fix(system-prompt): place environment facts after reusable instructions

.agents/notes/implemented/architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.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/2026-07-05-prompt-variables-and-tool-guidance-ownership.md
-2026-07-05-prompt-variables-and-tool-guidance-ownership.md: f4364453c5ddded2fc0cb1d733732059feee6632
-2026-07-05-prompt-variables-and-tool-guidance-ownership.zh.md: d8ee8ef906d02d5a96ae7db3fb705685618223ee
+2026-07-05-prompt-variables-and-tool-guidance-ownership.md: 9623018458a3bca44c0811ee5c245c716550a459
+2026-07-05-prompt-variables-and-tool-guidance-ownership.zh.md: eb90207c8560e24f92383195ec7f70bc96a2dcaf

.agents/notes/implemented/architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.md

@@ -30,9 +30,9 @@ Plugins register `{{name}}` values through `ctx.systemPrompt.variable(name, prov
 
 `dsh-agent-loop` registers the two built-ins, both pure projections of the context agent: `model` (= `options.model`) and `cwd` (= `session.header.cwd`). The example personas write `powered by the {{model}} model` — the model name is stated once, in the `model:` config key. `{{cwd}}` is demonstrated in the ACP example only: every ACP session carries the client's cwd, while config-pre-created stdio agents have none (a persona claiming `{{cwd}}` there fails the turn — by design). The variables stay on the loop plugin (unlike the sections below): they are runtime facts of the agents THIS loop drives, and a replacement loop supplies its own.
 
-### Persona as the order-0 section
+### Persona as a registry section
 
-`dsh-system-prompt` owns `harness:identity` at first-party order `-1000` and the configured `deployment:persona` at order 0, so both survive a replacement loop. Prompt rendering has one path, `renderPrompt(assembly)`, and the routed request header therefore records the exact prompt later replayed by `ctx.tokenMeter` for compaction pressure. An agent-scoped `deployment:persona` shadows the global default and lets subagent providers install a persona before publication. The [`dsh-system-prompt` README](../../../../packages/core/system-prompt/README.md) owns the sparse named placements for identity, policy, tool guidance, generated protocol, and final-output obligations.
+`dsh-system-prompt` owns `harness:identity` at first-party order `-1000` and the configured `deployment:persona` at order `10200`, so both survive a replacement loop. The [environment-suffix decision](../bug-fix/2026-09-06-environment-prompt-suffix.md) supersedes only the identity-first placement of the deployment persona; variable and tool-guidance ownership remain here. Prompt rendering has one path, `renderPrompt(assembly)`, and the routed request header therefore records the exact prompt later replayed by `ctx.tokenMeter` for compaction pressure. An agent-scoped `deployment:persona` shadows the global default and lets subagent providers install a persona before publication. The [`dsh-system-prompt` README](../../../../packages/core/system-prompt/README.md) owns the sparse named placements for identity, policy, tool guidance, generated protocol, and final-output obligations.
 
 ### Tool guidance ownership
 
@@ -58,7 +58,7 @@ Per-tool semantics and selection guidance live in tool descriptions. Prompt sect
 
 ## Shipped invariants
 
-- The tui-agent prompt renders identity, persona with the interpolated model, then fs/shell/web guidance through one assembly path.
+- First-party prompts render identity, reusable instructions, then environment-bearing sections including the interpolated persona through one assembly path.
 - Fork and fresh subagent descriptions reflect whether the provider inherits completed conversation turns; the tool appears, disappears, and is reworded with provider lifecycle changes.
 - Unknown, valueless, malformed, or unbalanced variable references name the section and throw; duplicate section, variable, and tool registrations also throw.
 - Snapshot replay is prompt-independent: it keys recorded chunk streams by turn and step without comparing the outgoing request.

.agents/notes/implemented/architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.zh.md

@@ -30,9 +30,9 @@ Status: implemented
 
 `dsh-agent-loop` 注册两个内置变量，均为上下文 agent 的纯投影：`model`（= `options.model`）和 `cwd`（= `session.header.cwd`）。示例 persona 写 `powered by the {{model}} model`——模型名称只在 `model:` 配置键中声明一次。`{{cwd}}` 仅在 ACP 示例中演示：每个 ACP 会话携带客户端的 cwd，而配置预创建的 stdio agent 没有 cwd（在那里声称 `{{cwd}}` 的 persona 会导致该轮次失败——这是有意为之）。变量留在 loop 插件上（不同于下面的 section）：它们是本循环驱动的 agent 的运行时事实，替换循环自行提供自己的变量。
 
-### Persona 作为 order-0 section
+### Persona 作为注册表 section
 
-`dsh-system-prompt` 拥有 first-party order 为 `-1000` 的 `harness:identity` 和 order 为 0 的配置 `deployment:persona`，因此两者在循环被替换时仍然存活。提示词渲染只有一条路径 `renderPrompt(assembly)`，已路由请求 header 因此会记录准确的提示词，稍后由 `ctx.tokenMeter` 为压缩（compaction）压力回放。agent 作用域的 `deployment:persona` 遮蔽全局默认值，允许 subagent 提供方在发布前安装 persona。[`dsh-system-prompt` README](../../../../packages/core/system-prompt/README.zh.md)规定身份、策略、工具指导、生成协议和最终输出义务的稀疏具名位置。
+`dsh-system-prompt` 拥有 first-party order 为 `-1000` 的 `harness:identity` 和 order 为 `10200` 的配置 `deployment:persona`，因此两者在循环被替换时仍然存活。[环境后缀决策](../bug-fix/2026-09-06-environment-prompt-suffix.zh.md)仅取代部署 persona 的 identity-first 位置；变量与工具指导的归属仍由本文规定。提示词渲染只有一条路径 `renderPrompt(assembly)`，已路由请求 header 因此会记录准确的提示词，稍后由 `ctx.tokenMeter` 为压缩（compaction）压力回放。agent 作用域的 `deployment:persona` 遮蔽全局默认值，允许 subagent 提供方在发布前安装 persona。[`dsh-system-prompt` README](../../../../packages/core/system-prompt/README.zh.md)规定身份、策略、工具指导、生成协议和最终输出义务的稀疏具名位置。
 
 ### 工具指导归属
 
@@ -58,7 +58,7 @@ Status: implemented
 
 ## 交付的不变式
 
-- tui-agent 的提示词通过一条组装路径依次渲染 identity、带插值模型名的 persona，然后是 fs/shell/web 指导。
+- 第一方提示词通过一条组装路径依次渲染 identity、可复用指令，再渲染包含插值 persona 的环境信息段落。
 - fork 和 fresh subagent 的描述反映提供方是否继承已完成的对话轮次；工具随提供方生命周期变化而出现、消失和重新措辞。
 - 未知、无值、格式错误或不平衡的变量引用会指明 section 名称并抛出异常；重复的 section、变量和工具注册同样抛出异常。
 - 快照回放与提示词无关：它按轮次和步骤索引已记录的分片流，不比较发出的请求。

.agents/notes/implemented/bug-fix/2026-09-06-environment-prompt-suffix.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 .agents/notes/implemented/bug-fix/2026-09-06-environment-prompt-suffix.md
+2026-09-06-environment-prompt-suffix.md: 92818a9b75bd228adb77c177c504de8c892d1fc0
+2026-09-06-environment-prompt-suffix.zh.md: edd82ef626f8fcc2e3e019f00e03b24db1f21078

.agents/notes/implemented/bug-fix/2026-09-06-environment-prompt-suffix.md

@@ -0,0 +1,33 @@
+# Agent Note: Environment facts follow reusable prompt instructions
+
+Status: implemented
+
+English | [中文](2026-09-06-environment-prompt-suffix.zh.md)
+
+## Problem
+
+The local Web URL, Harness checkout path, and persona model/workspace values differ across users and machines. Placing those facts before reusable tool instructions makes otherwise identical prompts diverge near their beginning, limiting the prefix available for cache reuse.
+
+## Decision
+
+The [system-prompt registry](../../../../packages/core/system-prompt/README.md) keeps the fixed Harness identity first and places first-party reusable instructions through `STRUCTURED_OUTPUT` before the environment-bearing suffix: `HARNESS_SOURCE` at `10000`, `WEB_SURFACE` at `10100`, and `DEPLOYMENT_PERSONA` at `10200`. Existing section names, interpolation, scoped shadowing, and exact `complete: true` persona overrides are unchanged. The order change applies to entire sections; it does not parse persona prose or add an OS variable or value.
+
+This decision supersedes only persona placement in the [prompt-variables and tool-guidance ownership note](../architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.md). That note remains active because its single-owner rule, strict interpolation, and tool-guidance responsibilities still apply.
+
+## Alternatives considered
+
+**Move only the source path and Web URL.** Shipped personas also contain the model and cwd, so leaving the persona near the beginning still breaks the reusable prefix across workspaces.
+
+**Split environment facts into a new API or infer variable sections from their text.** Existing named section orders cover the current producers. A new classification or persona parser adds behavior and configuration without a current consumer that needs it.
+
+**Move these facts into runtime-context messages.** That changes their message role and persistence placement rather than only their order. The existing system sections can preserve their content and ownership while moving after reusable instructions.
+
+## Consequences
+
+Cross-user byte-identical prefixes require matching tools, configuration, and preceding section text. Tool schemas, plan mode, deployment-specific guidance, and experimental Team state can still differ. Arbitrary extension orders and assembly listeners remain authoritative; this is a first-party placement policy, not a universal stable-prefix guarantee. Provider cache sharing and hit-rate improvements are not measured or promised.
+
+The deployment persona and Web/source guidance occur later, including after structured-output instructions. Structured output need not be the final string; complete persona overrides still suppress every other system section. Source and Web facts retain their existing distinction between the Harness checkout, session workspace, and current working directory.
+
+## Testing
+
+[Registry tests](../../../../packages/core/system-prompt/tests/system-prompt.spec.ts) compare identical reusable prefixes across changed checkout paths, URLs, models, cwd values, and a test-registered platform variable; they also cover strict interpolation and complete overrides. [Loop tests](../../../../packages/core/agent-loop/tests/loop.spec.ts) pin request ordering and session-cwd interpolation. [Persona tests](../../../../packages/preset/persona/tests/persona.spec.ts) cover scoped replacement and complete personas. [Recorded prompt snapshots](../../../../docs/testing.md) cover the emitted prompts in native-tool and generated-SDK compositions; they do not measure provider cache hits.

.agents/notes/implemented/bug-fix/2026-09-06-environment-prompt-suffix.zh.md

@@ -0,0 +1,33 @@
+# Agent Note: 环境事实位于可复用提示词指令之后
+
+Status: implemented
+
+[English](2026-09-06-environment-prompt-suffix.md) | 中文
+
+## 问题
+
+本地 Web URL、Harness checkout 路径和 persona 中的模型／工作区值因用户与机器而异。将这些事实放在可复用工具指令之前，会使其余内容相同的提示词在开头附近就出现差异，限制可供缓存复用的前缀。
+
+## 决策
+
+[系统提示词注册表](../../../../packages/core/system-prompt/README.zh.md)将固定 Harness 身份保留在最前，并把截至 `STRUCTURED_OUTPUT` 的第一方可复用指令放在环境信息后缀之前：`HARNESS_SOURCE` 位于 `10000`，`WEB_SURFACE` 位于 `10100`，`DEPLOYMENT_PERSONA` 位于 `10200`。既有段落名称、插值、作用域遮蔽以及精确的 `complete: true` persona 覆盖保持不变。顺序调整作用于完整段落；它不解析 persona 行文，也不添加 OS 变量或值。
+
+本决策仅取代[提示词变量与工具指导归属记录](../architecture/2026-07-05-prompt-variables-and-tool-guidance-ownership.zh.md)中的 persona 位置。该记录保持有效，因为它的单一归属规则、严格插值和工具指导职责仍然适用。
+
+## 曾考虑的替代方案
+
+**仅移动源码路径与 Web URL。** 交付的 persona 还包含模型和 cwd；若 persona 仍靠近开头，不同工作区之间的可复用前缀仍会被打断。
+
+**通过新 API 拆分环境事实，或从文本推断变量段落。** 既有具名段落顺序已覆盖当前提供方。新的分类或 persona 解析器会增加行为与配置，却没有当前消费方需要它。
+
+**将这些事实移到 runtime-context 消息。** 这会改变其消息角色和持久化位置，而不只是顺序。既有系统段落可以在保留内容与归属的同时移到可复用指令之后。
+
+## 后果
+
+跨用户字节相同的前缀要求工具、配置和前置段落文本一致。工具 schema、plan mode、部署专用指导和实验性 Team 状态仍可能不同。任意扩展顺序与组装监听器仍决定最终结果；这是一项第一方位置策略，而非通用稳定前缀保证。不测量或承诺提供方共享缓存及命中率提升。
+
+部署 persona 和 Web／源码指导出现得更晚，包括位于结构化输出指令之后。结构化输出无需成为最后一个字符串；完整 persona 覆盖仍会抑制其他所有系统段落。源码与 Web 事实保留 Harness checkout、会话工作区和当前工作目录之间的既有区分。
+
+## 测试
+
+[注册表测试](../../../../packages/core/system-prompt/tests/system-prompt.spec.ts)在 checkout 路径、URL、模型、cwd 值和测试注册的平台变量变化时比较相同的可复用前缀；同时覆盖严格插值与完整覆盖。[循环测试](../../../../packages/core/agent-loop/tests/loop.spec.ts)固定请求顺序和会话 cwd 插值。[Persona 测试](../../../../packages/preset/persona/tests/persona.spec.ts)覆盖作用域替换与完整 persona。[录制的提示词快照](../../../../docs/testing.zh.md)覆盖原生工具与生成 SDK 组合发出的提示词；它们不测量提供方缓存命中。

apps/web/tests/replay-round-trip.e2e.ts

@@ -101,17 +101,17 @@ describe('web e2e: fresh round trip through the real assembly', () => {
     }
   }, 200_000)
 
-  it('records the Web surface, source checkout, and session cwd in the request header', async () => {
+  it('ends the request header with the source checkout, Web surface, and session cwd', async () => {
     if (settledSessionId === undefined) throw new Error('the drive turn did not publish a session id')
     const agent = scaffold.ctx.agents.get(settledSessionId)
     if (agent === undefined) throw new Error(`the settled Web agent ${settledSessionId} is no longer live`)
     const system = agent.session.requestHeader()?.system
     if (system === undefined) throw new Error('the settled Web request has no system prompt')
-    const prefix = system.split('\n\n').slice(0, 4).join('\n\n')
+    const suffix = system.split('\n\n').slice(-3).join('\n\n')
       .split(REPO_ROOT).join('{{sourceRoot}}')
       .split(join(scaffold.workspaceCwd, 'workspace')).join('{{cwd}}')
       .split(scaffold.baseUrl).join('{{webUrl}}')
-    await compareOrRefreshGolden(WEB_CONTEXT_EXPECTED, prefix, MODE)
+    await compareOrRefreshGolden(WEB_CONTEXT_EXPECTED, suffix, MODE)
   })
 
   it('exposes the assembled Web URL to the real bash tool', async () => {

docs/config-catalog.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/config-catalog.md
-config-catalog.md: cb7c9f38d216cc09ee358b41a6e378970282291f
-config-catalog.zh.md: c3208f32f8b3e686e386cae7705b648fe3386616
+config-catalog.md: 12f630ecd571c460b084a93a80b1bec57d5e029c
+config-catalog.zh.md: 843a3b2a6590287781f98075414aa4fc5e34e688

docs/config-catalog.md

@@ -2517,7 +2517,7 @@ export interface Config {
   /** Include dynamic runtime-context snapshots in model history (default true). */
   includeRuntimeContext?: boolean
   /**
-   * Deployment-wide order-0 persona template. A scoped section named
+   * Deployment-wide persona template after first-party guidance. A scoped section named
    * `deployment:persona` shadows it; `{{variable}}` references are strict.
    */
   persona?: string
@@ -2530,7 +2530,7 @@ export interface Config {
 }
 ```
 
-Source: [`packages/core/system-prompt/src/index.ts:237`](../packages/core/system-prompt/src/index.ts)
+Source: [`packages/core/system-prompt/src/index.ts:238`](../packages/core/system-prompt/src/index.ts)
 
 <a id="deepseek-aidsh-terminal-bash"></a>
 

docs/config-catalog.zh.md

@@ -2519,7 +2519,7 @@ export interface Config {
   /** Include dynamic runtime-context snapshots in model history (default true). */
   includeRuntimeContext?: boolean
   /**
-   * Deployment-wide order-0 persona template. A scoped section named
+   * Deployment-wide persona template after first-party guidance. A scoped section named
    * `deployment:persona` shadows it; `{{variable}}` references are strict.
    */
   persona?: string
@@ -2532,7 +2532,7 @@ export interface Config {
 }
 ```
 
-来源：[`packages/core/system-prompt/src/index.ts:237`](../packages/core/system-prompt/src/index.ts)
+来源：[`packages/core/system-prompt/src/index.ts:238`](../packages/core/system-prompt/src/index.ts)
 
 <a id="deepseek-aidsh-terminal-bash"></a>
 

packages/boot/app-boot/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 packages/boot/app-boot/README.md
-README.md: 3cc8247eb366f278acc8bb86b30da75fe3ab3ee4
-README.zh.md: 1ea2ce4a90c5aa57edf9761ae2b85b04405e103c
+README.md: e6afceaebf7fc94f5a6d708639de128cb9005d72
+README.zh.md: a98be1fd5d0624f971af5d0ce5a81fdc0cafab54

packages/boot/app-boot/README.md

@@ -128,7 +128,7 @@ Indirectly, through the loaded plugin tree, which alone contributes model contex
 
 #### KV Cache effect
 
-Boot itself invalidates nothing in the request prefix. A consumer that calls `addHarnessSourceSection` places one short line near the system prompt's head, before per-request content, so it does not invalidate the cache across turns; any other request-prefix change is owned by the named consumer.
+Boot itself changes no request prefix. `addHarnessSourceSection` places its source path after first-party reusable instructions, so different checkouts leave those preceding bytes unchanged when tools and configuration match. Provider cache reuse is not guaranteed.
 
 ## Known Limitations and Deferred Work
 

packages/boot/app-boot/README.zh.md

@@ -128,7 +128,7 @@ profile 是同一套 dsh 安装提供不同应用界面的方式：`web`、`head
 
 #### KV Cache 影响
 
-启动本身不会使请求前缀中的任何内容失效。消费方调用 `addHarnessSourceSection` 时，会在系统提示词靠前位置、逐请求内容之前添加一行短文本，因此不会使跨轮次缓存失效；请求前缀的其他任何变化均由相应的具名消费方负责。
+启动本身不改变请求前缀。`addHarnessSourceSection` 将源码路径放在第一方可复用指令之后，因此工具与配置一致时，不同 checkout 不会改变前置字节。不保证提供方复用缓存。
 
 ## 已知限制与延期工作
 

packages/boot/app-boot/src/index.ts

@@ -840,9 +840,10 @@ export const HARNESS_SOURCE_SECTION = 'harness:source'
  * explicitly distinguishing it from the task workspace and current working
  * directory. The self-referential `dsh-tool-cordis` toolset reads and edits this
  * checkout. Call once on the settled boot context ({@link boot}); the section
- * uses the shared first-party placement just after the harness identity opener
- * and before the deployment persona. A booted tree with no `systemPrompt` service has no prompt to
- * augment, so this is then a no-op that returns `undefined`. The section is
+ * uses the shared first-party placement after reusable instructions
+ * and before the Web surface and deployment persona. A booted tree with no
+ * `systemPrompt` service has no prompt to augment, so this is then a no-op
+ * that returns `undefined`. The section is
  * registered against the `systemPrompt` service's fiber, so a dev HMR reload of
  * that plugin drops it until the next boot.
  * @param ctx - the settled boot context whose global system prompt to augment.

packages/boot/app-boot/tests/app-boot.spec.ts

@@ -894,10 +894,13 @@ describe('addHarnessSourceSection', () => {
   const SOURCE_ROOT = `${sep}opt${sep}harness-src`
   const EXPECTED = `The DeepSeek Harness implementation checkout is at ${SOURCE_ROOT}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.`
 
-  it('distinguishes the source path from the current workdir between identity and persona', async () => {
+  it('distinguishes the source path from the current workdir after reusable instructions', async () => {
     const ctx = new Context()
     try {
       await ctx.plugin(SystemPrompt, { persona: 'You are a coding agent.' })
+      ctx.systemPrompt.section({
+        name: 'tools:sdk', order: ctx.systemPrompt.getSectionOrder('TOOLS_SDK'), text: 'Reusable tool SDK.',
+      })
       const dispose = addHarnessSourceSection(ctx, SOURCE_ROOT)
       expect(dispose).toBeTypeOf('function')
       const systemPrompt = ctx.get('systemPrompt')!
@@ -910,7 +913,9 @@ describe('addHarnessSourceSection', () => {
       const personaAt = rendered.indexOf('You are a coding agent.')
       expect(identityAt).toBeGreaterThanOrEqual(0)
       expect(personaAt).toBeGreaterThanOrEqual(0)
-      expect(identityAt).toBeLessThan(sourceAt)
+      const sdkAt = rendered.indexOf('Reusable tool SDK.')
+      expect(sdkAt).toBeGreaterThan(identityAt)
+      expect(sdkAt).toBeLessThan(sourceAt)
       expect(sourceAt).toBeLessThan(personaAt)
     } finally {
       await ctx.fiber.dispose()

packages/bundle/acp-app/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 packages/bundle/acp-app/README.md
-README.md: d27beb0882b1ea1da894ff89ebaa62011df548bc
-README.zh.md: ac60cd408db2df746d989f27433841d1eaa13386
+README.md: 78a80e2699ebcf4da7f44ac6e06e957d3c346215
+README.zh.md: 05232e5ce71297ab7bf0adfce6b87b525582ad27

packages/bundle/acp-app/README.md

@@ -44,7 +44,7 @@ The complete supported method matrix, MCP trust model, update mapping, and stop
 
 #### What the model sees
 
-The profile supplies `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.` before the base tool and context contributions. The ACP row's route and each `session/new` cwd resolve the placeholders.
+The profile supplies `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.` after the first-party reusable instructions. The ACP row's route and each `session/new` cwd resolve the placeholders.
 
 #### Token effect
 

packages/bundle/acp-app/README.zh.md

@@ -44,7 +44,7 @@ ACP v1 SDK 客户端先初始化 `dsh --profile acp`，再用绝对 `cwd` 与可
 
 #### 模型看到什么
 
-在 base 的工具和上下文贡献之前，profile 提供 `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.`。ACP 配置项的路由与每个 `session/new` 的 cwd 会解析其中的占位符。
+在第一方可复用指令之后，profile 提供 `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.`。ACP 配置项的路由与每个 `session/new` 的 cwd 会解析其中的占位符。
 
 #### Token 影响
 

packages/bundle/sdk-app/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 packages/bundle/sdk-app/README.md
-README.md: 68ea9670ea3feee699bf1cce501472e265006882
-README.zh.md: c342e3d90508a8d147dd38800c3d314b619c6e36
+README.md: 97e2d62b344dd9976a8651a03345f6cb02133b67
+README.zh.md: e60cd47a1d831244591b5c521c597c8c9b594ab9

packages/bundle/sdk-app/README.md

@@ -42,7 +42,7 @@ The SDK uses the base `read`, `write`, and `edit` defaults. To add `str_replace_
 
 #### What the model sees
 
-The profile supplies `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.` before the base tool and context contributions. The exact SDK initialization route and session cwd resolve the placeholders. Default file tool schemas include `read`, `write`, and `edit`; they omit `str_replace_editor`.
+The profile supplies `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.` after the first-party reusable instructions. The exact SDK initialization route and session cwd resolve the placeholders. Default file tool schemas include `read`, `write`, and `edit`; they omit `str_replace_editor`.
 
 #### Token effect
 

packages/bundle/sdk-app/README.zh.md

@@ -42,7 +42,7 @@ SDK 使用 base 默认提供的 `read`、`write` 和 `edit`。要添加 `str_rep
 
 #### 模型看到什么
 
-profile 会在 base 工具与上下文贡献之前提供 `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.`。确切的 SDK 初始化路由与会话 cwd 会解析其中的占位符。默认文件工具 schema 包含 `read`、`write` 和 `edit`，不包含 `str_replace_editor`。
+profile 会在第一方可复用指令之后提供 `You are a coding agent powered by the {{model}} model. Your working directory is {{cwd}}.`。确切的 SDK 初始化路由与会话 cwd 会解析其中的占位符。默认文件工具 schema 包含 `read`、`write` 和 `edit`，不包含 `str_replace_editor`。
 
 #### Token 影响
 

packages/bundle/web-app/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 packages/bundle/web-app/README.md
-README.md: c352ec937ecfa51f36eae1970067a62aed51b643
-README.zh.md: cc3b5e5660cd7fdb38f6d9084669491b1df27ee7
+README.md: 57c63702e59133133b3c59f9bee2a3aff52e5a84
+README.zh.md: a345e55cb3487117b11cd7c05a8d6daa1a3b210a

packages/bundle/web-app/README.md

@@ -124,7 +124,7 @@ Read these pages when you want to go deeper into the shared core, the browser re
 
 #### What the model sees
 
-When `surfaceContext` is true, the `harness:source` section identifies the on-disk Harness implementation without claiming it is the working directory, and the `app:web-surface` global section (first-party order −800) orients the model to the GUI: the canonical local URL, the "this page" referent, the update contract (the reload receiver is always on; no-refresh reloads additionally need the `pnpm run dev:web` watcher), and the instruction not to start replacement servers. `DSH_WEB_URL` additionally appears in the managed bash environment with its description, resolved per invocation from the live server. When it is false, neither section nor the variable is registered.
+When `surfaceContext` is true, the `harness:source` section identifies the on-disk Harness implementation without claiming it is the working directory, and the `app:web-surface` global section (first-party order 10100, after reusable instructions) orients the model to the GUI: the canonical local URL, the "this page" referent, the update contract (the reload receiver is always on; no-refresh reloads additionally need the `pnpm run dev:web` watcher), and the instruction not to start replacement servers. `DSH_WEB_URL` additionally appears in the managed bash environment with its description, resolved per invocation from the live server. When it is false, neither section nor the variable is registered.
 
 #### Token effect
 
@@ -132,7 +132,7 @@ One source line and one prompt paragraph per session plus two managed-environmen
 
 #### KV Cache effect
 
-The prompt section sits near the system prompt's head and is stable for the life of the process (the port is a boot fact), so it does not invalidate the cache across turns.
+Source and Web sections follow first-party reusable instructions. Different checkout paths or local ports leave that preceding prefix unchanged when tools and configuration match; provider cache reuse is not guaranteed.
 
 ## Known Limitations and Deferred Work
 

packages/bundle/web-app/README.zh.md

@@ -124,7 +124,7 @@ URL 行与浏览器交接都是就绪信号：监督方一观察到该行就发
 
 #### 模型看到什么
 
-当 `surfaceContext` 为 true 时，`harness:source` 段落标明磁盘上的 Harness 实现，但不会声称它就是工作目录；全局段落 `app:web-surface`（first-party 顺序 −800）则向模型说明 GUI：规范的本地 URL、「this page」指代什么、更新约定（重载接收端始终开启；无刷新重载还需要 `pnpm run dev:web` watcher），以及不要启动替代服务器的指令。`DSH_WEB_URL` 还会连同描述出现在受管 bash 环境中，每次调用时从运行中的服务器解析。当它为 false 时，这两个段落和该变量都不会注册。
+当 `surfaceContext` 为 true 时，`harness:source` 段落标明磁盘上的 Harness 实现，但不会声称它就是工作目录；全局段落 `app:web-surface`（first-party 顺序 10100，位于可复用指令之后）则向模型说明 GUI：规范的本地 URL、「this page」指代什么、更新约定（重载接收端始终开启；无刷新重载还需要 `pnpm run dev:web` watcher），以及不要启动替代服务器的指令。`DSH_WEB_URL` 还会连同描述出现在受管 bash 环境中，每次调用时从运行中的服务器解析。当它为 false 时，这两个段落和该变量都不会注册。
 
 #### Token 影响
 
@@ -132,7 +132,7 @@ URL 行与浏览器交接都是就绪信号：监督方一观察到该行就发
 
 #### KV Cache 影响
 
-该提示词段落位于系统提示词靠前位置，且在进程整个生命周期内稳定（端口是启动期事实），因此不会使跨轮次缓存失效。
+源码与 Web 段落位于第一方可复用指令之后。工具与配置一致时，不同 checkout 路径或本地端口不会改变前置前缀；不保证提供方复用缓存。
 
 ## 已知限制与延期工作
 

packages/core/agent-loop/tests/loop.spec.ts

@@ -501,7 +501,7 @@ describe('agent loop', () => {
     expect(types).toContain('tool/result')
   })
 
-  it('renders harness identity, then the persona, then tool guidance — with {{variables}} resolved', async () => {
+  it('renders harness identity and tool guidance before the interpolated persona', async () => {
     const adapter = new MockAdapter([textResponse('ok')])
     // The persona is a TEMPLATE: {{model}} is the loop-registered variable
     // projecting this agent's configured model, so the model knows its own name.
@@ -521,7 +521,7 @@ describe('agent loop', () => {
     await waitForIdle(ctx, agent)
 
     const request = adapter.requests[0]
-    expect(request!.system).toBe('You are an AI agent powered by DeepSeek Harness.\n\nYou are a test agent on mock.\n\nUse the noop tool wisely.')
+    expect(request!.system).toBe('You are an AI agent powered by DeepSeek Harness.\n\nUse the noop tool wisely.\n\nYou are a test agent on mock.')
     expect(request!.tools?.map(t => t.name)).toEqual(['noop'])
   })
 

packages/core/system-prompt/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 packages/core/system-prompt/README.md
-README.md: 313a71e74f2fc889bd6d765fbb58a367a3afc36a
-README.zh.md: 41e3445135fadb0143c3ef3538a8c59a83e6d32b
+README.md: 75b7ec4dd477f716195ef4fea824848a3db7a64f
+README.zh.md: 215ba6e83fdfe284cb9a21f425c8c87790906100

packages/core/system-prompt/README.md

@@ -44,7 +44,7 @@ The config owns the fixed opener, runtime context, deployment persona, and tool
 |---|---|---|
 | `includeHarnessIdentity` | `true` | Include the fixed `You are an AI agent powered by DeepSeek Harness.` first-party opener at order −1000. Set false only when a compatibility deployment owns the complete system prompt. |
 | `includeRuntimeContext` | `true` | Include ordered dynamic runtime context in assembly |
-| `persona` | `''` | The global deployment-persona prompt fragment, rendered at order `0` |
+| `persona` | `''` | The global deployment-persona prompt fragment, rendered at order `10200` after first-party reusable instructions |
 | `toolOrder` | — | Explicit model-facing tool order with one `'<unlisted-tools>'` rest entry |
 
 The generated [configuration catalog](../../../docs/config-catalog.md#deepseek-aidsh-system-prompt) is the exhaustive source for every accepted field. A `toolOrder` list without exactly one rest entry or with duplicates fails at load; a listed name with no registered tool rejects every `assemble()`.
@@ -130,7 +130,7 @@ The package-level contract is enough for most consumers; read these when you nee
 
 #### What the model sees
 
-By default every assembly starts with the harness identity below, then the configured persona and ordered plugin sections after strict variable interpolation. `includeHarnessIdentity: false` omits only that fixed opener. Empty sections disappear; scoped sections and variables can shadow globals for one agent. The `system-prompt/assemble` waterfall determines the delivered prompt and tool schemas unless one effective section declares itself complete — that exact section then becomes the whole system prompt while the waterfall's contexts, tools, and variables remain. Ordered dynamic contexts are separate from sections and become sourced user-role snapshots only when present; `includeRuntimeContext: false` or a scoped suppressor removes them all.
+First-party sections render the harness identity, reusable instructions (including the generated tools SDK and structured-output guidance), then the environment-bearing suffix: harness source (`10000`), Web surface (`10100`), and deployment persona (`10200`). External section orders and assembly listeners remain authoritative. `includeHarnessIdentity: false` omits only that fixed opener. Empty sections disappear; scoped sections and variables can shadow globals for one agent. The `system-prompt/assemble` waterfall determines the delivered prompt and tool schemas unless one effective section declares itself complete — that exact section then becomes the whole system prompt while the waterfall's contexts, tools, and variables remain. Ordered dynamic contexts are separate from sections and become sourced user-role snapshots only when present; `includeRuntimeContext: false` or a scoped suppressor removes them all.
 
 ##### Harness identity
 
@@ -144,7 +144,7 @@ Identity is a fixed per-request cost when enabled. Persona and plugin text are r
 
 #### KV Cache effect
 
-Prefix-stable while identity, persona, variables, section text, and order render identically. Any change may invalidate reuse from the first changed system-prompt token.
+With matching tools, configuration, and preceding instructions, different source paths, local Web URLs, or persona variables leave the reusable first-party prefix unchanged. Any change may invalidate reuse from the first changed token; provider cache sharing and measured hit rates are not guaranteed.
 
 ### Tool schemas
 

packages/core/system-prompt/README.zh.md

@@ -44,7 +44,7 @@ kind: "package-reference"
 |---|---|---|
 | `includeHarnessIdentity` | `true` | 是否包含顺序为 −1000 的 first-party 固定开场白 `You are an AI agent powered by DeepSeek Harness.`。仅当兼容性部署拥有完整系统提示词时设为 false。 |
 | `includeRuntimeContext` | `true` | 是否在组装中包含有序动态 runtime 上下文 |
-| `persona` | `''` | 全局部署 persona 提示词片段，渲染在顺序 `0` |
+| `persona` | `''` | 全局部署 persona 提示词片段，渲染在第一方可复用指令之后的顺序 `10200` |
 | `toolOrder` | — | 显式面向模型工具顺序，含一个 `'<unlisted-tools>'` 其余项标记 |
 
 生成的[配置目录](../../../docs/config-catalog.zh.md#deepseek-aidsh-system-prompt)是每个受支持字段的穷尽式真源。没有恰好一个其余项或存在重复项的 `toolOrder` 列表会在加载时失败；已列名称没有对应已注册工具会使每次 `assemble()` 被拒绝。
@@ -130,7 +130,7 @@ ctx.systemPrompt.variable('cwd', ({ agent }) => agent?.session.header.cwd)
 
 #### 模型看到什么
 
-默认情况下，每次组装都从下方 harness 身份开始，然后在严格变量插值后追加已配置 persona 与有序插件段。`includeHarnessIdentity: false` 仅省略这个固定开场白。空段会消失；带作用域的段与变量可以为一个 agent 遮蔽全局项。`system-prompt/assemble` waterfall 决定交付的提示词与工具 schema，除非一个有效段声明自身为 complete——此时该确切段会成为完整的系统提示词，而 waterfall 得到的上下文、工具与变量保持不变。有序动态上下文与段分离，只在存在时才会成为带来源的 user 角色快照；`includeRuntimeContext: false` 或带作用域的抑制器会移除全部这类上下文。
+第一方段落依次渲染 harness 身份、可复用指令（包括生成的工具 SDK 和结构化输出指导），最后是携带环境信息的后缀：harness 源码（`10000`）、Web 表层（`10100`）和部署 persona（`10200`）。外部段落的顺序与组装监听器仍决定其最终结果。`includeHarnessIdentity: false` 仅省略这个固定开场白。空段会消失；带作用域的段与变量可以为一个 agent 遮蔽全局项。`system-prompt/assemble` waterfall 决定交付的提示词与工具 schema，除非一个有效段声明自身为 complete——此时该确切段会成为完整的系统提示词，而 waterfall 得到的上下文、工具与变量保持不变。有序动态上下文与段分离，只在存在时才会成为带来源的 user 角色快照；`includeRuntimeContext: false` 或带作用域的抑制器会移除全部这类上下文。
 
 ##### harness 身份
 
@@ -144,7 +144,7 @@ You are an AI agent powered by DeepSeek Harness.
 
 #### KV Cache 影响
 
-只要身份、persona、变量、段文本与顺序的渲染完全相同，前缀就保持稳定。任何变更都可能从第一个变化的系统提示词 token 起使复用失效。
+工具、配置与前置指令一致时，不同源码路径、本地 Web URL 或 persona 变量不会改变可复用的第一方前缀。任何变更都可能从第一个变化的 token 起使复用失效；不保证提供方共享缓存或实际命中率。
 
 ### 工具 schema
 

packages/core/system-prompt/src/index.ts

@@ -120,9 +120,6 @@ export interface PromptAssembly {
 
 const SECTION_ORDERS = {
   HARNESS_IDENTITY: -1000,
-  HARNESS_SOURCE: -900,
-  WEB_SURFACE: -800,
-  DEPLOYMENT_PERSONA: 0,
   PLAN_POLICY: 500,
   TEAM_POLICY: 600,
   PTC_ONLY: 800,
@@ -149,6 +146,10 @@ const SECTION_ORDERS = {
   TOOLS_SDK: 5000,
   DELIVERABLE_FILE_REFERENCES: 9000,
   STRUCTURED_OUTPUT: 9900,
+  // Local paths, endpoints, and interpolated personas follow reusable instructions.
+  HARNESS_SOURCE: 10000,
+  WEB_SURFACE: 10100,
+  DEPLOYMENT_PERSONA: 10200,
 } as const
 
 /** Name of a centrally allocated prompt-section position. */
@@ -240,7 +241,7 @@ export interface Config {
   /** Include dynamic runtime-context snapshots in model history (default true). */
   includeRuntimeContext?: boolean
   /**
-   * Deployment-wide order-0 persona template. A scoped section named
+   * Deployment-wide persona template after first-party guidance. A scoped section named
    * `deployment:persona` shadows it; `{{variable}}` references are strict.
    */
   persona?: string

packages/core/system-prompt/tests/system-prompt.spec.ts

@@ -14,13 +14,14 @@ import type { PromptContextOrderName, PromptSectionOrderName } from '@deepseek-a
 const BUILT_IN = ['harness:identity', 'deployment:persona']
 const IDENTITY = 'You are an AI agent powered by DeepSeek Harness.'
 const SECTION_ORDER_NAMES = [
-  'HARNESS_IDENTITY', 'HARNESS_SOURCE', 'WEB_SURFACE', 'DEPLOYMENT_PERSONA',
+  'HARNESS_IDENTITY',
   'PLAN_POLICY', 'TEAM_POLICY', 'PTC_ONLY', 'FILE_REFERENCE', 'TOOL_BASH',
   'TOOL_PWSH', 'TOOL_READ', 'TOOL_WRITE', 'TOOL_EDIT', 'TOOL_GLOB',
   'TOOL_GREP', 'TOOL_JOBS', 'TOOL_PTY', 'TOOL_WEB_SEARCH', 'TOOL_WEB_FETCH',
   'TOOL_LSP', 'TOOL_SESSION_QUERY', 'TOOL_GOAL', 'TOOL_CORDIS', 'TOOL_WORKFLOW',
   'TOOL_RALPH', 'TOOL_SUBAGENT', 'TOOL_REPORT', 'TOOLS_SDK',
   'DELIVERABLE_FILE_REFERENCES', 'STRUCTURED_OUTPUT',
+  'HARNESS_SOURCE', 'WEB_SURFACE', 'DEPLOYMENT_PERSONA',
 ] as const satisfies readonly PromptSectionOrderName[]
 const CONTEXT_ORDER_NAMES = [
   'SANDBOX_POLICY', 'APPROVAL_POLICY', 'SUBAGENT_DELEGATION',
@@ -40,6 +41,36 @@ describe('SystemPrompt', () => {
     expect(sorted.slice(1).every((order, index) => order - sorted[index]! >= 10)).toBe(true)
   })
 
+  it('keeps reusable instructions identical across local environments', async () => {
+    const ctx = new Context()
+    try {
+      await ctx.plugin(SystemPrompt, { persona: 'Model {{model}} in {{cwd}} on {{platform}}.' })
+      let environment = { model: 'model-a', cwd: '/alice/project', platform: 'darwin', source: '/alice/dsh', url: 'http://127.0.0.1:3080' }
+      for (const key of ['model', 'cwd', 'platform'] as const) {
+        ctx.systemPrompt.variable(key, () => environment[key])
+      }
+      const reusable = SECTION_ORDER_NAMES.filter(name =>
+        !['HARNESS_IDENTITY', 'HARNESS_SOURCE', 'WEB_SURFACE', 'DEPLOYMENT_PERSONA'].includes(name))
+      for (const name of [...reusable].reverse()) {
+        ctx.systemPrompt.section({ name, order: ctx.systemPrompt.getSectionOrder(name), text: name })
+      }
+      ctx.systemPrompt.section({
+        name: 'source', order: ctx.systemPrompt.getSectionOrder('HARNESS_SOURCE'), text: () => environment.source,
+      })
+      ctx.systemPrompt.section({
+        name: 'web', order: ctx.systemPrompt.getSectionOrder('WEB_SURFACE'), text: () => environment.url,
+      })
+      const first = renderPrompt(await ctx.systemPrompt.assemble())
+      environment = { model: 'model-b', cwd: 'C:/bob/project', platform: 'win32', source: 'C:/bob/dsh', url: 'http://127.0.0.1:4080' }
+      const second = renderPrompt(await ctx.systemPrompt.assemble())
+      const prefix = [IDENTITY, ...reusable].join('\n\n') + '\n\n'
+      expect(first).toBe(prefix + '/alice/dsh\n\nhttp://127.0.0.1:3080\n\nModel model-a in /alice/project on darwin.')
+      expect(second).toBe(prefix + 'C:/bob/dsh\n\nhttp://127.0.0.1:4080\n\nModel model-b in C:/bob/project on win32.')
+    } finally {
+      await ctx.fiber.dispose()
+    }
+  })
+
   it('keeps repository context placements unique and integral', async () => {
     const ctx = new Context()
     await ctx.plugin(SystemPrompt, {})
@@ -121,15 +152,15 @@ describe('SystemPrompt', () => {
     ctx.systemPrompt.tools(() => ({ schemas: [{ name: 'echo', description: 'echo back', parameters: {} }] }))
 
     const assembly = await ctx.systemPrompt.assemble()
-    expect(assembly.sections.map(s => s.name)).toEqual(['harness:identity', 'deployment:persona', 'rules', 'cwd'])
-    expect(assembly.sections.map(s => s.text)).toEqual([IDENTITY, 'You are DeepSeek Harness.', 'Be precise.', 'cwd: /tmp'])
+    expect(assembly.sections.map(s => s.name)).toEqual(['harness:identity', 'rules', 'cwd', 'deployment:persona'])
+    expect(assembly.sections.map(s => s.text)).toEqual([IDENTITY, 'Be precise.', 'cwd: /tmp', 'You are DeepSeek Harness.'])
     expect(assembly.contexts).toEqual([
       { name: 'earlier', text: 'context 1' },
       { name: 'later', text: 'context 2' },
     ])
     expect(assembly.tools).toEqual([{ name: 'echo', description: 'echo back', parameters: {} }])
     expect(assembly.variables).toEqual({})
-    expect(renderPrompt(assembly)).toBe(`${IDENTITY}\n\nYou are DeepSeek Harness.\n\nBe precise.\n\ncwd: /tmp`)
+    expect(renderPrompt(assembly)).toBe(`${IDENTITY}\n\nBe precise.\n\ncwd: /tmp\n\nYou are DeepSeek Harness.`)
     expect(renderContextSnapshot(assembly)).toBe('Current runtime context. This snapshot supersedes earlier runtime-context snapshots.\n\ncontext 1\n\ncontext 2')
   })
 
@@ -307,8 +338,8 @@ describe('SystemPrompt', () => {
 
     const passed: AssembleContext = {}
     const assembly = await ctx.systemPrompt.assemble(passed)
-    expect(seen).toEqual([['harness:identity', 'deployment:persona', 'base', 'from-a']])
-    expect(assembly.sections.map(s => s.name)).toEqual(['harness:identity', 'deployment:persona', 'base', 'from-a'])
+    expect(seen).toEqual([['harness:identity', 'base', 'deployment:persona', 'from-a']])
+    expect(assembly.sections.map(s => s.name)).toEqual(['harness:identity', 'base', 'deployment:persona', 'from-a'])
     expect(contexts[0]).toBe(passed) // the caller's context reaches listeners
   })
 
@@ -368,7 +399,7 @@ describe('SystemPrompt', () => {
     firstParameters.properties['leak'] = { type: 'string' }
 
     const second = await ctx.systemPrompt.assemble()
-    expect(second.sections.map(section => section.name)).toEqual(['harness:identity', 'deployment:persona', 'base'])
+    expect(second.sections.map(section => section.name)).toEqual(['harness:identity', 'base', 'deployment:persona'])
     expect(second.sections[0]!.text).toBe(IDENTITY)
     expect(second.contexts).toEqual([])
     expect(second.tools).toEqual([{ name: 't', description: 'tool', parameters: { type: 'object', properties: {} } }])

packages/preset/persona/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 packages/preset/persona/README.md
-README.md: 753e40ebcb1848635ba4497d686b6ac8a56c31e3
-README.zh.md: 141784dc6d4214aae83298713327a1896445f41b
+README.md: 6f15d24f25a063f6699b968a7cc5d5e58e0fec2f
+README.zh.md: 9263b272a9e37eaf76d9cd211f2f2b668cb6d8e0

packages/preset/persona/README.md

@@ -61,7 +61,7 @@ Use this row when a preset must change an agent's identity and not only its tool
 
 ### How the row registers
 
-`apply` registers one prompt section through `ctx.systemPrompt.section({ name: PERSONA_SECTION, order: ctx.systemPrompt.getSectionOrder('DEPLOYMENT_PERSONA'), text, complete? })` inside the mounting context's scope, so the section lands at order 0 — immediately after the harness identity opener — and only for agents joined to the preset. The shared section name makes a preset persona shadow the deployment's instead of landing beside it, while the service-owned order lookup keeps repository contributors on the central allocation. `includeRuntimeContext: false` calls `ctx.systemPrompt.suppressRuntimeContext()`.
+`apply` registers one prompt section through `ctx.systemPrompt.section({ name: PERSONA_SECTION, order: ctx.systemPrompt.getSectionOrder('DEPLOYMENT_PERSONA'), text, complete? })` inside the mounting context's scope, so the section lands at order 10200 — after first-party reusable instructions — and only for agents joined to the preset. The shared section name makes a preset persona shadow the deployment's instead of landing beside it, while the service-owned order lookup keeps repository contributors on the central allocation. `includeRuntimeContext: false` calls `ctx.systemPrompt.suppressRuntimeContext()`.
 
 ### Why the row is scope-only
 
@@ -96,7 +96,7 @@ Read these pages when the package-level contract is not enough; they move from t
 
 #### What the model sees
 
-The `deployment:persona` section at order 0, immediately after the harness identity opener, carrying exactly this row's configured `text` with prompt variables resolved. For an agent whose preset mounts this row, it replaces whatever persona the deployment configured. In complete mode, the model sees only this rendered section as its system prompt. Runtime context remains enabled by default; when disabled, a fresh agent receives no runtime-context snapshot from sandbox policy, approval policy, delegation, or another system-prompt context provider.
+The `deployment:persona` section at order 10200, after first-party reusable instructions, carrying exactly this row's configured `text` with prompt variables resolved. For an agent whose preset mounts this row, it replaces whatever persona the deployment configured. In complete mode, the model sees only this rendered section as its system prompt. Runtime context remains enabled by default; when disabled, a fresh agent receives no runtime-context snapshot from sandbox policy, approval policy, delegation, or another system-prompt context provider.
 
 #### Token effect
 
@@ -104,7 +104,7 @@ Fixed for a given preset: the persona's own tokens on every request that agent m
 
 #### KV Cache effect
 
-Prefix-stable for the life of an agent — the row mounts once, before the agent is published and therefore before its first request, and its text never changes while the agent runs. Two agents on different presets establish different prefixes from this section onward; neither can invalidate the other's reuse.
+Prefix-stable while the rendered template variables and text are unchanged. Different personas can share the preceding first-party instructions when tools and configuration match; provider cache sharing is not guaranteed.
 
 ## Known Limitations and Deferred Work
 

packages/preset/persona/README.zh.md

@@ -61,7 +61,7 @@ kind: "package-reference"
 
 ### 本行如何注册
 
-`apply` 在挂载上下文的 scope 内通过 `ctx.systemPrompt.section({ name: PERSONA_SECTION, order: ctx.systemPrompt.getSectionOrder('DEPLOYMENT_PERSONA'), text, complete? })` 注册一个提示词段落，因此该段落落在 order 0——紧随 harness 身份开场白之后——且只对加入该 preset 的 agent 生效。共享段落名让 preset 人设遮蔽部署人设，而不是落在它旁边；服务持有的 order 查询则让仓库自带贡献方服从集中分配。`includeRuntimeContext: false` 会调用 `ctx.systemPrompt.suppressRuntimeContext()`。
+`apply` 在挂载上下文的 scope 内通过 `ctx.systemPrompt.section({ name: PERSONA_SECTION, order: ctx.systemPrompt.getSectionOrder('DEPLOYMENT_PERSONA'), text, complete? })` 注册一个提示词段落，因此该段落落在 order 10200——位于第一方可复用指令之后——且只对加入该 preset 的 agent 生效。共享段落名让 preset 人设遮蔽部署人设，而不是落在它旁边；服务持有的 order 查询则让仓库自带贡献方服从集中分配。`includeRuntimeContext: false` 会调用 `ctx.systemPrompt.suppressRuntimeContext()`。
 
 ### 本行为何仅限 scope 内使用
 
@@ -96,7 +96,7 @@ kind: "package-reference"
 
 #### 模型看到什么
 
-位于 order 0 的 `deployment:persona` 段落，紧随 harness 身份开场白之后，携带本行配置的 `text`，其中的提示词变量已解析。对于其 preset 挂载了本行的 agent，它会替换部署所配置的任何人设。在完整模式下，模型只会看到这个渲染后的段落作为系统提示词。Runtime context 默认保持启用；禁用后，新建 agent 不会收到来自沙箱策略、批准策略、委派或其他 system-prompt 上下文提供方的 runtime-context 快照。
+位于 order 10200 的 `deployment:persona` 段落，在第一方可复用指令之后，携带本行配置的 `text`，其中的提示词变量已解析。对于其 preset 挂载了本行的 agent，它会替换部署所配置的任何人设。在完整模式下，模型只会看到这个渲染后的段落作为系统提示词。Runtime context 默认保持启用；禁用后，新建 agent 不会收到来自沙箱策略、批准策略、委派或其他 system-prompt 上下文提供方的 runtime-context 快照。
 
 #### Token 影响
 
@@ -104,7 +104,7 @@ kind: "package-reference"
 
 #### KV Cache 影响
 
-在一个 agent 的整个生命周期内保持前缀稳定——本行只挂载一次，发生在 agent 发布之前、因而也在它的首个请求之前，且在 agent 运行期间文本不再改变。两个使用不同 preset 的 agent 从该段落起建立各自不同的前缀，谁都无法让对方失去缓存复用。
+渲染后的模板变量与文本不变时，前缀保持稳定。不同 persona 在工具与配置一致时可以共享前置的第一方指令；不保证提供方共享缓存。
 
 ## 已知限制与延期工作
 

snapshots/sdk/bash-tool/system-prompt.expected.md

@@ -1,7 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -25,3 +23,5 @@ Use goal tools for one long-running completion objective in the current session.
 Use the workflow tool ONLY when the user explicitly asks for a workflow or for large multi-agent orchestration: you write a JavaScript script (the tool description documents the exact format) that fans work out across many subagents with phases and structured results. For one or two delegations, prefer plain subagent calls.
 
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
+
+You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.

snapshots/sdk/subagent-continuable-inheritance/system-prompt.1.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -30,3 +25,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/sdk/subagent-continuable/system-prompt.1.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -30,3 +25,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/sdk/subagent-dsh-sdk-diagnostic/system-prompt.expected.md

@@ -1,7 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -27,3 +25,5 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.

snapshots/sdk/subagent-dsh-sdk-dynamic-route/system-prompt.1.expected.md

@@ -1,7 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-Echo where you run.
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -27,3 +25,5 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+Echo where you run.

snapshots/sdk/subagent-dsh-sdk-dynamic-route/system-prompt.expected.md

@@ -1,7 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding agent powered by the mock-delegate model. Your working directory is {{cwd}}.
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -25,3 +23,5 @@ Use goal tools for one long-running completion objective in the current session.
 Use the workflow tool ONLY when the user explicitly asks for a workflow or for large multi-agent orchestration: you write a JavaScript script (the tool description documents the exact format) that fans work out across many subagents with phases and structured results. For one or two delegations, prefer plain subagent calls.
 
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
+
+You are a coding agent powered by the mock-delegate model. Your working directory is {{cwd}}.

snapshots/sdk/subagent-list-agents/system-prompt.1.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -30,3 +25,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/sdk/subagent-send-message/system-prompt.1.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -30,3 +25,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/sdk/text-turn/system-prompt.expected.md

@@ -1,7 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -25,3 +23,5 @@ Use goal tools for one long-running completion objective in the current session.
 Use the workflow tool ONLY when the user explicitly asks for a workflow or for large multi-agent orchestration: you write a JavaScript script (the tool description documents the exact format) that fans work out across many subagents with phases and structured results. For one or two delegations, prefer plain subagent calls.
 
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
+
+You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.

snapshots/session/agent-instructions/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -31,14 +26,13 @@ Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop o
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
 
-<!-- request/header change 1 -->
-
-You are an AI agent powered by DeepSeek Harness.
-
 You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
 
 Verify your work by running the code or tests. Keep answers brief and factual.
 
+<!-- request/header change 1 -->
+
+You are an AI agent powered by DeepSeek Harness.
 
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
@@ -65,3 +59,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/both-mode-turn/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -538,3 +533,7 @@ declare const tools: {
   [K in ToolName]: (args: ToolArgsMap[K]) => Promise<ToolOutputMap[K]>;
 }
 ```
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/compaction-recovery/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -31,14 +26,13 @@ Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop o
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
 
-<!-- request/header change 1 -->
-
-You are an AI agent powered by DeepSeek Harness.
-
 You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
 
 Verify your work by running the code or tests. Keep answers brief and factual.
 
+<!-- request/header change 1 -->
+
+You are an AI agent powered by DeepSeek Harness.
 
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
@@ -65,3 +59,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/cordis-inspect-jsdoc/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -724,3 +719,7 @@ declare const tools: {
   [K in ToolName]: (args: ToolArgsMap[K]) => Promise<ToolOutputMap[K]>;
 }
 ```
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/fs-glob-sampling/system-prompt.expected.md

@@ -1,7 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a concise snapshot agent working in {{cwd}}.
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -23,3 +21,5 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a concise snapshot agent working in {{cwd}}.

snapshots/session/lsp-definition/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -32,3 +27,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/product-subagent-codex/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -30,3 +25,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/ptc-python-turn/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 `run_code` is the only tool you can call directly — a tool call naming any other tool fails. Reach every tool the SDK declares below from inside the program.
 
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
@@ -609,3 +604,7 @@ class Tools(Protocol):
 
 tools: Tools
 ```
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/ptc-read-image/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash-vision-exp model. Your working directory is {{cwd}}.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 `run_code` is the only tool you can call directly — a tool call naming any other tool fails. Reach every tool the SDK declares below from inside the program.
 
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
@@ -540,3 +535,7 @@ declare const tools: {
   [K in ToolName]: (args: ToolArgsMap[K]) => Promise<ToolOutputMap[K]>;
 }
 ```
+
+You are a coding assistant powered by the deepseek-v4-flash-vision-exp model. Your working directory is {{cwd}}.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/ptc-turn/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 `run_code` is the only tool you can call directly — a tool call naming any other tool fails. Reach every tool the SDK declares below from inside the program.
 
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
@@ -540,3 +535,7 @@ declare const tools: {
   [K in ToolName]: (args: ToolArgsMap[K]) => Promise<ToolOutputMap[K]>;
 }
 ```
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/pty-tools-sandbox-backend/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -32,3 +27,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/pwsh-tool-turn/system-prompt.expected.md

@@ -1,7 +1,7 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a concise snapshot agent working in {{cwd}}.
-
 Non-zero exits are reported as `[exit code: N]` markers; investigate failures before moving on. On Windows a killed process settles as `[exit code: 1]` without a signal marker; treat a bare exit 1 after an interruption as a termination, not a command failure.
 
 Track every background job id you start. You are notified in-session when a job finishes — do not busy-poll or sleep on one; keep working on independent steps and do not duplicate a running job's work. Before giving a final answer, collect every still-relevant job with job_output (set wait: true only when you are genuinely blocked on it), and job_kill jobs that stopped mattering.
+
+You are a concise snapshot agent working in {{cwd}}.

snapshots/session/ralph-loop/system-prompt.1.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -32,3 +27,7 @@ Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop o
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
 
 When you have your final answer, you MUST report it by calling the `structured_output` tool with arguments matching its parameter schema exactly. Do not finish with a plain text answer: only the tool call counts as your result.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/ralph-loop/system-prompt.2.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -32,3 +27,7 @@ Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop o
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
 
 When you have your final answer, you MUST report it by calling the `structured_output` tool with arguments matching its parameter schema exactly. Do not finish with a plain text answer: only the tool call counts as your result.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/read-image/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash-vision-exp model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -30,3 +25,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash-vision-exp model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/session-query-spill/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -32,3 +27,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/text-turn/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -30,3 +25,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-flash model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/session/web-fetch/system-prompt.expected.md

@@ -1,10 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
-
-Verify your work by running the code or tests. Keep answers brief and factual.
-
-
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
 
 Use the read tool — not shell commands like cat — to inspect text files. Results include line numbers. Use offset and limit to continue reading large files.
@@ -28,3 +23,7 @@ Use the workflow tool ONLY when the user explicitly asks for a workflow or for l
 Use the ralph tool ONLY when the direct human explicitly asks for a Ralph loop or fresh-agent iterative execution. Each Ralph round starts a fresh child with no conversation seed and uses the shared workspace as durable memory. Completion and blockers are worker reports, not independent evaluation. Use same-session goal tools for ordinary long-running objectives, and plain subagents or workflows for bounded delegation and fan-out.
 
 Use subagent in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
+
+You are a coding assistant powered by the deepseek-v4-pro model. Your working directory is {{cwd}}. Your bash tool runs under a file sandbox — a `[sandbox: file access denied …]` result is policy, not a command bug.
+
+Verify your work by running the code or tests. Keep answers brief and factual.

snapshots/web/cordis-tool-round/system-prompt.expected.md

@@ -1,11 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-The DeepSeek Harness implementation checkout is at {{sourceRoot}}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.
-
-You are interacting with the user through the DeepSeek Harness Web GUI at {{webUrl}}. When the user refers to "this page", "this GUI", or "this app" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.
-
-You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
 Tokens prefixed with @ are workspace paths the user explicitly referenced, relative to the workspace root. A trailing slash marks a directory: list it when its contents matter. Anything else is a file: use the read tool when its contents are needed, and do not claim to have inspected it before reading. @"..." quotes a path containing spaces.
 
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
@@ -143,3 +137,9 @@ Use subagent in the background by default. Start independent delegations togethe
 Use subagent_fork in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
 
 When you successfully create or modify files, mention the primary outputs in your final response. To make those and any other changed-file references clickable in Web, format them as Markdown inline code using the exact file-tool path, or a basename when unique among the files changed in that turn.
+
+The DeepSeek Harness implementation checkout is at {{sourceRoot}}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.
+
+You are interacting with the user through the DeepSeek Harness Web GUI at {{webUrl}}. When the user refers to "this page", "this GUI", or "this app" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.
+
+You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.

snapshots/web/fresh-round-trip/system-prompt.expected.md

@@ -1,11 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-The DeepSeek Harness implementation checkout is at {{sourceRoot}}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.
-
-You are interacting with the user through the DeepSeek Harness Web GUI at {{webUrl}}. When the user refers to "this page", "this GUI", or "this app" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.
-
-You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
 Tokens prefixed with @ are workspace paths the user explicitly referenced, relative to the workspace root. A trailing slash marks a directory: list it when its contents matter. Anything else is a file: use the read tool when its contents are needed, and do not claim to have inspected it before reading. @"..." quotes a path containing spaces.
 
 Check the [exit code: N] marker on every bash result; investigate failures before moving on.
@@ -37,3 +31,9 @@ Use subagent in the background by default. Start independent delegations togethe
 Use subagent_fork in the background by default. Start independent delegations together in one assistant message and continue useful work while they run. Set `run_in_background: false` only when your next action depends on that subagent's result. When a background run settles, the runtime sends you a notice containing its outcome and any final assistant message.
 
 When you successfully create or modify files, mention the primary outputs in your final response. To make those and any other changed-file references clickable in Web, format them as Markdown inline code using the exact file-tool path, or a basename when unique among the files changed in that turn.
+
+The DeepSeek Harness implementation checkout is at {{sourceRoot}}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.
+
+You are interacting with the user through the DeepSeek Harness Web GUI at {{webUrl}}. When the user refers to "this page", "this GUI", or "this app" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.
+
+You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.

snapshots/web/fresh-round-trip/web-context.expected.md

@@ -1,5 +1,3 @@
-You are an AI agent powered by DeepSeek Harness.
-
 The DeepSeek Harness implementation checkout is at {{sourceRoot}}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.
 
 You are interacting with the user through the DeepSeek Harness Web GUI at {{webUrl}}. When the user refers to "this page", "this GUI", or "this app" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.

snapshots/web/ptc-round/system-prompt.expected.md

@@ -1,11 +1,5 @@
 You are an AI agent powered by DeepSeek Harness.
 
-The DeepSeek Harness implementation checkout is at {{sourceRoot}}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.
-
-You are interacting with the user through the DeepSeek Harness Web GUI at {{webUrl}}. When the user refers to "this page", "this GUI", or "this app" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.
-
-You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.
-
 `run_code` is the only tool you can call directly — a tool call naming any other tool fails. Reach every tool the SDK declares below from inside the program.
 
 Tokens prefixed with @ are workspace paths the user explicitly referenced, relative to the workspace root. A trailing slash marks a directory: list it when its contents matter. Anything else is a file: use the read tool when its contents are needed, and do not claim to have inspected it before reading. @"..." quotes a path containing spaces.
@@ -543,3 +537,9 @@ declare const tools: {
 ```
 
 When you successfully create or modify files, mention the primary outputs in your final response. To make those and any other changed-file references clickable in Web, format them as Markdown inline code using the exact file-tool path, or a basename when unique among the files changed in that turn.
+
+The DeepSeek Harness implementation checkout is at {{sourceRoot}}. The checkout location and current working directory are separate values and may differ; never infer the working directory from this path. Use pwd to determine the current working directory. Use this checkout only to inspect or extend DSH itself.
+
+You are interacting with the user through the DeepSeek Harness Web GUI at {{webUrl}}. When the user refers to "this page", "this GUI", or "this app" without naming another target, they mean this GUI. The browser provides no implicit DOM, route, or screenshot context. The client-plugin HMR receiver is active, but client-plugin changes reload without a refresh only while `pnpm run dev:web` is also running from this same checkout to rebuild their bundles; verify that watcher before promising automatic updates. Every other change — the apps/web shell and plain packages — requires rebuilding the affected Web artifacts and verifying this existing URL after a page refresh. Starting another server does not update this GUI. The apps/web Vite entry builds the shell but is not a standalone application because only dsh web injects window.__DSH_BOOT__. Do not start a replacement server unless the user asks; if one is needed, use a managed background job and verify its exact URL.
+
+You are a coding agent powered by the deepseek-v4-flash model. Your working directory is {{cwd}}.