ci: make the claude-pr-loop fix agent adversarial and scope-aware

[allxsmith]

Description

Item 1 of 3 from the Bun/robobun-inspired loop hardening.

The fix agent in claude-pr-loop.yml behaved like an order-taker — it tried to satisfy every CodeRabbit finding, including the out-of-scope "make Button/Link fully generic-polymorphic" ask that #188 explicitly deemed unnecessary. It burned its entire turn budget attempting that refactor and failed with error_max_turns (twice on #225, even after 40→80).

Fix: add a BE ADVERSARIAL directive to the fix prompt:

• Read the linked issue's scope and refute findings that exceed it (even technically-valid ones).
• Treat the opus deep review's clean verdict as a strong prior — findings in areas it cleared need a concrete, reproducible bug to act on.
• Never undertake a large type-system/architecture refactor to satisfy one reviewer.
• Refute = reply with a code-cited reason + push no code; let contested findings escalate to needs-human-review instead of thrashing.

This mirrors the "adversarial code review" + "verify the problem is real" model that Bun/robobun use (per this session's deep-research).

Affected package(s):

• Other: CI / GitHub Actions (.github/workflows/claude-pr-loop.yml)

Related Issue(s)

Refs #188. Unblocks #225 (the fixer will now refute the Major and converge the rest).

Type of Change

• Build tooling

Checklist

• Self-reviewed; YAML validated
• Prompt-only change to the fix job; no permission or tool changes

🤖 Generated with Claude Code

https://claude.ai/code/session_01NVR5yWevceZEFbTJmpviiM

---

Generated by Claude Code

Summary by CodeRabbit

• Chores
  • Improved the automated review workflow so it handles issue scope more strictly and only raises concrete, reproducible findings.
  • Added clearer behavior for cases where a finding is not supported, helping avoid unnecessary back-and-forth on out-of-scope changes.
  • Reduced the chance of broad, unrelated changes being suggested during automated fix attempts.

[coderabbitai]

Caution

Review failed

The pull request is closed.

ℹ️ Recent review info

⚙️ Run configuration

Configuration used: Path: .coderabbit.yaml

Review profile: CHILL

Plan: Pro Plus

Run ID: 35263c77-83d0-4197-83ea-9550154d2e46

📥 Commits

Reviewing files that changed from the base of the PR and between 337b231 and bfc39fa.

📒 Files selected for processing (1)

• .github/workflows/claude-pr-loop.yml

---

Walkthrough

The Claude PR loop workflow's fix prompt was updated to instruct Claude to act as a skeptical reviewer, treating linked issue scope as a strict contract, requiring reproducible findings, prohibiting large refactors, and defining explicit refute behavior for out-of-scope or disputed findings.

Changes

Claude Fix Prompt Update

Layer / File(s)	Summary
Adversarial reviewer prompt instructions .github/workflows/claude-pr-loop.yml	Fix prompt now enforces strict scope adherence, requires concrete reproducible defects, prohibits large refactors/architecture changes, and defines refute behavior that leaves contested threads open for human escalation.

Estimated code review effort: 1 (Trivial) | ~5 minutes

Possibly related PRs

• allxsmith/bestax#216: Modifies the same claude-pr-loop.yml fix prompt/refute logic introduced in this earlier PR for the autonomous Claude PR loop workflow.

✨ Finishing Touches

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch claude/adversarial-fixer

---

_{Comment @coderabbitai help to get the list of available commands.}

[github-actions]

Preview Deployment

Preview URL: https://73f96988.bestax.pages.dev

[github-actions]

🎉 This PR is included in version 5.2.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀

[github-actions]

🎉 This PR is included in version 3.2.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀

[github-actions]

🎉 This PR is included in version 1.0.0 🎉

The release is available on:

• npm package (@latest dist-tag)
• GitHub release

Your semantic-release bot 📦🚀