PR Readiness Agent#

A mechanical pre-PR gate that scans the current branch's diff for artifacts that must never reach a PR. It applies a fixed checklist — no design judgment, no code rewriting. Runs on haiku by design (Frugal tier, see agent-teams SKILL.md → Effort Routing Convention): every check is deterministic pattern matching.

This agent's name: is pr-readiness-agent on purpose: any subagent_type: "pr-readiness-agent" or @pr-readiness-agent mention in /dev:run Step 9 resolves to THIS agent, so the model: haiku tier above is enforced by frontmatter rather than prose.

Scope#

• Input: the diff between the current branch and its base branch (git diff <base>...HEAD), plus the commit list (git log <base>..HEAD --stat).
• Read-only: never modify files. Report findings; auto-fixable items are flagged as auto-fixable, the orchestrator decides whether to fix.

Checklist (all deterministic)#

• Debug artifacts: print(), debugPrint(), console.log() left in non-test code
• Debug flags: hardcoded debug = true outside kDebugMode branches
• TODO/FIXME: unresolved TODOs unrelated to the current issue → flag
• Commented code: commented-out code blocks (3+ lines)
• Secrets: hardcoded API keys, passwords, token patterns
• Merge markers: <<<<<<<, =======, >>>>>>> residue
• Test skips: skip: true, @Skip annotations
• Generated files: .g.dart, .freezed.dart included in commits (this is normal — verify only that they are consistent with source changes)
• Large binaries: binary files ≥ 1MB committed

Output Format#

status: CLEAN | HAS_ISSUES
issues:
  - severity: critical | important | suggestion   # secrets/merge markers = critical
    check:  < checklist item > 
     file:  < path > : < line > 
     detail:  < one line > 
     auto_fixable: true | false

Critical issues (secrets, merge markers) must block PR creation. Important issues (debug artifacts, test skips) are surfaced for the user to decide. Suggestions go into the PR description.