Add PR template to filter low-quality submissions

Requires contributors to articulate the problem they're solving,
confirm human review, document eval methodology, and check for
duplicate PRs. Informed by patterns in ~90 closed-without-merge PRs.

.github/PULL_REQUEST_TEMPLATE.md

@@ -0,0 +1,87 @@
+<!--
+BEFORE SUBMITTING: Read every word of this template. PRs that leave
+sections blank, contain multiple unrelated changes, or show no evidence
+of human involvement will be closed without review.
+-->
+
+## What problem are you trying to solve?
+<!-- Describe the specific problem you encountered. If this was a session
+     issue, include: what you were doing, what went wrong, the model's
+     exact failure mode, and ideally a transcript or session log.
+
+     "Improving" something is not a problem statement. What broke? What
+     failed? What was the user experience that motivated this? -->
+
+## What does this PR change?
+<!-- 1-3 sentences. What, not why — the "why" belongs above. -->
+
+## Is this change appropriate for the core library?
+<!-- Superpowers core contains general-purpose skills and infrastructure
+     that benefit all users. Ask yourself:
+
+     - Would this be useful to someone working on a completely different
+       kind of project than yours?
+     - Is this project-specific, team-specific, or tool-specific?
+     - Does this integrate or promote a third-party service?
+
+     If your change is a new skill for a specific domain, workflow tool,
+     or third-party integration, it belongs in its own plugin — not here.
+     See the plugin development docs for how to publish it separately. -->
+
+## What alternatives did you consider?
+<!-- What other approaches did you try or evaluate before landing on this
+     one? Why were they worse? If you didn't consider alternatives, say so
+     — but know that's a red flag. -->
+
+## Does this PR contain multiple unrelated changes?
+<!-- If yes: stop. Split it into separate PRs. Bundled PRs will be closed.
+     If you believe the changes are related, explain the dependency. -->
+
+## Existing PRs
+- [ ] I have reviewed all open AND closed PRs for duplicates or prior art
+- Related PRs: <!-- #number, #number, or "none found" -->
+
+<!-- If a related closed PR exists, explain what's different about your
+     approach and why it should succeed where the other didn't. -->
+
+## Environment tested
+
+| Harness (e.g. Claude Code, Cursor) | Harness version | Model | Model version/ID |
+|-------------------------------------|-----------------|-------|------------------|
+|                                     |                 |       |                  |
+
+## Evaluation
+- What was the initial prompt you (or your human partner) used to start
+  the session that led to this change?
+- How many eval sessions did you run AFTER making the change?
+- How did outcomes change compared to before the change?
+
+<!-- "It works" is not evaluation. Describe the before/after difference
+     you observed across multiple sessions. -->
+
+## Rigor
+
+- [ ] If this is a skills change: I used `superpowers:writing-skills` and
+      completed adversarial pressure testing (paste results below)
+- [ ] This change was tested adversarially, not just on the happy path
+- [ ] I did not modify carefully-tuned content (Red Flags table,
+      rationalizations, "human partner" language) without extensive evals
+      showing the change is an improvement
+
+<!-- If you changed wording in skills that shape agent behavior, show your
+     eval methodology and results. These are not prose — they are code. -->
+
+## Human review
+- [ ] A human has reviewed the COMPLETE proposed diff before submission
+
+<!--
+STOP. If the checkbox above is not checked, do not submit this PR.
+
+PRs will be closed without review if they:
+- Show no evidence of human involvement
+- Contain multiple unrelated changes
+- Promote or integrate third-party services or tools
+- Submit project-specific or personal configuration as core changes
+- Leave required sections blank or use placeholder text
+- Modify behavior-shaping content without eval evidence
+-->