox` 使用 `ctx.sandbox` 后端；能力级别的拒绝使用 `tools/pre-execute` |
-| 权限系统 / AskUserQuestion | 从 `tools/pre-execute` 返回 `ask` 并通过 `ctx.approval` 应答；为普通用户提问注册一个独立的面向模型的 ask 工具 |
-| Plan mode | 已交付：[`@deepseek-ai/dsh-plan-mode`](../../packages/plan/plan-mode/README.md) — 落日志的 `plan/mode` 状态、`plan:policy` 引导段、`/plan [message]` 入口、`/plan off` 直接退出，以及经用户评审的 `exit_plan_mode` 出口；强制约束留在独立的沙箱/审批轴上 |
-| 子 agent 委派 | `ctx.subagents` 提供方注册表（`dsh-subagent-spawn`/`-fork`/`-acp`）+ `dsh-tool-subagent` 向模型暴露一个已配置的提供方 |
-| MCP | 每个服务器一个插件：发现工具 → `ctx.tools.register()` |
-| Skill（技能） | section + 工具注册；调用时通过 `inject()` 注入 skill 内容 |
-| 记忆 | section provider + 工具 |
-| 定时任务（cron） | 插件注册面向模型的调度工具；定时器触发 → 空闲时 `followup(…, {source: {kind: 'cron', …}})`／忙碌时 `inject()` 通知 |
-| UI（GUI；CLI 输出 JSONL） | 监听 `session/event`（助手分片、边界、工具活动）；输入 → `followup()` |
-| 遥测 / 可回放 trace | `session/event` → JSONL；回放 = `sessions.create(id, { seed })` |
-| 模型适配器 | 通过 `registerAdapter` 注册 `LlmAdapter` 子类（`dsh-llm-deepseek`、`dsh-llm-pi-ai`） |
-| 插件热重载 | 每个注册都是一个 `ctx.effect` → vendor 的 HMR（热模块替换）直接生效 |

docs/cordis-catalog/events.md

@@ -398,7 +398,7 @@ A declarative agent entry failed before it could publish a live agent. Consumers
 
 Types: [SessionId](../core-data-structures/core.md)
 
-Source: [`packages/core/agent-loop/src/index.ts:157`](../../packages/core/agent-loop/src/index.ts)
+Source: [`packages/core/agent-loop/src/index.ts:158`](../../packages/core/agent-loop/src/index.ts)
 
 ## `approval/*`
 

docs/cordis-catalog/services.md

@@ -44,7 +44,7 @@ async resume(ownerCtx: Context, options: ResumeAgentOptions): Promise<AgentHandl
 
 Types: [Agent](../core-data-structures/core.md) · [AgentOptions](../core-data-structures/core.md) · [SessionHeader](../core-data-structures/persistence.md) · [SessionId](../core-data-structures/core.md)
 
-Source: [`packages/core/agent-loop/src/index.ts:252`](../../packages/core/agent-loop/src/index.ts)
+Source: [`packages/core/agent-loop/src/index.ts:253`](../../packages/core/agent-loop/src/index.ts)
 
 ## `ctx.agents` — `AgentRegistry`
 
@@ -1103,7 +1103,7 @@ abstract confine(argv: readonly string[], policy: SandboxPolicy): ConfinedArgv
 
 Types: [ConfinedArgv](../core-data-structures/sandbox.md) · [SandboxPolicy](../core-data-structures/sandbox.md)
 
-Source: [`packages/sandbox/sandbox/src/index.ts:131`](../../packages/sandbox/sandbox/src/index.ts)
+Source: [`packages/sandbox/sandbox/src/index.ts:148`](../../packages/sandbox/sandbox/src/index.ts)
 
 ## `ctx.sandboxPolicy` — `SandboxPolicyService`
 
@@ -1681,7 +1681,7 @@ fork(source: SessionForkSource, boundary?: number, childSessionId?: SessionId):
 
 Types: [CreateSessionOptions](../core-data-structures/persistence.md) · [Session](../core-data-structures/session.md) · [SessionId](../core-data-structures/core.md)
 
-Source: [`packages/core/session/src/index.ts:767`](../../packages/core/session/src/index.ts)
+Source: [`packages/core/session/src/index.ts:796`](../../packages/core/session/src/index.ts)
 
 ## `ctx.sessionTitle` — `SessionTitleService`
 
@@ -1735,6 +1735,14 @@ Source: [`packages/session-title/session-title/src/index.ts:261`](../../packages
 Abstract settings service. Providers implement raw-document storage (`load`/`persist`) and push external changes through Settings.publish; the base class owns namespace registration, resolution, validation, change detection, and the `settings/updated` commit event.
 
 ```ts cordis-catalog
+/**
+ * Prepare the provider's user-editable document for a native editor. File
+ * providers may materialize an absent document before returning its path;
+ * non-file providers return undefined.
+ * @returns the absolute local document path, or undefined for non-file storage.
+ */
+prepareDocument(): Promise<string | undefined>
+
 /**
  * Register a namespace schema and receive its owner scope. The registration
  * is an effect on the calling plugin's fiber: disposing that fiber removes

docs/core-data-structures/llm-streaming.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/core-data-structures/llm-streaming.md
-llm-streaming.md: 28f775f81049d3aa0198e75fbb6c4544302e772d
-llm-streaming.zh.md: db6b28acb844033025b3c5ffef075c9c3153151e
+llm-streaming.md: 8f5b917ee19044eeef70aadf68cff752b8fba76a
+llm-streaming.zh.md: 10e04f0856a96f86c2145cdca51edb9e28020e67

docs/core-data-structures/llm-streaming.md

@@ -64,10 +64,10 @@ Every adapter MUST obey these, and every consumer may rely on them:
 - **Provider stalls are bounded at the transport.** Both shipping remote adapters expose positive finite `streamIdleTimeoutMs` with a five-minute default. The watchdog arms only while iterator `next()` is outstanding, uses one stable signal for the whole request, maps its own expiry to `TIMEOUT`, and keeps an earlier caller abort as `ABORTED`.
 - **Context overflow has one canonical code.** Both DeepSeek adapters classify explicit provider detail through `isContextWindowExceededError()` and surface `CONTEXT_WINDOW_EXCEEDED`, whether the failure arrives as a thrown HTTP `LlmError` or an in-band finish error. Consumers route on the code, never provider text.
 - **An empty completion is a retryable error, not a silent success.** Both adapters map a terminal `stop` finish that carried no content blocks to `finish {kind:'error'}` with the canonical `EMPTY_RESPONSE` code, and `dsh-llm-retry` retries it by default; see [empty model responses are retryable](../../.agents/notes/implemented/bug-fix/2026-07-24-empty-model-response-is-retryable.md).
-- **Every provider HTTP request carries the app-attribution header.** Adapters send `attributionHeaders()` (below) - the `User-Agent` baseline - and prove it with a wire-level test (mock server asserting the received header, or the library's header hook for a library-backed adapter).
+- **Every provider HTTP request carries the app-attribution header.** Adapters send `attributionHeaders()` (below), the `User-Agent` baseline.
 - **Replay state is adapter-owned.** A successful `finish` may carry lossless-JSON state needed to reconstruct a native provider response. The loop stores it with the assembled assistant message. On a later request, `LlmService` passes the state only when the historical provider and target provider are currently registered to the exact same adapter instance. That adapter validates the state and owns any cross-model or cross-provider conversion; other adapters receive the provider-neutral content and provenance without the private state.
 
-This contract is pinned down by two deliberately independent implementations: `dsh-llm-deepseek` (direct fetch, SSE framing via `eventsource-parser`) and `dsh-llm-pi-ai` (a generic multi-provider adapter through `@earendil-works/pi-ai`). The library-backed adapter exercises the finish-chunk error path, while transport-boundary tests prove each idle watchdog stops its actual request.
+Two independent implementations obey this contract: `dsh-llm-deepseek` uses direct fetch with SSE framing through `eventsource-parser`, while `dsh-llm-pi-ai` provides a generic multi-provider adapter through `@earendil-works/pi-ai`. Both carry cancellation and the idle watchdog to the provider request.
 
 ## `ResolvedRetryPolicy`
 

docs/core-data-structures/llm-streaming.zh.md

@@ -64,10 +64,10 @@ interface LlmFailure {
 - **提供方停顿在传输层受到时限约束。** 两个已交付的远程适配器都暴露正数且有限的 `streamIdleTimeoutMs`，默认五分钟。watchdog 只在 iterator `next()` 尚未完成时启动，整个请求使用同一个稳定 signal，把自身到期映射为 `TIMEOUT`，并把更早发生的调用方中止保留为 `ABORTED`。
 - **上下文溢出只有一个规范 code。** 两个 DeepSeek 适配器都通过 `isContextWindowExceededError()` 对提供方的显式细节分类并暴露 `CONTEXT_WINDOW_EXCEEDED`，无论失败以抛出的 HTTP `LlmError` 还是带内 finish error 到达。消费方按 code 路由，绝不依赖提供方文本。
 - **空 completion 是可重试错误，而不是静默的成功结果。** 两个适配器都把没有携带任何内容块的终止性 `stop` 结束映射为携带规范 `EMPTY_RESPONSE` code 的 `finish {kind:'error'}`，`dsh-llm-retry` 默认会重试它；详见[空模型响应可重试](../../.agents/notes/implemented/bug-fix/2026-07-24-empty-model-response-is-retryable.md)。
-- **每个提供方 HTTP 请求都携带应用归属头。** 适配器发送 `attributionHeaders()`（见下文）作为 `User-Agent` 基线，并通过协议级测试加以证明（mock 服务器断言收到的 header，或对基于库的适配器使用库的 header 钩子）。
+- **每个提供方 HTTP 请求都携带应用归属头。** 适配器发送下文的 `attributionHeaders()`，即 `User-Agent` 基线。
 - **回放状态归适配器所有。** 成功的 `finish` 可以携带重建提供方原生响应所需的无损 JSON 状态。循环会将其与组装后的 assistant 消息一起存储。后续请求中，仅当历史提供方与目标提供方当前注册到完全相同的适配器实例时，`LlmService` 才会传递该状态。该适配器负责校验状态并拥有所有跨模型或跨提供方转换；其他适配器只会收到提供方无关的内容与 provenance，不会收到私有状态。
 
-该契约由两个有意保持独立的实现锁定：`dsh-llm-deepseek`（直接 fetch，SSE（Server-Sent Events）分帧经由 `eventsource-parser`）和 `dsh-llm-pi-ai`（通过 `@earendil-works/pi-ai` 实现的通用多提供方适配器）。基于库的适配器覆盖 finish 分片错误路径，而传输边界测试证明每个空闲 watchdog 都会停止其实际请求。
+两个彼此独立的实现遵循该契约：`dsh-llm-deepseek` 使用直接 fetch，并通过 `eventsource-parser` 进行 SSE（Server-Sent Events）分帧；`dsh-llm-pi-ai` 则通过 `@earendil-works/pi-ai` 提供通用多提供方适配器。两者都会把取消与空闲 watchdog 传递至提供方请求。
 
 ## `ResolvedRetryPolicy`
 

docs/core-data-structures/persistence.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/core-data-structures/persistence.md
-persistence.md: 237ce268691fa6f4b0546c6c1c14c21eaa462612
-persistence.zh.md: 9cb555f2e157844704896d9b04feb822428540e5
+persistence.md: ed19af4e739c153ce1beec7c1e8de06f0c0bca53
+persistence.zh.md: 83d8ba3c0eabe57b9df661fb86ec05d3db98d8b6

docs/core-data-structures/persistence.md

@@ -4,7 +4,7 @@ English | [中文](persistence.zh.md)
 
 The **durability seam** for the event log. [session.md](session.md) describes the in-memory `Session` — the append-only `SessionEvent` log that is the source of truth. This page describes how that log is made durable: the abstract `SessionPersistence` service, its backends, the flush checkpoint, crash recovery, and the metadata header that travels alongside the log. The event vocabulary the log carries is enumerated, member by member, in the generated [persistence log event catalog](../persistence-catalog.md).
 
-The seam is a textbook [capability seam](../../.agents/notes/implemented/architecture/2026-06-13-capability-seams.md): one abstract service ([dsh-session-persistence](../../packages/session-persistence/session-persistence), `ctx.sessionPersistence`) defining locate/create/append, crash-repairing load, non-mutating inspect, and lightweight list/snapshot observation over the existing `SessionEvent` — **no parallel persisted type** — and two interchangeable backends that pass the same `runPersistenceContract` suite. See the [session-persistence Agent Note](../../.agents/notes/implemented/architecture/2026-06-14-session-persistence.md).
+The seam is a textbook [capability seam](../../.agents/notes/implemented/architecture/2026-06-13-capability-seams.md): one abstract service ([dsh-session-persistence](../../packages/session-persistence/session-persistence), `ctx.sessionPersistence`) defining locate/create/append, crash-repairing load, non-mutating inspect, and lightweight list/snapshot observation over the existing `SessionEvent` — **no parallel persisted type** — and two interchangeable backends implementing the same contract. See the [session-persistence Agent Note](../../.agents/notes/implemented/architecture/2026-06-14-session-persistence.md).
 
 ## The flush checkpoint
 

docs/core-data-structures/persistence.zh.md

@@ -4,7 +4,7 @@
 
 事件日志的**持久性 seam**。[session.md](session.md) 描述了内存中的 `Session`：仅追加的 `SessionEvent` 日志即为真源。本页描述如何使该日志持久化：抽象的 `SessionPersistence` 服务、它的后端、flush 检查点、崩溃恢复，以及随日志一同存储的元数据头。日志承载的事件词汇在生成的[持久化日志事件目录](../persistence-catalog.md)中逐项列举。
 
-该 seam 是典型的[能力 seam](../../.agents/notes/implemented/architecture/2026-06-13-capability-seams.md)：一个抽象服务（[dsh-session-persistence](../../packages/session-persistence/session-persistence)，`ctx.sessionPersistence`）在现有 `SessionEvent` 上定义 locate/create/append、会执行崩溃修复的 load、不会修改数据的 inspect，以及轻量的 list/snapshot 观察——**没有平行的持久化类型**——以及两个可互换、通过同一套 `runPersistenceContract` 的后端。见 [session-persistence Agent Note（agent 决策记录）](../../.agents/notes/implemented/architecture/2026-06-14-session-persistence.md)。
+该 seam 是典型的[能力 seam](../../.agents/notes/implemented/architecture/2026-06-13-capability-seams.md)：一个抽象服务（[dsh-session-persistence](../../packages/session-persistence/session-persistence)，`ctx.sessionPersistence`）在现有 `SessionEvent` 上定义 locate/create/append、会执行崩溃修复的 load、不会修改数据的 inspect，以及轻量的 list/snapshot 观察——**没有平行的持久化类型**——以及两个实现同一契约的可互换后端。见 [session-persistence Agent Note（agent 决策记录）](../../.agents/notes/implemented/architecture/2026-06-14-session-persistence.md)。
 
 ## flush 检查点
 

docs/core-data-structures/sandbox.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/core-data-structures/sandbox.md
-sandbox.md: 9bc05fa06f22fdc9ac9e8aacd482c1e7c2f2edec
-sandbox.zh.md: ff44db13115073f3a77d2bda575cc91ec41da418
+sandbox.md: 9e5feafe046f18dad49aeaf281793f0b8e03c240
+sandbox.zh.md: a1314d1b78eb0d46ea4c8ca5aa330ee83bf89132

docs/core-data-structures/sandbox.md

@@ -87,7 +87,27 @@ interface SandboxPolicy extends SandboxExecutionPolicy {
 
 ## Wrapped argv and classification dialects
 
-`ConfinedArgv` is what the consumer spawns. Besides the replacement argv, it carries the backend's enforcement fact and two orthogonal stderr dialects. `denialSignatures` identify the confined command being blocked while the sandbox works correctly. `runnerFailureSignatures` identify the sandbox runner refusing or failing before it executes the command; consumers check these first and surface a sandbox infrastructure failure, never an ordinary task failure.
+`RunnerFailureRule` combines evidence that a runner failed before executing the command. A consumer requires a nonzero exit, the optional allowed-exit-code gate, and a case-insensitive fatal signature within one remaining stderr line. Case-insensitive exact full-line informational exclusions are removed first, so a benign runner notice cannot prove failure by itself. The matched line remains available as error detail; classification does not rewrite stderr.
+
+```ts type-equiv
+/**
+ * Evidence that identifies a sandbox runner failing before it executes the
+ * wrapped command. A consumer first applies {@link allowedExitCodes} when
+ * present, removes {@link informationalLines} by case-insensitive exact line
+ * equality, then matches {@link fatalSignatures} case-insensitively within
+ * each remaining stderr line. Exit status alone never proves runner failure.
+ */
+interface RunnerFailureRule {
+  /** Nonzero process exit codes on which this rule may match; omitted permits any nonzero exit. */
+  allowedExitCodes?: readonly number[]
+  /** Non-empty substrings identifying a fatal runner diagnostic on one stderr line. */
+  fatalSignatures: readonly string[]
+  /** Benign stderr lines excluded by exact full-line equality before fatal matching. */
+  informationalLines?: readonly string[]
+}
+```
+
+`ConfinedArgv` is what the consumer spawns. Besides the replacement argv, it carries the backend's enforcement fact and two orthogonal stderr classifiers. `denialSignatures` identify the confined command being blocked while the sandbox works correctly. `runnerFailureRules` identify the sandbox runner refusing or failing before it executes the command; consumers check these first and surface a sandbox infrastructure failure, never an ordinary task failure.
 
 ```ts type-equiv
 /**
@@ -110,18 +130,19 @@ interface ConfinedArgv {
    */
   denialSignatures: readonly string[]
   /**
-   * Case-insensitive signatures for runner failure before command execution.
-   * Consumers check these before denial signatures: runner failure means the
+   * Structured runner-failure evidence rules. Consumers require a matching
+   * fatal stderr line (after informational exclusions) and any rule-specific
+   * exit-code gate before checking denial signatures: runner failure means the
    * command never ran, while denial means confinement worked and blocked it.
    */
-  runnerFailureSignatures: readonly string[]
+  runnerFailureRules: readonly RunnerFailureRule[]
 }
 ```
 
-An operator-configured local runner must supply at least one `runnerFailureSignatures` entry for its own pre-exec refusal dialect; the provider adds outer-shell missing and unexecutable forms automatically. This makes an executable custom runner rejecting its profile distinguishable from the wrapped command exiting with the same status.
+The [local provider](../../packages/sandbox/sandbox-local/README.md) owns operator configuration and maps its runner dialect into these rules. The [sandboxed bash consumer](../../packages/bash/bash-sandbox/README.md) owns spawn and result attribution.
 
 ## Provider and fail-closed errors
 
-`ctx.sandbox.confine(argv, policy)` returns a `ConfinedArgv` or throws `SandboxUnavailableError` with code `SANDBOX_UNAVAILABLE` when no usable backend exists. A selected runner can also fail closed at execution time, in which case its failure signature carries the same infrastructure meaning. Silent unconfined passthrough is never legal for a confined policy.
+`ctx.sandbox.confine(argv, policy)` returns a `ConfinedArgv` or throws `SandboxUnavailableError` with code `SANDBOX_UNAVAILABLE` when no usable backend exists. Consumers may also classify a failure while spawning or observing the returned argv; that attribution belongs to the consumer contract. Silent unconfined passthrough is never legal for a confined policy.
 
-Provider probing arbitrates between multiple candidates and is cached for the provider lifetime. A platform with one candidate may select it directly; execution-time refusal retains the safety property. The local provider reports bwrap and Seatbelt as full and preserves the Landlock launcher's full/partial kernel verdict.
+Provider selection, probing, caching, and backend-specific enforcement reports belong to the [local provider](../../packages/sandbox/sandbox-local/README.md).

docs/core-data-structures/sandbox.zh.md

@@ -87,7 +87,27 @@ interface SandboxPolicy extends SandboxExecutionPolicy {
 
 ## 包装后的 argv 与分类方言
 
-`ConfinedArgv` 是消费方实际 spawn 的内容。除了替换后的 argv，它还携带后端的强制执行事实和两种正交的 stderr 方言。`denialSignatures` 用于识别沙箱正常工作、受限命令被阻止的情况。`runnerFailureSignatures` 用于识别沙箱运行器在执行命令之前拒绝或失败的情况；消费方应先检查后者，将其作为沙箱基础设施故障上报，而非普通任务失败。
+`RunnerFailureRule` 汇集用于判定 runner 在执行命令前失败的证据。消费方要求进程以非零状态退出，并同时满足可选的允许退出码门控，以及余下某一 stderr 行中不区分大小写的致命签名。系统会先按不区分大小写的整行精确匹配移除信息性排除项，因此无害的 runner 通知本身不能证明失败。匹配到的行仍可用作错误详情；分类过程不会重写 stderr。
+
+```ts type-equiv
+/**
+ * Evidence that identifies a sandbox runner failing before it executes the
+ * wrapped command. A consumer first applies {@link allowedExitCodes} when
+ * present, removes {@link informationalLines} by case-insensitive exact line
+ * equality, then matches {@link fatalSignatures} case-insensitively within
+ * each remaining stderr line. Exit status alone never proves runner failure.
+ */
+interface RunnerFailureRule {
+  /** Nonzero process exit codes on which this rule may match; omitted permits any nonzero exit. */
+  allowedExitCodes?: readonly number[]
+  /** Non-empty substrings identifying a fatal runner diagnostic on one stderr line. */
+  fatalSignatures: readonly string[]
+  /** Benign stderr lines excluded by exact full-line equality before fatal matching. */
+  informationalLines?: readonly string[]
+}
+```
+
+`ConfinedArgv` 是消费方实际 spawn 的内容。除了替换后的 argv，它还携带后端的强制执行事实和两种正交的 stderr 分类器。`denialSignatures` 用于识别沙箱正常工作时受限命令被阻止的情况。`runnerFailureRules` 用于识别沙箱 runner 在执行命令之前拒绝或失败的情况；消费方应先检查后者，将其作为沙箱基础设施故障上报，而非普通任务失败。
 
 ```ts type-equiv
 /**
@@ -110,18 +130,19 @@ interface ConfinedArgv {
    */
   denialSignatures: readonly string[]
   /**
-   * Case-insensitive signatures for runner failure before command execution.
-   * Consumers check these before denial signatures: runner failure means the
+   * Structured runner-failure evidence rules. Consumers require a matching
+   * fatal stderr line (after informational exclusions) and any rule-specific
+   * exit-code gate before checking denial signatures: runner failure means the
    * command never ran, while denial means confinement worked and blocked it.
    */
-  runnerFailureSignatures: readonly string[]
+  runnerFailureRules: readonly RunnerFailureRule[]
 }
 ```
 
-运维人员配置的本地运行器必须为自身的 pre-exec 拒绝方言提供至少一条 `runnerFailureSignatures` 条目；提供方会自动添加外层 shell 报告的「命令不存在」和「不可执行」形式。这使得可执行的自定义运行器拒绝其 profile 的情况能够与被包装命令以相同状态码退出的情况区分开来。
+[本地提供方](../../packages/sandbox/sandbox-local/README.md)拥有运维配置，并将其 runner 方言映射到这些规则。[沙箱化 bash 消费方](../../packages/bash/bash-sandbox/README.md)拥有 spawn 与结果归因。
 
 ## 提供方与 fail-closed 错误
 
-`ctx.sandbox.confine(argv, policy)` 返回一个 `ConfinedArgv`，或在没有可用后端时抛出 `SandboxUnavailableError`（错误码 `SANDBOX_UNAVAILABLE`）。已选定的运行器也可能在执行时 fail-closed，此时其失败签名承载相同的基础设施含义。对于受限策略，静默的无隔离透传永远不合法。
+`ctx.sandbox.confine(argv, policy)` 返回一个 `ConfinedArgv`，或在没有可用后端时抛出 `SandboxUnavailableError`（错误码 `SANDBOX_UNAVAILABLE`）。消费方也可以在 spawn 或观察所返回的 argv 时对失败进行分类；该归因属于消费方契约。对于受限策略，静默的无隔离透传永远不合法。
 
-提供方探测在多个候选后端之间仲裁，结果在提供方生命周期内缓存。只有一个候选后端的平台可以直接选定它；执行时拒绝仍保留安全属性。本地提供方将 bwrap 和 Seatbelt 报告为 full，并保留 Landlock 启动器的 full/partial 内核裁定。
+提供方选择、探测、缓存和后端专有的强制执行报告归[本地提供方](../../packages/sandbox/sandbox-local/README.md)所有。

docs/core-data-structures/session.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/core-data-structures/session.md
-session.md: 6369c956df03c0d786db696c208b000300d5bfbc
-session.zh.md: 9a9a06ca3d58dfb3ef880868de5a41a009208528
+session.md: c5491b8d6b44c0a86ce5804320533925a0e6e287
+session.zh.md: 3a713c77e972096cf95223701e8aff4cb9d0ff87

docs/core-data-structures/session.md

@@ -351,13 +351,14 @@ interface SurfaceFoldResult {
 
 ## `Session` public API
 
-The body-stripped declaration keeps the plain class's public constructor, state accessors, append boundary, and history projections synchronized with source. Store operations remain in the generated [`ctx.sessions` service catalog](../cordis-catalog/services.md#ctxsessions--sessionstore).
+The body-stripped declaration keeps the plain class's detached factory, state accessors, append boundary, and history projections synchronized with source. Store operations remain in the generated [`ctx.sessions` service catalog](../cordis-catalog/services.md#ctxsessions--sessionstore).
 
 ```ts public-api
 /**
  * An event-sourced session: an append-only log of {@link SessionEvent}s.
  *
- * Plain class (not a Service) — create instances via `ctx.sessions.create()`.
+ * Plain class (not a Service) — create live instances via
+ * `ctx.sessions.create()` and detached instances via {@link create}.
  * Seeding with an existing event log replays/forks a session.
  * @typert object
  */
@@ -367,7 +368,7 @@ declare class Session {
   /**
    * Detached, deep-frozen creation metadata (format version, cwd, lineage,
    * seed boundary). Supplied by the store via `ctx.sessions.create()`. When a
-   * `Session` is constructed bare (tests, ad-hoc replay), a minimal header is
+   * `Session` is created without a store-owned header, a minimal header is
    * synthesized (stamped with the current {@link SESSION_FORMAT_VERSION}) so
    * `session.header` is always present. Kept out of the event log — it is a
    * storage concern, not replayable conversation state.
@@ -400,7 +401,15 @@ declare class Session {
    * holds an ordinary published write.
    */
   readonly firstLiveSeq: number;
-  constructor(id: SessionId, seed?: readonly SessionEvent[], header?: SessionHeader);
+  /**
+   * Create a detached session by validating and snapshotting borrowed seed
+   * events and storage metadata.
+   * @param id - session identity.
+   * @param seed - optional borrowed replay or fork events.
+   * @param header - optional borrowed storage metadata.
+   * @returns a detached session.
+   */
+  static create(id: SessionId, seed?: readonly SessionEvent[], header?: SessionHeader): Session;
   /**
    * An immutable snapshot of the append-only event log. The snapshot is reused
    * until the next append; a previously returned array does not grow later.

docs/core-data-structures/session.zh.md

@@ -353,13 +353,14 @@ interface SurfaceFoldResult {
 
 ## `Session` 公共 API
 
-去除方法体的声明与源码中的普通类保持同步，覆盖其公共构造函数、状态访问器、追加边界和历史投影。存储操作仍由生成的 [`ctx.sessions` 服务目录](../cordis-catalog/services.md#ctxsessions--sessionstore)记录。
+去除方法体的声明与源码中的普通类保持同步，覆盖其脱离态工厂、状态访问器、追加边界和历史投影。存储操作仍由生成的 [`ctx.sessions` 服务目录](../cordis-catalog/services.md#ctxsessions--sessionstore)记录。
 
 ```ts public-api
 /**
  * An event-sourced session: an append-only log of {@link SessionEvent}s.
  *
- * Plain class (not a Service) — create instances via `ctx.sessions.create()`.
+ * Plain class (not a Service) — create live instances via
+ * `ctx.sessions.create()` and detached instances via {@link create}.
  * Seeding with an existing event log replays/forks a session.
  * @typert object
  */
@@ -369,7 +370,7 @@ declare class Session {
   /**
    * Detached, deep-frozen creation metadata (format version, cwd, lineage,
    * seed boundary). Supplied by the store via `ctx.sessions.create()`. When a
-   * `Session` is constructed bare (tests, ad-hoc replay), a minimal header is
+   * `Session` is created without a store-owned header, a minimal header is
    * synthesized (stamped with the current {@link SESSION_FORMAT_VERSION}) so
    * `session.header` is always present. Kept out of the event log — it is a
    * storage concern, not replayable conversation state.
@@ -402,7 +403,15 @@ declare class Session {
    * holds an ordinary published write.
    */
   readonly firstLiveSeq: number;
-  constructor(id: SessionId, seed?: readonly SessionEvent[], header?: SessionHeader);
+  /**
+   * Create a detached session by validating and snapshotting borrowed seed
+   * events and storage metadata.
+   * @param id - session identity.
+   * @param seed - optional borrowed replay or fork events.
+   * @param header - optional borrowed storage metadata.
+   * @returns a detached session.
+   */
+  static create(id: SessionId, seed?: readonly SessionEvent[], header?: SessionHeader): Session;
   /**
    * An immutable snapshot of the append-only event log. The snapshot is reused
    * until the next append; a previously returned array does not grow later.

docs/defensive-patterns.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/defensive-patterns.md
-defensive-patterns.md: cc34877fb0d6a2e1740d8fa138f879363c8e69a3
-defensive-patterns.zh.md: 277e4904d033e26d379e0495cb2b22370ac159d0
+defensive-patterns.md: f9737cd6ba8d2bf1f926c962b3842885778f9af4
+defensive-patterns.zh.md: 26a8933401cef39287bab8a48a07527e0c86dc1f

docs/defensive-patterns.md

@@ -18,7 +18,7 @@ When an interface documents two valid ways to signal something — an adapter ma
 
 ## Dispose must reach quiescence, not just request it
 
-A teardown that issues kills/aborts but returns before the work stops leaves orphans. Make cleanup async and await the children's exit (kill → await `done`), and close listener/notification registries BEFORE killing so late completions stay silent. Tests prove disposal waited (pid gone right after `await fiber.dispose()`), not merely that the process eventually dies.
+A teardown that issues kills/aborts but returns before the work stops leaves orphans. Make cleanup async and await the children's exit (kill → await `done`), and close listener/notification registries BEFORE killing so late completions stay silent.
 
 ## Contain callback exceptions at the boundary
 

docs/defensive-patterns.zh.md

@@ -18,7 +18,7 @@
 
 ## Dispose 必须达到完全停稳，而不仅仅是请求停止
 
-如果清理流程只发出终止或中止信号便返回，而不等待工作真正停止，就会留下孤儿进程。清理逻辑应采用异步流程，并等待子进程退出（发出终止信号后等待 `done`）；还应在终止进程前关闭监听器和通知注册表，使迟到的完成事件保持静默。测试必须证明 dispose 确实等待了，例如 `await fiber.dispose()` 返回后进程 ID 已不存在，而不能只证明该进程最终会退出。
+如果清理流程只发出终止或中止信号便返回，而不等待工作真正停止，就会留下孤儿进程。清理逻辑应采用异步流程，并等待子进程退出（发出终止信号后等待 `done`）；还应在终止进程前关闭监听器和通知注册表，使迟到的完成事件保持静默。
 
 ## 在边界处隔离回调异常
 

docs/development.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/development.md
-development.md: 77087a95fdb4cfbeb02111ef2560d6989cde5686
-development.zh.md: 18e130e2f3aa1c53f6efa70366cdb629df71ad85
+development.md: 30a2bd0a2c97df8d3d75ec50f47b861b3a65590e
+development.zh.md: 5582a85429c97c3e31517a495c69392b80885f7d

docs/development.md

@@ -2,16 +2,18 @@
 
 English | [中文](development.zh.md)
 
-This onboarding guide helps project contributors get started with the local environment, daily workflow, and CI flow; see the Agent Notes for design rationale and technical trade-offs.
+The setup tutorial takes a new contributor from prerequisites to a checked checkout. The contributor reference that follows covers repository layout, daily workflow, and CI shape. Design rationale and implementation details belong to the linked Agent Notes and scripts.
 
-## Prerequisites
+## Setup tutorial
+
+### Prerequisites
 
 - Node.js supports 22.19+ and 24+. CI covers 22.19, 24, and 26; see the [Node engine floor Agent Note](../.agents/notes/implemented/process/2026-07-06-node-engine-floor.md).
 - Corepack-enabled pnpm. The repo pins `pnpm@11.7.0` in `package.json`; run `corepack enable` if `pnpm --version` does not resolve through Corepack.
 - Git 2.26 or newer; hook setup enables Git's worktree-specific configuration extension.
 - Optional: a DeepSeek API key for the Web, headless, and ACP automation demos and real-API e2e tests.
 
-## First-time setup
+### First-time setup
 
 Install dependencies from the repo root:
 
@@ -19,7 +21,7 @@ Install dependencies from the repo root:
 pnpm install
 ```
 
-The install also runs the root `postinstall` script, which installs lefthook from the repo dev dependency through `scripts/install-lefthook.mjs`. With `CI=true` or `GITHUB_ACTIONS=true`, the wrapper returns before Git discovery because automated jobs do not consume contributor hooks. Otherwise, it requires Git 2.26 or newer and gives the current worktree an explicit hook directory under its own Git directory; linked worktrees therefore use their own lefthook binary and configuration instead of rewriting common hooks. The first install enables Git's worktree-specific configuration extension and repository format 1; see the [worktree-local hooks Agent Note](../.agents/notes/implemented/process/2026-07-27-worktree-local-lefthook.md).
+The install also configures worktree-local lefthook hooks through `scripts/install-lefthook.mjs`. The [worktree-local hooks Agent Note](../.agents/notes/implemented/process/2026-07-27-worktree-local-lefthook.md) owns the safety and migration contract.
 
 If hooks are missing because dependencies were restored from cache or `postinstall` was skipped, install them manually:
 
@@ -27,11 +29,7 @@ If hooks are missing because dependencies were restored from cache or `postinsta
 node scripts/install-lefthook.mjs
 ```
 
-The wrapper refuses user-owned `core.hooksPath` values. An inherited system, global, or common-repository path requires `DSH_LEFTHOOK_ALLOW_HOOKS_PATH_OVERRIDE=1`. When Git seeds a new worktree with another registered worktree's marker-backed hook path, the wrapper replaces that copied value with the new worktree's own path; command-scoped and other worktree-scoped paths must be integrated or removed explicitly.
-
-Before enabling worktree config, migrate direct `extensions.*` in a format-0 common config, direct `core.worktree` or `core.bare=true`, and any non-empty dormant `config.worktree`. The common config and every worktree config must be regular files, while the owned hook directory may contain only unaliased regular files.
-
-After moving a checkout, rerun the wrapper to relocate its owned path and regenerate hooks. For a stale or invalid installer lock, first confirm no installer is running, then remove the reported lock and retry. If installation and hook-path rollback both fail, inspect the reported worktree config before retrying. The [worktree-local hooks Agent Note](../.agents/notes/implemented/process/2026-07-27-worktree-local-lefthook.md) owns the full safety contract.
+If the wrapper rejects existing Git configuration or reports a stale lock, follow its diagnostic and the linked Agent Note rather than editing worktree metadata speculatively. After moving a checkout, rerun the wrapper to regenerate the owned path.
 
 Run typecheck once after a fresh clone:
 
@@ -39,9 +37,13 @@ Run typecheck once after a fresh clone:
 pnpm run typecheck
 ```
 
-That first typecheck runs the whole-repo `tsc -b` graph: it emits every package/vendor `lib/types` and checks examples, tests, and scripts through the two no-emit aggregates described below.
+Setup is complete when `pnpm run typecheck` exits successfully.
 
-## TypeScript project layout
+## Contributor reference
+
+### TypeScript project layout
+
+The repository typecheck runs the whole-repo `tsc -b` graph: it emits every package/vendor `lib/types` and checks examples, tests, and scripts through two no-emit aggregates.
 
 The repository's TypeScript configuration has exactly three roles; every tsconfig file plays one of them.
 
@@ -68,7 +70,7 @@ pnpm run build
 
 `pnpm run hygiene` includes `publint`, which validates package entrypoints against the built `lib/*.js` files, and `verify-node-next-types`, which validates built declarations against a temporary NodeNext consumer. A fresh worktree has no bundled JS or declarations until `pnpm run build` runs; ordinary commits and pushes do not require that build unless their selected checks consume it.
 
-## Environment variables
+### Environment variables
 
 The real DeepSeek adapter and key-backed agent demos read credentials from the environment or from a gitignored `.env` at the repo root:
 
@@ -79,7 +81,7 @@ DEEPSEEK_BASE_URL=https://... # optional
 
 `DEEPSEEK_BASE_URL` is optional and defaults to the public API. Never commit real credentials. The real-API e2e suites self-skip when `DEEPSEEK_API_KEY` is not set.
 
-## Git hooks
+### Git hooks
 
 lefthook is configured in `lefthook.yml` as a fast local checkpoint:
 
@@ -92,44 +94,15 @@ The hooks intentionally do not run tests, snapshots, documentation checks, build
 
 Contributors can opt into the comprehensive local gate set with `pnpm run check:all`. The command is independent of both Git hooks and is not an agent instruction.
 
-## CI gates
+### CI gates
 
 The keyless [CI workflow](../.github/workflows/ci.yml) groups independent gates into broad lanes and runs a smaller compatibility signal across supported Node versions. Artifact consumers wait for one build within their lane. The separate real-API workflow runs `pnpm run test:e2e` with its configured worker bound. See [scripts/run-gates.ts](../scripts/run-gates.ts) and the workflow files for the current gate and job inventory.
 
-## Daily commands
+### Daily commands
 
-Use these from the repo root:
+The root [contributor instructions](../AGENTS.md#commands) summarize common commands, while [`package.json`](../package.json) and [scripts/run-gates.ts](../scripts/run-gates.ts) own the current script and gate inventories. Select the smallest checks that cover the changed surface. Documentation changes use `pnpm run doc-sync`; package-public behavior changes also update the owning README or JSDoc, and built-artifact checks require `pnpm run build` first.
 
-```sh
-pnpm run test           # unit tests
-pnpm run test:coverage  # unit tests with per-file coverage gates
-pnpm run test:e2e       # real-API tests; self-skips without DEEPSEEK_API_KEY
-pnpm run check:all      # comprehensive opt-in gate set; not wired to Git hooks
-pnpm run typecheck      # tsc -b over the root solution: emits package/vendor lib/types, checks both aggregates
-pnpm run lint           # oxlint .
-pnpm run lint:fix       # formatting-only ESLint, then oxlint . --fix
-pnpm run doc-typecheck  # compile checked TypeScript snippets in Markdown docs
-pnpm run gen-cordis-catalog     # regenerate docs/cordis-catalog/events.md + services.md from source
-pnpm run verify-cordis-catalog  # fail if either cordis catalog is stale
-pnpm run verify-export-jsdoc    # fail if a module-level package export lacks complete JSDoc
-pnpm run gen-doc-graphs     # regenerate generated relationship docs from source and curated graph definitions
-pnpm run verify-doc-graphs  # fail if generated relationship docs are stale
-pnpm run verify-md-wrap  # fail on hard-wrapped prose paragraphs in docs/README markdown
-pnpm run verify-mermaid  # fail if a ```mermaid diagram has invalid Mermaid syntax
-pnpm run verify-type-equiv  # fail if a ```ts type-equiv doc block drifts from its source type
-pnpm run verify-doc-budgets  # fail if a budgeted standing doc exceeds its word ceiling
-pnpm run gen-translation-brief   # print the minimal-update briefing for out-of-sync translation pairs (--apply splices code-only edits)
-pnpm run doc-sync       # all Markdown/doc gates, scheduled concurrently; the doc-sync leaf list in scripts/run-gates.ts is the full list
-pnpm run gen-module-graph     # regenerate docs/module-graph.md from package peerDeps
-pnpm run verify-module-graph  # fail if docs/module-graph.md is stale
-pnpm run build          # emit lib/types intermediates, then bundle lib/index.* runtime files
-pnpm run verify-node-next-types  # fail if built declarations are not NodeNext-consumable
-pnpm run hygiene        # knip, publint, workspace constraints, and NodeNext declaration check
-```
-
-When changing package public behavior, update the relevant README or JSDoc in the same change. `pnpm run doc-sync` catches checked TypeScript snippets, generated doc freshness, markdown wrap/link drift, type equivalence, translation pairing, Mermaid syntax, and doc budgets, but broader prose/API sync still needs review.
-
-## Demos
+### Demos
 
 The one-shot Headless coding agent needs `DEEPSEEK_API_KEY` in the environment or repo-root `.env`:
 
@@ -149,7 +122,7 @@ The ACP automation server exposes fresh agent sessions over JSON-RPC stdio and a
 pnpm run demo:acp
 ```
 
-## TODO markers
+### TODO markers
 
 Use one of three comment tags to flag known issues in the code, ordered by urgency:
 
@@ -159,7 +132,7 @@ Use one of three comment tags to flag known issues in the code, ordered by urgen
 
 Pick the tag that matches the urgency so anyone scanning the code can tell a release blocker from a someday-maybe.
 
-## Documenting types verbatim (`ts type-equiv`)
+### Documenting types verbatim (`ts type-equiv`)
 
 The [core data structures](core-data-structures/core.md) docs paste source-equivalent declarations together with their original JSDoc so a reader sees the exact shape and source contract. To keep a paste from drifting when source changes, fence it as ` ```ts type-equiv ` (instead of ` ```ts `) and register it in `scripts/type-equiv.manifest.json` with the source file and symbol it mirrors:
 
@@ -168,7 +141,3 @@ The [core data structures](core-data-structures/core.md) docs paste source-equiv
 ```
 
 `pnpm run verify-type-equiv` (part of `doc-sync`) then extracts that symbol's declaration and attached JSDoc from source via the TypeScript parser and asserts the block matches both. For a class whose implementation bodies do not belong in the catalog, use ` ```ts public-api ` and set `"projection": "public-api"`; the checked projection retains the public fields, constructor, accessors, methods, and original class/member JSDoc while omitting bodies and private or protected members. Comparison ignores whitespace and non-JSDoc comments but requires every original JSDoc comment, including member documentation, so readers see the source contract beside the exact shape. The gate enforces a 1:1 correspondence by document, symbol, and projection between primary blocks and manifest entries; a paired `.zh.md` block reuses its unsuffixed sibling's entry only when the whole tracked fence sequence is byte-identical and ordered identically. `doc-typecheck` applies the same derivative rule to compilable fences, while skipping both source-equivalence fence kinds from compilation and its opt-out ratio. When you change a documented declaration or its JSDoc, the gate fails until you update the paste; when you add or remove a primary block, update the manifest in the same change.
-
-## Architecture context
-
-Read `docs/architecture.md` before changing anything under `packages/`. The codebase is built around Cordis plugins, event-sourced sessions, typed service seams, and explicit extension points.

docs/development.zh.md

@@ -2,16 +2,18 @@
 
 [English](development.md) | 中文
 
-本指南覆盖参与 DeepSeek Harness 开发所需的本地环境搭建、日常工作流与 CI 流程；设计动机与技术权衡请查阅相应 Agent Note。
+搭建教程引导新贡献者从准备前置条件开始，直到检出通过检查。后面的贡献者参考介绍仓库布局、日常工作流和 CI 形态。设计依据与实现细节属于链接的 Agent Note 和脚本。
 
-## 前置条件
+## 搭建教程
+
+### 前置条件
 
 - Node.js 支持 22.19+ 与 24+。CI 覆盖 22.19、24 和 26；见 [Node 引擎下限 Agent Note](../.agents/notes/implemented/process/2026-07-06-node-engine-floor.md)。
 - 启用了 Corepack 的 pnpm。仓库在 `package.json` 中固定使用 `pnpm@11.7.0`；如果 `pnpm --version` 无法通过 Corepack 解析，请先运行 `corepack enable`。
 - Git 2.26 或更高版本；钩子设置会启用 Git 的 worktree 专属配置扩展。
 - 可选：一个 DeepSeek API key，用于 Web、headless 和 ACP（Agent Client Protocol）自动化 agent（智能体）演示以及真实 API 的 e2e 测试。
 
-## 首次搭建
+### 首次搭建
 
 在仓库根目录安装依赖：
 
@@ -19,7 +21,7 @@
 pnpm install
 ```
 
-安装过程同时会运行根目录的 `postinstall` 脚本，该脚本通过 `scripts/install-lefthook.mjs` 从仓库 dev 依赖安装 lefthook。当 `CI=true` 或 `GITHUB_ACTIONS=true` 时，该脚本会在探测 Git 前返回，因为自动化任务不会使用贡献者钩子。否则，包装脚本要求使用 Git 2.26 或更高版本，并会为当前 worktree 在其自身的 Git 目录下设置显式钩子目录；因此，关联 worktree 会使用各自的 lefthook 二进制文件和配置，而不会改写共用钩子。首次安装会启用 Git 的 worktree 专属配置扩展和仓库格式 1；见 [worktree 本地钩子 Agent Note](../.agents/notes/implemented/process/2026-07-27-worktree-local-lefthook.md)。
+安装过程还会通过 `scripts/install-lefthook.mjs` 配置 worktree 本地的 lefthook 钩子。其安全与迁移契约由 [worktree 本地钩子 Agent Note](../.agents/notes/implemented/process/2026-07-27-worktree-local-lefthook.md) 负责。
 
 如果依赖是从缓存恢复或 `postinstall` 被跳过而导致缺少钩子，请手动安装：
 
@@ -27,11 +29,7 @@ pnpm install
 node scripts/install-lefthook.mjs
 ```
 
-包装层会拒绝用户自有的 `core.hooksPath` 值。继承自系统、全局或共用仓库配置的路径必须设置 `DSH_LEFTHOOK_ALLOW_HOOKS_PATH_OVERRIDE=1`。当 Git 使用另一个已注册 worktree 中由所有权标记佐证的钩子路径初始化新 worktree 时，包装层会将这个复制值替换为新 worktree 自有的路径；命令作用域和其他 worktree 作用域的路径必须显式集成或移除。
-
-启用 worktree 配置之前，请迁移格式 0 共用配置中直接设置的 `extensions.*`，并迁移直接设置的 `core.worktree` 或 `core.bare=true`，以及任何非空且尚未生效的 `config.worktree`。共用配置和每个 worktree 配置都必须是常规文件，而自有钩子目录只能包含不带别名的常规文件。
-
-检出目录移动后，请重新运行包装层，使其重新定位自有路径并重新生成钩子。对于陈旧或无效的安装程序锁，请先确认没有安装程序正在运行，再移除报告的锁并重试。若安装和钩子路径回滚都失败，请在重试前检查报告的 worktree 配置。完整安全契约由 [worktree 本地钩子 Agent Note](../.agents/notes/implemented/process/2026-07-27-worktree-local-lefthook.md) 统一定义。
+如果包装脚本拒绝现有 Git 配置或报告陈旧锁，请遵循其诊断和所链接的 Agent Note，不要凭猜测编辑 worktree 元数据。移动检出目录后，请重新运行包装脚本以重新生成自有路径。
 
 新克隆后请先运行一次类型检查：
 
@@ -39,9 +37,13 @@ node scripts/install-lefthook.mjs
 pnpm run typecheck
 ```
 
-首次类型检查会执行全仓 `tsc -b tsconfig.json` 图：发射每个 package/vendor 的 `lib/types`，并通过下述两个 no-emit 聚合检查示例、测试和脚本。
+`pnpm run typecheck` 成功退出即表示搭建完成。
 
-## TypeScript 项目布局
+## 贡献者参考
+
+### TypeScript 项目布局
+
+仓库类型检查会执行全仓 `tsc -b` 图：它会发射每个 package/vendor 的 `lib/types`，并通过两个 no-emit 聚合检查示例、测试和脚本。
 
 仓库的 TypeScript 配置只有三种角色；每个 tsconfig 文件恰好扮演其中一种。
 
@@ -68,7 +70,7 @@ pnpm run build
 
 `pnpm run hygiene` 包含 `publint`（用构建出的 `lib/*.js` 文件校验 package 入口点）和 `verify-node-next-types`（用一个临时的 NodeNext 消费方校验构建出的声明文件）。新 worktree 在 `pnpm run build` 运行之前没有打包的 JS 和声明文件；普通提交和推送无需构建，除非所选检查会使用这些产物。
 
-## 环境变量
+### 环境变量
 
 真实的 DeepSeek 适配器和需要密钥的 agent 演示从环境变量或仓库根目录一个被 gitignore 的 `.env` 文件读取凭证：
 
@@ -79,7 +81,7 @@ DEEPSEEK_BASE_URL=https://... # optional
 
 `DEEPSEEK_BASE_URL` 可选，默认为公开 API。请勿提交真实凭证。未设置 `DEEPSEEK_API_KEY` 时，真实 API 的 e2e 套件会自动跳过。
 
-## Git 钩子
+### Git 钩子
 
 lefthook 在 `lefthook.yml` 中配置，作为快速的本地检查点：
 
@@ -92,44 +94,15 @@ vendor manifest 守卫检查 `vendor/*/src` 下的改动是否连同对应的 `v
 
 贡献者可以选择运行 `pnpm run check:all`，执行全面的本地门禁集。该命令独立于两个 Git 钩子，也不是对 agent 的指令。
 
-## CI 门禁
+### CI 门禁
 
 keyless [CI 工作流](../.github/workflows/ci.yml) 将独立门禁分组到若干宽粒度 lane，并在受支持的 Node 版本上运行一组较小的兼容性检查。产物消费方在各自 lane 内等待一次 build。单独的真实 API 工作流按其配置的 worker 上限运行 `pnpm run test:e2e`。当前门禁和 job 清单以 [scripts/run-gates.ts](../scripts/run-gates.ts) 和工作流文件为准。
 
-## 日常命令
+### 日常命令
 
-在仓库根目录使用：
+根目录的[贡献者说明](../AGENTS.md#commands)概述常用命令，[`package.json`](../package.json) 与 [scripts/run-gates.ts](../scripts/run-gates.ts) 则负责当前脚本和门禁清单。请选择覆盖变更表面的最小检查集。文档变更使用 `pnpm run doc-sync`；package 公开行为变更还需更新所属 README 或 JSDoc，而基于构建产物的检查需要先运行 `pnpm run build`。
 
-```sh
-pnpm run test           # unit tests
-pnpm run test:coverage  # unit tests with per-file coverage gates
-pnpm run test:e2e       # real-API tests; self-skips without DEEPSEEK_API_KEY
-pnpm run check:all      # comprehensive opt-in gate set; not wired to Git hooks
-pnpm run typecheck      # tsc -b over the root solution: emits package/vendor lib/types, checks both aggregates
-pnpm run lint           # oxlint .
-pnpm run lint:fix       # formatting-only ESLint, then oxlint . --fix
-pnpm run doc-typecheck  # compile checked TypeScript snippets in Markdown docs
-pnpm run gen-cordis-catalog     # regenerate docs/cordis-catalog/events.md + services.md from source
-pnpm run verify-cordis-catalog  # fail if either cordis catalog is stale
-pnpm run verify-export-jsdoc    # fail if a module-level package export lacks complete JSDoc
-pnpm run gen-doc-graphs     # regenerate generated relationship docs from source and curated graph definitions
-pnpm run verify-doc-graphs  # fail if generated relationship docs are stale
-pnpm run verify-md-wrap  # fail on hard-wrapped prose paragraphs in docs/README markdown
-pnpm run verify-mermaid  # fail if a ```mermaid diagram has invalid Mermaid syntax
-pnpm run verify-type-equiv  # fail if a ```ts type-equiv doc block drifts from its source type
-pnpm run verify-doc-budgets  # fail if a budgeted standing doc exceeds its word ceiling
-pnpm run gen-translation-brief   # print the minimal-update briefing for out-of-sync translation pairs (--apply splices code-only edits)
-pnpm run doc-sync       # all Markdown/doc gates, scheduled concurrently; the doc-sync leaf list in scripts/run-gates.ts is the full list
-pnpm run gen-module-graph     # regenerate docs/module-graph.md from package peerDeps
-pnpm run verify-module-graph  # fail if docs/module-graph.md is stale
-pnpm run build          # emit lib/types intermediates, then bundle lib/index.* runtime files
-pnpm run verify-node-next-types  # fail if built declarations are not NodeNext-consumable
-pnpm run hygiene        # knip, publint, workspace constraints, and NodeNext declaration check
-```
-
-修改 package 的公开行为时，请在同一个变更中更新相关 README 或 JSDoc。`pnpm run doc-sync` 能检测到被检查的 TypeScript 片段、生成文档的新鲜度、Markdown 换行/链接漂移、type-equiv、翻译配对、Mermaid 语法和文档预算，但更广泛的行文/API 同步仍需评审把关。
-
-## 演示
+### 演示
 
 单次运行的 Headless coding agent 需要环境变量或仓库根目录 `.env` 中的 `DEEPSEEK_API_KEY`：
 
@@ -149,7 +122,7 @@ ACP 自动化服务器通过 JSON-RPC stdio 提供全新 agent 会话，同样
 pnpm run demo:acp
 ```
 
-## TODO 标记
+### TODO 标记
 
 请使用以下三种注释标签之一标记代码中的已知问题，按紧急程度排序：
 
@@ -159,7 +132,7 @@ pnpm run demo:acp
 
 请选择与紧急程度匹配的标签，让浏览代码的人一眼分清「发布阻塞」和「有空再说」。
 
-## 逐字记录类型（`ts type-equiv`）
+### 逐字记录类型（`ts type-equiv`）
 
 [核心数据结构](core-data-structures/core.md)文档会把与源码等价的声明及其原始 JSDoc 一并粘贴，让读者看到确切形状和源码契约。为防止粘贴内容在源码变化时漂移，请将其围栏为 ` ```ts type-equiv `（而不是 ` ```ts `），并在 `scripts/type-equiv.manifest.json` 中登记它镜像的源文件和符号：
 
@@ -168,7 +141,3 @@ pnpm run demo:acp
 ```
 
 `pnpm run verify-type-equiv`（`doc-sync` 的一环）随后通过 TypeScript 解析器从源码提取该符号的声明及其附带的 JSDoc，并断言代码块同时匹配两者。对于不应把实现体写进目录的类，请使用 ` ```ts public-api ` 并设置 `"projection": "public-api"`；门禁检查的投影会保留公共字段、构造函数、访问器、方法以及类和成员的原始 JSDoc，同时省略实现体和私有或受保护成员。比对会忽略空白和非 JSDoc 注释，但要求保留每条原始 JSDoc（包括成员文档），让读者同时看到源码契约和确切形状。该门禁按文档、符号和投影，在主块与 manifest 条目之间强制 1:1 对应；只有当配对 `.zh.md` 块的完整受跟踪围栏序列与其无后缀兄弟文件按字节一致且顺序相同时，才会复用后者的条目。`doc-typecheck` 对可编译围栏应用同一派生规则，同时跳过两种源码等价围栏的编译，并将其排除在 opt-out 比例之外。当你改动一个已记录的类型声明或其 JSDoc 时，门禁会失败直到你更新粘贴内容；当你增删一个主块时，请在同一个变更里更新 manifest。
-
-## 架构上下文
-
-在修改 `packages/` 目录下的任何内容之前，请先阅读 `docs/architecture.md`。这套代码围绕 Cordis 插件、事件溯源的会话、类型化的服务 seam 与显式扩展点构建。

docs/event-producer-consumer.md

@@ -7,7 +7,7 @@ This matrix shows which packages dispatch each harness-owned event and which pac
 
 | Event | Mode | Declared in | Dispatchers | Listeners |
 | --- | --- | --- | --- | --- |
-| `agent-loop/config-start-failed` | `emit` | [`packages/core/agent-loop/src/index.ts:157`](../packages/core/agent-loop/src/index.ts) | [`agent-loop`](../packages/core/agent-loop) (`events.dispatch`) | - |
+| `agent-loop/config-start-failed` | `emit` | [`packages/core/agent-loop/src/index.ts:158`](../packages/core/agent-loop/src/index.ts) | [`agent-loop`](../packages/core/agent-loop) (`events.dispatch`) | - |
 | `agent/cancel-requested` | `emit` | [`packages/core/agent/src/types.ts:353`](../packages/core/agent/src/types.ts) | [`agent-loop`](../packages/core/agent-loop) (`emitAgentEvent`) | [`goal-session`](../packages/goal/goal-session) |
 | `agent/created` | `emit` | [`packages/core/agent/src/types.ts:284`](../packages/core/agent/src/types.ts) | [`agent`](../packages/core/agent) (`events.dispatch`) | [`goal-session`](../packages/goal/goal-session) |
 | `agent/disposed` | `emit` | [`packages/core/agent/src/types.ts:293`](../packages/core/agent/src/types.ts) | [`agent`](../packages/core/agent) (`events.dispatch`) | [`agent-loop`](../packages/core/agent-loop), [`goal-session`](../packages/goal/goal-session), [`subagent`](../packages/subagent/subagent) |

docs/i18n/README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/i18n/README.md
-README.md: d1fff61a1d3db7fbe93d3c95b144598c2ca41c8b
-README.zh.md: 9f1d61f1b854ead0ef5851157cce2e9a6de83d89
+README.md: 2bac578034441bbe786ce54c051cc622bf9275c0
+README.zh.md: b7566bf6f88e03eb06227b63011f78252b4dcd07

docs/i18n/README.md

@@ -43,7 +43,7 @@ The gate's limit, stated plainly: **a green gate means the pair was confirmed co
 
 **Excluded** (never paired, and the gate rejects a `.zh.md` or `.i18n.yaml` for them):
 
-- `docs/cordis-catalog/`, `docs/tool-catalog/`, `docs/config-catalog.md`, `docs/persistence-catalog.md`, `docs/module-graph.md`, `docs/agent-lifecycle.md`, `docs/capability-seams.md`, `docs/event-producer-consumer.md`, `docs/graph-atlas.md`, and `docs/tool-execution-pipeline.md` — generated files; their generators emit English only today, so a hand-written translation would go stale on every regeneration. The planned follow-up is to teach the generators to emit Chinese alongside English, at which point these leave the exclusion list.
+- `docs/cordis-catalog/`, `docs/tool-catalog/`, `docs/config-catalog.md`, `docs/persistence-catalog.md`, `docs/module-graph.md`, `docs/agent-lifecycle.md`, `docs/capability-seams.md`, `docs/event-producer-consumer.md`, `docs/graph-atlas.md`, and `docs/tool-execution-pipeline.md` — generated files whose generators emit English only; a hand-written translation would go stale on regeneration.
 - `docs/AGENTS.md`, `.agents/notes/**/AGENTS.md`, and their `CLAUDE.md` instruction symlinks — agent instructions, maintained in English only like the root `AGENTS.md`.
 - `docs/i18n/terminology.md` and [style-samples.md](style-samples.md) — both are bilingual by construction.
 - [translation-prompt.md](translation-prompt.md) — the automated pipeline's prompt template; its body is machine-consumed verbatim, so a paired translation would change pipeline behavior.

docs/i18n/README.zh.md

@@ -43,7 +43,7 @@
 
 **排除**（永不配对，门禁拒绝为它们建 `.zh.md` 或 `.i18n.yaml`）：
 
-- `docs/cordis-catalog/`、`docs/tool-catalog/`、`docs/config-catalog.md`、`docs/persistence-catalog.md`、`docs/module-graph.md`、`docs/agent-lifecycle.md`、`docs/capability-seams.md`、`docs/event-producer-consumer.md`、`docs/graph-atlas.md` 与 `docs/tool-execution-pipeline.md`：生成文件；生成器目前只输出英文，手写译文在每次重新生成时必然陈旧。计划中的后续工作是让生成器同时输出中文，届时这些文件移出排除清单。
+- `docs/cordis-catalog/`、`docs/tool-catalog/`、`docs/config-catalog.md`、`docs/persistence-catalog.md`、`docs/module-graph.md`、`docs/agent-lifecycle.md`、`docs/capability-seams.md`、`docs/event-producer-consumer.md`、`docs/graph-atlas.md` 与 `docs/tool-execution-pipeline.md`：生成文件，其生成器只输出英文；手写译文会在重新生成时变得陈旧。
 - `docs/AGENTS.md`、`.agents/notes/**/AGENTS.md` 以及指向它们的 `CLAUDE.md` 指令符号链接：agent 指令，与根 `AGENTS.md` 一样只以英文维护。
 - `docs/i18n/terminology.md` 与 [style-samples.md](style-samples.md)：二者本身即为中英对照文档。
 - [translation-prompt.md](translation-prompt.md)：自动翻译流水线的提示词模板；正文逐字进入模型请求，配对翻译会改变流水线行为。

docs/postmortem/0004-landlock-partial-notice-misclassified-child-failures.i18n.yaml

@@ -0,0 +1,6 @@
+# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
+# side as of the last confirmed-consistent state. Both languages carry equal authority;
+# after editing either side, bring the other along and re-record with:
+#   pnpm run verify-translation-pairing --write docs/postmortem/0004-landlock-partial-notice-misclassified-child-failures.md
+0004-landlock-partial-notice-misclassified-child-failures.md: db810fdc896f9734d1b581617838d72166f4efc9
+0004-landlock-partial-notice-misclassified-child-failures.zh.md: 4a31fb038c44b036e6a183040947295a47221967

docs/postmortem/0004-landlock-partial-notice-misclassified-child-failures.md

@@ -0,0 +1,55 @@
+# Post-mortem 0004: Landlock partial-enforcement notice misclassified child failures
+
+English | [中文](0004-landlock-partial-notice-misclassified-child-failures.zh.md)
+
+Status: resolved
+
+## Executive summary
+
+On kernels with an older Landlock ABI, the launcher prints a benign partial-enforcement notice before executing every child. The harness treated that shared `landlock-run:` prefix plus any nonzero child exit as launcher failure, so ordinary outcomes such as ripgrep's exit 1 for no matches surfaced as `SANDBOX_UNAVAILABLE`; the then-bash-backed filesystem search also hid that structured error behind `SEARCH_FAILED`. Broad signature rules and missing partial-ABI composition coverage let the defect through. Runner classification now requires status-gated fatal evidence after exact informational exclusions, and an assembled keyless scenario pins the surviving bash path. Filesystem search uses packaged ripgrep through the subprocess seam and does not cross sandboxed bash.
+
+## Summary
+
+The native launcher contract distinguishes two kinds of stderr lines. A partially enforcing kernel prints exactly `landlock-run: partial enforcement (older Landlock ABI)` and continues into the child. A launcher failure prints another `landlock-run:` line and exits 125 without executing the child.
+
+The harness represented both with one case-insensitive `landlock-run: ` substring. Its consumer classified any nonzero exit carrying that substring as runner failure. The child's status was therefore attached to the launcher's informational line: `false`, ripgrep's no-match exit 1, invalid-pattern exit 2, and even a child-selected exit 125 could be blamed on the sandbox despite successful confinement and execution.
+
+At the time of the incident, filesystem search added a second attribution error. Its bash-backed `runRipgrep()` caught every rejected bash run that was not aborted and replaced it with a generic cwd/shell-start `SEARCH_FAILED`, including the structured `SandboxUnavailableError` produced by the sandbox executor.
+
+## Impact
+
+On partial-ABI Landlock hosts, legitimate nonzero child outcomes could appear as sandbox infrastructure failure. `glob` and `grep` were especially visible because ripgrep uses exit 1 as successful empty search. When a real sandbox failure did occur through filesystem search, callers lost its `SANDBOX_UNAVAILABLE` code and received an incorrect startup diagnosis.
+
+The defect did not weaken confinement or run a command unconfined. Its security effect was availability and diagnostic integrity: a valid confined result was rejected or mislabeled.
+
+## Timeline
+
+- The native launcher contract defined exit 125 for launcher failures, a fatal `landlock-run:` line for every such failure, and the exact partial-enforcement notice for successful child execution.
+- The sandbox provider reduced that contract to `runnerFailureSignatures: ['landlock-run: ']`; the bash consumer combined the prefix with any nonzero exit and reported stderr's first line.
+- Unit tests covered clean success, denial diagnostics, and fatal runner prefixes. Real-runner tests self-skipped without a usable kernel and did not force partial enforcement followed by a nonzero child.
+- A minimal POSIX wrapper that prints the notice and `exec`s its payload reproduced the failure with `false` and ripgrep no-match.
+- Structured rules plus shared foreground/background classification and assembled replay coverage closed the surviving sandbox attribution gap. Filesystem search uses packaged ripgrep through `ctx.subprocess`; the fix leaves that path outside sandboxed bash.
+
+## Root cause
+
+The public sandbox result type could express only a bag of substrings. It could not state that Landlock failure requires exit 125, that evidence must occur within one fatal line, or that one exact line under the same prefix is informational. The boolean consumer consequently joined unrelated facts from different processes and selected the first stderr line for detail even when a later line was the fatal evidence.
+
+The test matrix mirrored that representation. Fake providers emitted either no runner line or an unambiguously fatal prefix; they never emitted a benign runner line before a child-controlled nonzero exit. Real Landlock coverage depended on the host ABI, so full-ABI hosts could not exercise the notice. In the incident-era search implementation, filesystem-search tests modeled raw spawn errors but not a structured error thrown by the real sandboxed bash composition.
+
+Stderr remains an in-band attribution channel. A confined child can deliberately reproduce a runner's gated fatal line and exit status, causing an availability/diagnostic false attribution. The tighter conjunction prevents the accidental collision in this incident but does not authenticate the writer; an out-of-band status protocol remains separate hardening, not a sandbox-bypass fix.
+
+## Guardrails added
+
+- [`RunnerFailureRule`](../core-data-structures/sandbox.md#wrapped-argv-and-classification-dialects) carries optional allowed exit codes, case-insensitive per-line fatal signatures, and case-insensitive exact informational-line exclusions.
+- [`dsh-sandbox-local`](../../packages/sandbox/sandbox-local/) maps Landlock to exit 125 plus a non-notice `landlock-run:` line while bwrap, Seatbelt, and custom runners remain signature-only.
+- [`dsh-bash-sandbox`](../../packages/bash/bash-sandbox/) directly spawns the provider argv, so a pre-start rejection uses the spawn-error channel instead of localized shell diagnostics. Settled foreground and background execution share one evidence-returning classifier; fatal evidence outranks denial, and foreground errors report the matched fatal line without changing captured stderr.
+- [`dsh-tool-fs-search`](../../packages/fs/tool-fs-search/) uses packaged ripgrep through `ctx.subprocess` and remains outside the sandboxed bash seam.
+- The native-boundary regression cases live in [`partial-landlock.spec.ts`](../../packages/bash/bash-sandbox/tests/partial-landlock.spec.ts), including informational notices, fatal evidence, and foreground/background classification.
+- The assembled product path is pinned by the [`partial-landlock` snapshot composition](../../examples/acp-agent/partial-landlock.cordis.snapshot.yml), independently of filesystem-search implementation choices.
+
+## Lessons
+
+- Process attribution requires a conjunction of independent evidence; a shared prefix is not a protocol.
+- Informational and fatal diagnostics can share a namespace, so exclusions must be exact and narrow while unknown fatal lines stay fail-closed.
+- An adapter must preserve structured failures owned by the seam below it instead of replacing them with its own nearest generic category.
+- Platform-dependent behavior needs a deterministic fake at the native boundary plus one assembled product path; a self-skipping real-kernel test cannot carry that regression alone.

docs/postmortem/0004-landlock-partial-notice-misclassified-child-failures.zh.md

@@ -0,0 +1,55 @@
+# 事故复盘（postmortem） 0004：Landlock 部分强制执行通知导致子进程失败被误归类
+
+[English](0004-landlock-partial-notice-misclassified-child-failures.md) | 中文
+
+Status: resolved
+
+## 摘要
+
+在 Landlock ABI 较旧的内核上，launcher 会在执行每个子进程前打印一条无害的部分强制执行通知。harness 把共享的 `landlock-run:` 前缀与任意非零子进程退出组合起来，判定为 launcher 失败，因此 ripgrep 在没有匹配项时以 1 退出等普通结果会呈现为 `SANDBOX_UNAVAILABLE`；当时仍由 bash 支撑的文件系统搜索还会用 `SEARCH_FAILED` 遮蔽这个结构化错误。过于宽泛的签名规则，以及缺少较旧 ABI 下部分强制执行的组合测试覆盖，让该缺陷得以流入。runner 分类现在会先精确排除信息性行，再要求由退出状态门控的致命证据，并由一个组装后的无密钥场景固定仍然存在的 bash 路径。文件系统搜索通过 subprocess seam 运行打包的 ripgrep，不经过沙箱化 bash。
+
+## 概述
+
+原生 launcher 契约区分两类 stderr 行。内核只能部分强制执行时，会精确打印 `landlock-run: partial enforcement (older Landlock ABI)`，然后继续执行子进程。launcher 失败则打印另一行 `landlock-run:` 诊断，在不执行子进程的情况下以 125 退出。
+
+harness 用一个不区分大小写的 `landlock-run: ` 子串表示这两种情况。消费方只要发现非零退出同时携带该子串，就会归类为 runner 失败。因此，子进程的退出状态被错误地关联到 launcher 的信息性行：`false`、ripgrep 无匹配时的退出码 1、无效 pattern 的退出码 2，乃至由子进程自行选择的退出码 125，都可能在约束与执行均成功的情况下被错误归因为沙箱故障。
+
+事故发生时，文件系统搜索又造成第二处归因错误。当时由 bash 支撑的 `runRipgrep()` 会捕获 bash 执行器除中止外抛出的所有错误，并将其替换为关于 cwd 或 shell 启动的通用 `SEARCH_FAILED`，其中也包括沙箱执行器产生的结构化 `SandboxUnavailableError`。
+
+## 影响
+
+在 Landlock ABI 只能部分强制执行的主机上，合法的非零子进程结果可能表现为沙箱基础设施故障。`glob` 和 `grep` 尤其容易暴露该问题，因为 ripgrep 把退出码 1 用作成功的空搜索。当文件系统搜索中确实发生沙箱故障时，调用方也会丢失其 `SANDBOX_UNAVAILABLE` 错误码，转而收到错误的启动诊断。
+
+该缺陷没有削弱约束，也没有让命令在无约束状态下运行。其安全影响在于可用性与诊断完整性：有效的受限结果会被拒绝或错误标记。
+
+## 时间线
+
+- 原生 launcher 契约规定：launcher 失败使用退出码 125，每次此类失败都会打印一行致命的 `landlock-run:` 诊断；成功执行子进程时则打印精确的部分强制执行通知。
+- 沙箱提供方把该契约简化为 `runnerFailureSignatures: ['landlock-run: ']`；bash 消费方将此前缀与任意非零退出组合，并报告 stderr 的第一行。
+- 单元测试覆盖了无诊断的成功、拒绝诊断和致命 runner 前缀。真实 runner 测试在没有可用内核时会自行跳过，也没有强制构造「部分强制执行通知后跟非零子进程退出」的情况。
+- 一个最小 POSIX 包装脚本会打印该通知并 `exec` 其负载；它通过 `false` 与 ripgrep 无匹配场景复现了故障。
+- 结构化规则、前台与后台共享的分类逻辑和组装后的回放覆盖共同弥补了仍然存在的沙箱归因缺口。文件系统搜索通过 `ctx.subprocess` 运行打包的 ripgrep；本修复让该路径继续位于沙箱化 bash 之外。
+
+## 根因
+
+公开的沙箱结果类型只能表达一组子字符串。它无法表示 Landlock 失败必须使用退出码 125、证据必须出现在一行致命诊断内，或同一前缀下有一行精确文本属于信息性通知。消费方的布尔判定逻辑因此把来自不同进程且互不相关的事实组合在一起；即便致命证据位于后续行，它仍选用 stderr 的第一行作为详细信息。
+
+测试矩阵与这种表示方式一致。模拟提供方要么不输出 runner 行，要么输出含义明确的致命前缀，从不在由子进程控制的非零退出前输出无害 runner 行。真实 Landlock 覆盖依赖主机 ABI，因此使用完整 ABI 的主机无法覆盖该通知。在事故发生时的搜索实现中，文件系统搜索测试模拟了原始 spawn 错误，却没有覆盖真实沙箱化 bash 组合抛出的结构化错误。
+
+stderr 仍是带内归因通道。受限子进程可以故意复现 runner 的门控致命诊断行与退出状态，造成可用性或诊断误归因。更严格的多项证据合取可以避免本次事故中的意外冲突，但无法验证写入者身份；带外状态协议仍属于独立的加固工作，而非沙箱绕过修复。
+
+## 已添加的防护措施
+
+- [`RunnerFailureRule`](../core-data-structures/sandbox.md#wrapped-argv-and-classification-dialects) 携带可选的允许退出码、不区分大小写的逐行致命签名，以及按不区分大小写的整行精确匹配排除的信息性行。
+- [`dsh-sandbox-local`](../../packages/sandbox/sandbox-local/) 把 Landlock 映射为退出码 125 加一行非通知的 `landlock-run:` 诊断，而 bwrap、Seatbelt 和自定义 runner 仍仅依据签名。
+- [`dsh-bash-sandbox`](../../packages/bash/bash-sandbox/) 直接 spawn 提供方 argv，因此启动前遭拒时使用 spawn 错误通道，而非本地化的 shell 诊断。已结算的前台与后台执行共用一个返回证据的分类器；致命证据优先于拒绝，前台错误会报告匹配到的致命行，同时保持捕获的 stderr 不变。
+- [`dsh-tool-fs-search`](../../packages/fs/tool-fs-search/) 通过 `ctx.subprocess` 运行打包的 ripgrep，并继续位于沙箱化 bash seam 之外。
+- 原生边界回归用例位于 [`partial-landlock.spec.ts`](../../packages/bash/bash-sandbox/tests/partial-landlock.spec.ts)，包括信息性通知、致命证据和前台／后台分类。
+- 组装后的产品路径由 [`partial-landlock` 快照组合](../../examples/acp-agent/partial-landlock.cordis.snapshot.yml)固定，独立于文件系统搜索的实现选择。
+
+## 教训
+
+- 进程归因需要多项独立证据同时成立；共享前缀不是协议。
+- 信息性诊断与致命诊断可以共享同一命名空间，因此排除规则必须精确且范围狭窄，同时对未知的致命行保持失败关闭。
+- 适配器必须保留下层 seam 所拥有的结构化失败，而不能用自身最接近的通用类别将其替换。
+- 平台相关行为需要在原生边界放置确定性的模拟实现，并覆盖一条组装后的产品路径；会自行跳过的真实内核测试无法独自固定该回归。

docs/postmortem/README.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/postmortem/README.md
-README.md: 4858f8841e92a895f2d1a840b59b42758e83d952
-README.zh.md: 45fe23275e6236d5210ba32622b51457d170aaeb
+README.md: ffde0057304856b7c7718e3dd1f4743c48ee193f
+README.zh.md: 58635af16212f990d0f4e2621f4bfed437631645

docs/postmortem/README.md

@@ -15,3 +15,4 @@ Every post-mortem opens with an **Executive summary**: one short paragraph a bus
 | [0001](0001-acp-default-export-drops-inject.md) | ACP server crashed on connect: `export default` dropped the plugin's `inject` |
 | [0002](0002-js-expression-disabled-filesystem-tools.md) | Filesystem snapshot tools were permanently disabled by a literal `!!js` object |
 | [0003](0003-web-agent-gui-feedback-loop.md) | Web agent validated a replacement server instead of the GUI hosting its session |
+| [0004](0004-landlock-partial-notice-misclassified-child-failures.md) | Landlock partial-enforcement notice misclassified child failures |

docs/postmortem/README.zh.md

@@ -15,3 +15,4 @@
 | [0001](0001-acp-default-export-drops-inject.md) | ACP（Agent Client Protocol）服务器在连接时崩溃：`export default` 丢失了插件的 `inject` |
 | [0002](0002-js-expression-disabled-filesystem-tools.md) | 文件系统快照工具被一个字面量 `!!js` 对象永久禁用 |
 | [0003](0003-web-agent-gui-feedback-loop.md) | Web agent（智能体）验证了替代服务器，而非承载其会话的 GUI |
+| [0004](0004-landlock-partial-notice-misclassified-child-failures.md) | Landlock 部分强制执行通知导致子进程失败被误归类 |

docs/testing.i18n.yaml

@@ -2,5 +2,5 @@
 # side as of the last confirmed-consistent state. Both languages carry equal authority;
 # after editing either side, bring the other along and re-record with:
 #   pnpm run verify-translation-pairing --write docs/testing.md
-testing.md: 728c65f1911cbaa098f653af9436a92336fbe068
-testing.zh.md: 72b7c3bcfa69f4c65fc128dbcbb19c080fb0fbc4
+testing.md: 514ca4e1df7505b02350470d5de4a5ee3647634b
+testing.zh.md: 6d24d8e74d726d53fa3500f57e34238483c5481c

docs/testing.md

@@ -9,7 +9,7 @@ How this repo tests, tier by tier, and the rules that keep a green suite meaning
 - **Unit** (`pnpm run test`): vitest over package and example specs under their `tests/**` directories plus repository script specs under `scripts/**/*.spec.ts`; tests stay with the code area they exercise. Every registry gets an HMR-safety test (dispose the contributing fiber, assert cleanup). Prefer edge cases, error paths, event ordering, concurrency races, and permanent contract regressions (see `packages/core/agent-loop/tests/contract-regressions.spec.ts`).
 - **Coverage gate** (`pnpm run test:coverage`): the gating run, per-file 100% on `packages/*/*/src`. An uncovered line is often dead code the gate is correctly flagging for deletion, not a missing test to bolt on. Line coverage is necessary, never sufficient — it proves lines ran, not that the feature works as shipped. Per-file 100% on `packages/bash/pwsh-local/src` needs a real `pwsh`: without one its executor suites self-skip and `vitest.config.ts` exempts the file so pwsh-less hosts stay green, while CI runners ship pwsh and enforce the full bar.
 - **Real-API e2e** (`pnpm run test:e2e`): with-key tests against live provider APIs — the DeepSeek model plus provider-specific smokes that gate on their own keys (`EXA_API_KEY`, `PERPLEXITY_API_KEY`, …); each suite self-skips without its key so keyless CI stays green ([real-API e2e Agent Note](../.agents/notes/implemented/testing/2026-06-19-real-api-e2e-ci.md)).
-- **Snapshot** (`pnpm run test:snapshot`): keyless expected outputs cover external behavior — transport contracts and presentation, while persisted logs pin assembled backend behavior. ACP boots the real automation-server example, replays a recorded session, and diffs normalized JSON-RPC plus the re-persisted log ([ACP snapshot Agent Note](../.agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.md)); headless pins `stream-json` through its real one-shot process. TUI journeys replay primary/child JSONL through the real loop and tools, then project ANSI into semantic terminal-state outputs; package snapshots retain transient states and a real PTY covers the process boundary ([TUI snapshot Agent Note](../.agents/notes/archived/testing/2026-07-18-tui-terminal-state-snapshots.md)). Use `pnpm run test:snapshot:record` when a model transcript changes and `pnpm run test:snapshot:refresh` when replay input remains valid; review every JSONL and expected-output diff. One ACP scenario (`text-turn`) pins full system-prompt/tool-schema content; other fixtures tokenize it so an edit churns one line ([pinned-header Agent Note](../.agents/notes/archived/testing/2026-07-06-pin-request-header-content-in-one-scenario.md)).
+- **Snapshot** (`pnpm run test:snapshot`): keyless expected outputs cover external behavior — transport contracts and presentation, while persisted logs pin assembled backend behavior. ACP boots the real automation-server example, replays a recorded session, and diffs normalized JSON-RPC plus the re-persisted log ([ACP snapshot Agent Note](../.agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.md)); headless pins `stream-json` through its real one-shot process. Use `pnpm run test:snapshot:record` when a model transcript changes and `pnpm run test:snapshot:refresh` when replay input remains valid; review every JSONL and expected-output diff. One ACP scenario (`text-turn`) pins full system-prompt/tool-schema content; other fixtures tokenize it so an edit churns one line ([pinned-header Agent Note](../.agents/notes/archived/testing/2026-07-06-pin-request-header-content-in-one-scenario.md)).
 - **Web browser snapshot** (`pnpm run test:web`; required Linux PR gate): Chromium compares replayed browser output with `apps/web/tests/snapshots/`. CI forces read-only `DSH_SNAPSHOT=replay`, never writing expected outputs; record/refresh stay local and every diff is reviewed ([web e2e lane](../.agents/notes/implemented/testing/2026-07-24-web-gui-browser-e2e-lane.md), [CI gate decision](../.agents/notes/implemented/testing/2026-07-30-web-browser-snapshot-ci-gate.md)). `test:web` [builds first](../.agents/notes/implemented/bug-fix/2026-07-28-themed-scrollbars-and-reserved-gutter.md) for plugin CSS.
 
 Committed session-format JSONL uses the canonical packed-row layout, and the keyless snapshot gate discovers every such fixture by its `session` header. In-flight branches carrying older fixture edits merge current `master` and run the [temporary migrator](../scripts/migrate-packed-session-fixtures.ts) through `pnpm run migrate:packed-session-fixtures`; the [removal proposal](../.agents/notes/proposed/process/2026-07-26-remove-packed-session-fixture-migrator.md) retires that command and these links after all affected branches converge.

docs/testing.zh.md

@@ -9,7 +9,7 @@
 - **单元测试**（`pnpm run test`）：vitest 运行包（package）和示例各自的 `tests/**` 目录下的测试，以及匹配 `scripts/**/*.spec.ts` 的仓库脚本测试；测试文件与其所覆盖的代码区域放在一起。每个注册表都有一个 HMR（热模块替换）安全测试（dispose（资源释放）贡献的 fiber，断言清理完成）。优先覆盖边界情况、错误路径、事件顺序、并发竞态，以及永久性契约回归（见 `packages/core/agent-loop/tests/contract-regressions.spec.ts`）。
 - **覆盖率门禁**（`pnpm run test:coverage`）：门禁级运行，对 `packages/*/*/src` 按文件 100% 覆盖。未覆盖的行往往是门禁正确标记出的死代码（应删除），而非需要补写的测试。行覆盖率是必要条件，但永远不是充分条件：它证明行被执行过，不证明功能按交付预期工作。`packages/bash/pwsh-local/src` 的按文件 100% 覆盖需要真实的 `pwsh`：缺少它时其 executor 套件会自动跳过，`vitest.config.ts` 会豁免该文件以使无 pwsh 的主机保持绿色，而 CI runner 自带 pwsh，仍按完整标准执行门禁。
 - **真实 API e2e**（`pnpm run test:e2e`）：带密钥测试调用真实提供方 API，包括 DeepSeek 模型以及各提供方特有的冒烟测试；这些测试各自由自己的密钥控制（`EXA_API_KEY`、`PERPLEXITY_API_KEY` 等），缺少密钥时套件会自动跳过，使 keyless CI 保持绿色（[真实 API e2e Agent Note](../.agents/notes/implemented/testing/2026-06-19-real-api-e2e-ci.md)）。
-- **快照**（`pnpm run test:snapshot`）：无密钥预期输出覆盖对外行为（传输契约与呈现），持久化日志则固定组装后的后端行为。ACP 启动真实的自动化服务器示例、回放录制会话，并对归一化 JSON-RPC 与重新持久化的日志执行 diff（[ACP 快照 Agent Note](../.agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.md)）；headless 通过真实单次运行进程固定 `stream-json`。TUI 旅程通过真实循环与工具回放主会话与子会话 JSONL，再将 ANSI 投影为语义化终端状态输出；包级快照保留瞬态状态，真实 PTY 覆盖进程边界（[TUI 快照 Agent Note](../.agents/notes/archived/testing/2026-07-18-tui-terminal-state-snapshots.md)）。当模型 transcript（文本记录）发生变化时使用 `pnpm run test:snapshot:record`，回放输入仍然有效时使用 `pnpm run test:snapshot:refresh`；请审查每一处 JSONL 与预期输出差异。一个 ACP 场景（`text-turn`）固定完整的系统提示词与工具 schema 内容；其他 fixture（测试前置数据）将其 token 化，因此修改只会扰动一行（[pinned-header Agent Note](../.agents/notes/archived/testing/2026-07-06-pin-request-header-content-in-one-scenario.md)）。
+- **快照**（`pnpm run test:snapshot`）：无密钥预期输出覆盖对外行为（传输契约与呈现），持久化日志则固定组装后的后端行为。ACP 启动真实的自动化服务器示例、回放录制会话，并对归一化 JSON-RPC 与重新持久化的日志执行 diff（[ACP 快照 Agent Note](../.agents/notes/implemented/testing/2026-06-19-acp-snapshot-tests.md)）；headless 通过真实单次运行进程固定 `stream-json`。当模型 transcript（文本记录）发生变化时使用 `pnpm run test:snapshot:record`，回放输入仍然有效时使用 `pnpm run test:snapshot:refresh`；请审查每一处 JSONL 与预期输出差异。一个 ACP 场景（`text-turn`）固定完整的系统提示词与工具 schema 内容；其他 fixture（测试前置数据）将其 token 化，因此修改只会扰动一行（[pinned-header Agent Note](../.agents/notes/archived/testing/2026-07-06-pin-request-header-content-in-one-scenario.md)）。
 - **Web 浏览器快照**（`pnpm run test:web`；必需的 Linux PR（Pull Request）门禁）：Chromium 将回放后的浏览器输出与 `apps/web/tests/snapshots/` 比较。CI 强制只读的 `DSH_SNAPSHOT=replay`，绝不写入预期输出；record/refresh 留在本地，每处 diff 都须评审（[web e2e 车道](../.agents/notes/implemented/testing/2026-07-24-web-gui-browser-e2e-lane.md)、[CI 门禁决策](../.agents/notes/implemented/testing/2026-07-30-web-browser-snapshot-ci-gate.md)）。`test:web` 会[先构建](../.agents/notes/implemented/bug-fix/2026-07-28-themed-scrollbars-and-reserved-gutter.md)以交付插件 CSS。
 
 签入仓库的会话格式 JSONL 使用规范打包行布局，无密钥快照门禁会通过 `session` header 发现每一份此类 fixture。仍携带旧版 fixture 改动的在途分支应合并当前 `master`，并通过 `pnpm run migrate:packed-session-fixtures` 运行[临时迁移器](../scripts/migrate-packed-session-fixtures.ts)；待所有受影响分支收敛后，[移除提案](../.agents/notes/proposed/process/2026-07-26-remove-packed-session-fixture-migrator.md)会移除该命令及这些链接。

docs/typert-catalog-integration-design.i18n.yaml

@@ -1,6 +0,0 @@
-# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
-# side as of the last confirmed-consistent state. Both languages carry equal authority;
-# after editing either side, bring the other along and re-record with:
-#   pnpm run verify-translation-pairing --write docs/typert-catalog-integration-design.md
-typert-catalog-integration-design.md: c7d601730655f61f3b875ad5ad6997d3c888bfea
-typert-catalog-integration-design.zh.md: abaddfe1f740d4bd7cff5b2db8fd91e34626aab8

docs/typert-catalog-integration-design.md

@@ -1,133 +0,0 @@
-# Typert Catalog Integration Design
-
-English | [中文](typert-catalog-integration-design.zh.md)
-
-## Current State and Problem
-
-Typert already provides separate host/client `FaceModel` instances, a `TypeGraph` with explicit cross-face references, and analysis support for services, events, `@typert object`, generics, inheritance, and External types. The TypeScript compiler API should only translate source code into this standard model; downstream consumers should not traverse the TypeScript AST again.
-
-The repository currently has two catalog pipelines that analyze TypeScript source directly: the static API catalog consumed by `tool-cordis`, and the generation and freshness gate for `docs/cordis-catalog/events.md` and `docs/cordis-catalog/services.md`. They analyze the same services, events, and related types, but maintain separate collection and rendering logic, so they cannot prove that the Typert model is sufficient to represent the existing domain semantics.
-
-The first phase makes both pipelines consume the Typert model while keeping the three committed artifacts character-for-character identical to their pre-migration versions:
-
-- `docs/cordis-catalog/events.md`
-- `docs/cordis-catalog/services.md`
-- `packages/cordis/tool-cordis/src/api-catalog.ts`
-
-This phase does not require product plugins to publish Typert subpaths, example applications to load Typert, or changes to the runtime dependencies of `tool-cordis`.
-
-## Options
-
-### Drive `tool-cordis` from the Runtime Registry
-
-Each plugin publishes and loads Typert artifacts, then `tool-cordis` reads the current runtime model from `ctx.typert`. This path reflects the set of plugins actually loaded, but it requires every product package represented in the catalog to add package exports, generated artifacts, registry contributions, and application assembly. That integration surface is much larger than the analysis capability being validated now.
-
-### Publish Typert Artifacts Repository-Wide, Then Aggregate Them Statically
-
-All product packages generate host/client JS and DTS during the normal build/typecheck process, then the catalog generator aggregates those artifacts. This path establishes the complete publication protocol up front, but it also changes many package manifests and the build topology at once, coupling catalog migration to repository-wide Typert publication.
-
-### Analyze at Build Time, Then Project the Catalog
-
-`WorkspaceAnalyzer` builds a `WorkspaceModel` and `TypeGraph` from the host TypeScript project. The repository-specific `CordisCatalogProjector` consumes only that model and generates the three texts. `tool-cordis` continues to import the committed static `api-catalog.ts`, so the runtime does not need the Typert service.
-
-This phase uses build-time projection. It directly verifies that the standard Typert model can replace the existing AST collector while leaving runtime publication and automatic loading to separate follow-up decisions.
-
-## Phase-One Architecture
-
-```text
-tsconfig.host.json
-        │
-        ▼
-WorkspaceAnalyzer ── TypeScript compiler API 的唯一边界
-        │
-        ▼
-WorkspaceModel + TypeGraph
-        │
-        ▼
-CordisCatalogProjector ── 不依赖 TypeScript AST
-        ├── docs/cordis-catalog/events.md
-        ├── docs/cordis-catalog/services.md
-        └── packages/cordis/tool-cordis/src/api-catalog.ts
-```
-
-The objects have the following responsibilities:
-
-- `WorkspaceAnalyzer` analyzes packages, exports, services, events, type declarations, and reference relationships, and produces a compiler-independent model.
-- `WorkspaceModel` and `TypeGraph` are the standard data structures shared by all generation and scanning analyses. They preserve developer-authored generics, inheritance, and type trees without retaining the TypeScript AST.
-- The root entry point of `@deepseek-ai/dsh-typert-generator` exports `CordisCatalogProjector`, which performs model-driven selection, sorting, summary extraction, source location handling, JSDoc completeness checks, type-link closure, and rendering in three text formats. Its implementation remains in a dedicated Cordis catalog file, but it does not create another package subpath or embed a list of repository type names.
-- `scripts/gen-cordis-catalog.ts` provides `LINK_MAP`, `FOUNDATION_TYPE_NAMES`, `TYPE_LINK_EXEMPTIONS`, and the inherited Cordis list, injects them explicitly into the projector through `CordisCatalogPolicy`, and owns the write/check CLI behavior. The vendor Cordis core pages continue to be generated by a separate pinned-source projector.
-- `tool-cordis` imports only the static `api-catalog.ts` and does not depend on `typert-registry` or `typert-loader`.
-
-`CordisCatalogProjector` is a repository-specific downstream consumer and is not part of Typert's general-purpose model. When adding another category, first extend the standard model, then add the corresponding projector. The Typert analyzer must not absorb Cordis documentation formats or `tool-cordis` presentation logic.
-
-## Model Additions
-
-In addition to type structure, the catalog's character-for-character projection needs the declaration forms written by developers and exact source locations. The standard model therefore retains event/service locations, body-free text for events and members, parameter initializers, and the export status and canonical text of type declarations. `SourceDeclarationModel` also indexes top-level exported declarations for ambiguity checks and static type closure, without promoting them to domain graph roots.
-
-```ts
-interface SourceLocation {
-  readonly file: string
-  readonly line: number
-  readonly column: number
-}
-
-interface EventModel {
-  readonly location: SourceLocation
-  readonly text: string
-}
-```
-
-Repository-wide analysis supports building bounded `ts.Program` instances in package batches, then merging them through source-location-stable graph ids into a face model equivalent to monolithic analysis. This capability changes only the memory boundary of the compiler program; it does not change package, declaration, or type graph semantics.
-
-All information required by the projector must come from `WorkspaceModel` or `TypeGraph`. If a fact required for character-for-character compatibility cannot be expressed by the model, extend the standard model; do not reintroduce `ts.Node`, `ts.Symbol`, or `ts.TypeChecker` in the projector or script.
-
-## Character-for-Character Migration Oracle
-
-Before migration, retain the three texts produced by the old generator against the same source state. After migration, run the new analyzer and projector and require the three outputs to be byte-for-byte identical. Newlines, spaces, ordering, JSDoc, source pointers, and generated headers are all part of the comparison.
-
-`pnpm run verify-cordis-catalog` retains its `--check` mode, which reads the three committed artifacts and compares them directly with the newly computed results. A missing file or any differing character makes the artifact stale, and the error points to the single `pnpm run gen-cordis-catalog` repair command.
-
-Tests pin both of the following layers:
-
-- Typert fixture snapshots pin the `WorkspaceModel`, `TypeGraph`, JS, DTS, and Zod outputs, proving the behavior of the standard model and general-purpose emitters.
-- Cordis catalog tests or snapshots pin the projector's three complete texts, proving that the repository-specific product projection does not bypass the standard model and providing directly reviewable textual evidence.
-
-The three committed artifacts are the migration oracle between the old and new implementations and the continuing freshness oracle after migration. The old `gen-cordis-api` AST collector is removed. The scripts and commands with that name remain only as compatibility entry points for the unified projector because the generated file header itself contains the command; retaining the entry point preserves the character-for-character oracle without creating a second source of truth.
-
-## Exact Change List
-
-### Typert Generator
-
-- Add the locations, authored declaration text, parameter initializers, export status, and top-level source declaration index needed for character-for-character projection, with coverage in analyzer and model snapshots.
-- Support bounded package-batch analysis and prove that direct and batched models are equivalent.
-- Confirm that the catalog's required service declarations, public instance members, JSDoc, generics, inheritance, and referenced types are all available from the model.
-- Keep the TypeScript compiler API encapsulated within the analyzer; the public model and projector inputs do not expose compiler objects.
-
-### Cordis Catalog Projector
-
-- Select the complete set of Cordis services and events from the host `WorkspaceModel`.
-- Preserve the old generator's JSDoc rules: events must have `@mode` and payload `@param` tags; service methods must have a matching `@param` for every parameter; non-void returns must have `@returns`.
-- Compute the type links used by signatures and the transitive public type closure required by `tool-cordis` from the type graph.
-- Receive caller-maintained type classifications and the inherited surface through an explicit `CordisCatalogPolicy`; do not maintain the repository documentation taxonomy inside the generator package.
-- Preserve the existing output rules for source pointers, signatures, summaries, ordering, declaration truncation, and the inherited context catalog.
-- Project once and render the events Markdown, services Markdown, and TypeScript API catalog, preventing drift between documentation and tool data.
-
-### Commands and Consumers
-
-- `scripts/gen-cordis-catalog.ts` maintains repository policy data, assembles the analyzer and projector, and writes/checks all three artifacts together. Parsing, validation, and rendering logic lives in the generator's dedicated Cordis source file and is exported uniformly from the package root entry point.
-- Narrow `scripts/gen-cordis-api.ts` to a logic-free compatibility entry point for the unified CLI; the root `gen-cordis-api` and `verify-cordis-api` aliases point to that entry point.
-- Restore the static catalog default in `tool-cordis` and remove its dependencies on `ctx.typert`, `typert-registry`, and runtime package-model completeness.
-- `gen-doc-graphs` obtains the projector's model-level result once and reuses its services and events; it must not continue to import the AST collector or analyze the repository again.
-
-### Narrow the Scope of Phase-One Changes
-
-- Remove the newly added `./typert` and `./client/typert` exports and `lib/typert.*` files from product plugin package.json files.
-- Remove `typert-registry` and `typert-loader` assembly from examples.
-- Normal build/typecheck does not run repository-wide `gen-typert` or require product-package Typert artifacts to exist before it runs on a clean tree.
-- Retain `packages/typert/generator`, `packages/typert/registry`, and `packages/typert/loader`, along with their independent fixture, emitter, and runtime registration tests.
-
-## Future Extensions
-
-The runtime registry remains the receiving and query layer for generated JS/Zod, and the loader remains the automatic loading mechanism; neither supplies data to the first-phase static catalog. When product packages need runtime reflection, they can opt in by publishing `package/typert` and `package/client/typert`, which the loader then registers with `ctx.typert`.
-
-Future integration does not change the phase-one layering: only the analyzer handles TypeScript, the standard model serves both static generation and scan analysis, and the emitter produces runtime artifacts from that same model. Whether to extend publication to more packages, enable the loader by default, or extend the runtime registry's query capabilities are separate review decisions and remain decoupled from the Cordis catalog migration.

docs/typert-catalog-integration-design.zh.md

@@ -1,133 +0,0 @@
-# Typert catalog 接入设计
-
-[English](typert-catalog-integration-design.md) | 中文
-
-## 现状与问题
-
-Typert 已经具备独立的 host/client `FaceModel`、可显式跨 face 引用的 `TypeGraph`，以及 service、event、`@typert object`、泛型、继承和 External 类型的分析能力。TypeScript compiler API 只应负责把源码转换成这套标准模型；后续消费者不应再次遍历 TypeScript AST。
-
-仓库目前有两条直接分析 TypeScript 源码的 catalog 链路：`tool-cordis` 使用的静态 API catalog，以及 `docs/cordis-catalog/events.md`、`docs/cordis-catalog/services.md` 的生成与 freshness gate。它们分析的是同一批 service、event 和相关类型，却分别维护收集与渲染逻辑，不能证明 Typert 模型足以承载现有业务语义。
-
-第一阶段的目标是让这两条链路共同消费 Typert 模型，并保持三份已提交产物与迁移前字符级一致：
-
-- `docs/cordis-catalog/events.md`
-- `docs/cordis-catalog/services.md`
-- `packages/cordis/tool-cordis/src/api-catalog.ts`
-
-本阶段不要求业务插件发布 Typert 子路径，不要求示例应用加载 Typert，也不改变 `tool-cordis` 的运行时依赖关系。
-
-## 可选路径
-
-### 运行时 registry 驱动 `tool-cordis`
-
-每个插件发布并加载 Typert 产物，`tool-cordis` 再从 `ctx.typert` 读取当前运行时模型。这条路径可以反映实际加载的插件集合，但会要求所有参与 catalog 的业务包增加 package exports、生成产物、registry contribution 和应用装配，接入面远大于当前要验证的分析能力。
-
-### 全仓发布 Typert 产物后静态汇总
-
-所有业务包在普通 build/typecheck 中生成 host/client JS 与 DTS，再由 catalog 生成器汇总这些产物。这条路径能够提前建立完整的发布协议，但会同时修改大量 package manifest 和构建拓扑，使 catalog 迁移与 Typert 的全仓发布绑定。
-
-### 构建期分析后投影 catalog
-
-`WorkspaceAnalyzer` 从 host TypeScript project 构建 `WorkspaceModel` 与 `TypeGraph`，仓库专用的 `CordisCatalogProjector` 只消费该模型并生成三份文本。`tool-cordis` 继续导入已提交的静态 `api-catalog.ts`，运行时不需要 Typert service。
-
-本阶段采用构建期投影。它直接验证 Typert 标准模型能否替代现有 AST collector，同时把运行时 publication 和自动加载留在独立的后续决策中。
-
-## 第一阶段架构
-
-```text
-tsconfig.host.json
-        │
-        ▼
-WorkspaceAnalyzer ── TypeScript compiler API 的唯一边界
-        │
-        ▼
-WorkspaceModel + TypeGraph
-        │
-        ▼
-CordisCatalogProjector ── 不依赖 TypeScript AST
-        ├── docs/cordis-catalog/events.md
-        ├── docs/cordis-catalog/services.md
-        └── packages/cordis/tool-cordis/src/api-catalog.ts
-```
-
-各对象的职责如下：
-
-- `WorkspaceAnalyzer` 负责 package、export、service、event、类型声明和引用关系的分析，并产生 compiler-independent model。
-- `WorkspaceModel` 与 `TypeGraph` 是所有生成和扫描分析共用的标准数据结构，保留开发者写出的泛型、继承和类型树，不保存 TypeScript AST。
-- `@deepseek-ai/dsh-typert-generator` 根入口导出的 `CordisCatalogProjector` 负责模型驱动的选择、排序、摘要、源位置、JSDoc 完整性、类型链接闭包和三种文本格式；实现仍单独放在 Cordis catalog 专用文件中，但不形成额外的 package subpath，也不内置仓库类型名单。
-- `scripts/gen-cordis-catalog.ts` 提供 `LINK_MAP`、`FOUNDATION_TYPE_NAMES`、`TYPE_LINK_EXEMPTIONS` 和 inherited Cordis 清单，通过 `CordisCatalogPolicy` 显式注入 projector，并负责 write/check 的命令行行为；vendor Cordis core 页面仍由独立的 pinned-source projector 生成。
-- `tool-cordis` 只导入静态 `api-catalog.ts`，不依赖 `typert-registry` 或 `typert-loader`。
-
-`CordisCatalogProjector` 是仓库业务消费者，不进入 Typert 通用模型。新增其他类别时，先扩展标准模型，再增加对应 projector；Typert analyzer 不吸收 Cordis 文档格式或 `tool-cordis` 展示逻辑。
-
-## 模型补充
-
-Catalog 的字符级投影除了类型结构，还需要开发者写下的声明形式和精确源码位置。标准模型因此保留 event/service location、event/member 的 body-free text、parameter initializer，以及 type declaration 的 export 状态和 canonical text；`SourceDeclarationModel` 另外索引顶层导出声明，供歧义检查和静态类型闭包使用，但不把它们提升为业务 graph root。
-
-```ts
-interface SourceLocation {
-  readonly file: string
-  readonly line: number
-  readonly column: number
-}
-
-interface EventModel {
-  readonly location: SourceLocation
-  readonly text: string
-}
-```
-
-全仓分析支持按 package 分批构建有界 `ts.Program`，再依靠源码位置稳定的 graph id 合并为与一次性分析等价的 face model。该能力只改变 compiler program 的内存边界，不改变 package、declaration 或 type graph 语义。
-
-projector 所需信息必须来自 `WorkspaceModel` 或 `TypeGraph`。如果字符级兼容需要的事实无法从模型表达，应补充标准模型；不得在 projector 或脚本中重新引入 `ts.Node`、`ts.Symbol` 或 `ts.TypeChecker`。
-
-## 字符级迁移 oracle
-
-迁移前，在同一份源码状态下保留旧生成器产生的三份文本。迁移后运行新的 analyzer 与 projector，要求三份输出逐字节相等；换行、空格、排序、JSDoc、source pointer 和生成头都属于比较内容。
-
-`pnpm run verify-cordis-catalog` 的 `--check` 模式继续读取三份 committed artifact，并与本次计算结果直接比较。任一文件缺失或任一字符不同都视为 stale，错误信息指向统一的 `pnpm run gen-cordis-catalog` 修复命令。
-
-测试同时固定以下两层：
-
-- Typert fixture snapshots 固定 `WorkspaceModel`、`TypeGraph`、JS、DTS 与 Zod 输出，证明标准模型和通用 emitter 的行为。
-- Cordis catalog 测试或 snapshot 固定 projector 的三份完整文本，证明仓库业务投影没有绕过标准模型，并给出可直接评审的文本证据。
-
-三份 committed artifact 是旧实现与新实现的迁移 oracle，也是迁移完成后的持续 freshness oracle。旧 `gen-cordis-api` AST collector 被删除；同名脚本和命令只作为统一 projector 的兼容入口保留，因为生成文件头本身包含该命令，保留入口可以维持字符级 oracle 而不产生第二套真源。
-
-## 精确改造清单
-
-### Typert generator
-
-- 补齐字符级投影所需的 location、authored declaration text、parameter initializer、export 状态和顶层 source declaration index，并在 analyzer 与 model snapshots 中覆盖。
-- 支持有界 package batch 分析，并证明 direct 与 batched model 等价。
-- 确认 catalog 所需的 service 声明、public instance member、JSDoc、泛型、继承和引用类型均可从 model 读取。
-- 保持 TypeScript compiler API 封装在 analyzer 内；公共 model 和 projector 输入不暴露 compiler 对象。
-
-### Cordis catalog projector
-
-- 从 host `WorkspaceModel` 选择完整的 Cordis service/event 集合。
-- 保留旧生成器的 JSDoc 规则：event 必须有 `@mode` 和 payload `@param`，service method 必须有参数对应的 `@param`，非 void 返回必须有 `@returns`。
-- 从 type graph 计算签名涉及的类型链接和 `tool-cordis` 所需的传递 public type closure。
-- 通过显式 `CordisCatalogPolicy` 接收调用方维护的类型分类和 inherited surface，不在 generator 包内维护仓库文档 taxonomy。
-- 保留 source pointer、签名、摘要、排序、声明截断和 inherited context catalog 的既有输出规则。
-- 一次投影并渲染 events Markdown、services Markdown 与 TypeScript API catalog，避免文档和工具数据漂移。
-
-### 命令与消费方
-
-- `scripts/gen-cordis-catalog.ts` 维护仓库 policy 数据、组装 analyzer/projector，并同时 write/check 三份产物；解析、校验和渲染逻辑位于 generator 的 Cordis 专用源文件，并统一从 package 根入口导出。
-- 将 `scripts/gen-cordis-api.ts` 收窄为统一 CLI 的无逻辑兼容入口；根目录的 `gen-cordis-api`、`verify-cordis-api` aliases 指向该入口。
-- `tool-cordis` 恢复静态 catalog 默认值，移除对 `ctx.typert`、`typert-registry` 和运行时 package model 完整性的依赖。
-- `gen-doc-graphs` 一次取得 projector 的 model-level 结果并复用 services/events，不能继续导入 AST collector 或重复分析全仓。
-
-### 收窄本阶段改动面
-
-- 撤销业务插件 package.json 中新增的 `./typert`、`./client/typert` exports 和 `lib/typert.*` files。
-- 撤销 examples 中的 `typert-registry`、`typert-loader` 装配。
-- 普通 build/typecheck 不运行全仓 `gen-typert`，也不要求 clean tree 预先存在业务包 Typert artifact。
-- 保留 `packages/typert/generator`、`packages/typert/registry`、`packages/typert/loader` 及其独立 fixture、emitter 和 runtime registration 测试。
-
-## 后续扩展
-
-Runtime registry 继续作为生成 JS/Zod 后的接收与查询层，loader 继续作为自动装载机制；两者不承担第一阶段静态 catalog 的数据来源。业务包需要运行时反射时，可以按 package opt-in 发布 `package/typert` 与 `package/client/typert`，再由 loader 注册到 `ctx.typert`。
-
-后续接入不改变本阶段的分层：TypeScript 只进入 analyzer，标准模型同时服务静态生成与扫描分析，runtime artifact 由 emitter 从同一模型产生。是否把更多 package 接入 publication、是否默认启用 loader，以及 runtime registry 最终提供哪些查询能力，分别评审，不与 Cordis catalog 迁移捆绑。