Add gh-aw guidance to Claude prompts

[github-actions]

Summary

• add formatting, rigor, workflow-edit guardrails, and MCP pagination guidance to all Claude composite-action prompt blocks
• keep existing footer behavior unchanged

Testing

• make lint

• Fixes Update Claude prompts to have best practices from our gh-aw workflows #132

AI generated by Issue Assistant

[strawgate]

/ai keep it specific to Claude please, Claude has no issues with pagination on mcp

[github-actions]

APPROVE

---

What is this? | From workflow: PR Review Agent

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not. for issue #142

[github-actions]

Removed the MCP pagination guidance from the Claude composite-action prompts to keep the instructions Claude-specific.

Tests: make lint.

---

What is this? | From workflow: PR Assistant

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not. for issue #142

[github-actions]

Summary: The Verify compiled workflows job fails in Ensure lock files are up to date because make compile leaves the working tree dirty; the log shows new/updated files under .github/workflows/.

Evidence from the job log:

Working tree dirty after make compile.
 M .github/workflows/agentics-maintenance.yml
 M .github/workflows/gh-aw-*.lock.yml
 ?? .github/workflows/trigger-release-update.yml

Recommended fix: run make compile and commit the resulting .github/workflows/agentics-maintenance.yml, updated .github/workflows/gh-aw-*.lock.yml files, and the new .github/workflows/trigger-release-update.yml. (I did not push changes because this workflow forbids editing .github/workflows/.)

Tests: make lint

---

What is this? | From workflow: PR Checks Fixer

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not.

[strawgate]

/ai claude has no concept of noop let's audit and make sure we didnt include any other nonsense

[github-actions]

Removed noop references from the Claude prompt rigor guidance and replaced them with explicit “no actionable findings” language across the affected composite actions. Tests: make lint.

---

What is this? | From workflow: PR Assistant

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not. for issue #142

[github-actions]

Verdict: REQUEST_CHANGES.

The prompt updates are missing MCP pagination guidance in the newly ported guidance block, which leaves a known failure mode (oversized MCP responses) unaddressed.

---

What is this? | From workflow: PR Review Agent

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not. for issue #142

[github-actions]

CI failure is caused by generated workflow artifacts being out of sync, not by a runtime/test failure.

• Failing job: Verify compiled workflows (run 22197059946)
• Failing step: Ensure lock files are up to date
• Log evidence: after make compile, CI checks git status --porcelain and exits because the tree is dirty (multiple .github/workflows/*.lock.yml updates plus missing/new trigger-*.yml files).

What I verified locally:

• Checked out PR head branch main-cd0472fd23de6cfb
• Ran make compile
• Confirmed it reproduces the same dirty-worktree condition CI reports (including modified lock files and new trigger workflow files)

Root cause:

• This PR updates Claude prompt sources under claude-workflows/**/action.yml but does not include the regenerated compiled workflow outputs/lockfiles.

Recommended fix:

1. Run make compile
2. Commit all generated workflow updates under .github/workflows/ (lockfiles + trigger files)
3. Push and re-run CI

I did not push a fix from this run because the required changes are in .github/workflows/, which this fixer is constrained from modifying.

---

What is this? | From workflow: PR Checks Fixer

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not.

[strawgate]

/ai how does this look? are we doing anything crazy?

[github-actions]

Short answer: this looks sane, not crazy.

I checked the current PR diff and the changes are tightly scoped to the 12 Claude action prompts under claude-workflows/**/action.yml, consistently adding the same three guidance blocks ((formatting_guidelines), (rigor), (workflow_edit_guardrails)) without altering tool permissions or execution behavior. The only likely point of debate is policy preference (e.g., MCP pagination wording), not a functional regression in the prompt plumbing.

---

What is this? | From workflow: PR Assistant

Give us feedback! React with 🚀 if perfect, 👍 if helpful, 👎 if not. for issue #142