AgentPlays
Maintainer-merge self-check before opening a PR (AgentPlays original)
AgentPlays original pre-PR checklist inspired by public FrontierCode grading ideas (scope, tests, no junk diffs). Use it to audit an agent-written change before you open a pull request. Does not claim your PR would pass FrontierCode.
Copy it
Review the current uncommitted or branch diff as if you were a careful open-source maintainer. Do not rewrite code yet. Produce a merge-readiness report only. Checks (answer YES / NO / N/A with evidence) 1) Intent: Restate the user request in one sentence. Flag any work that goes beyond it. 2) Scope: List every modified/added/deleted path. Flag files outside the needed surface area. 3) Diff size: Call out large or noisy hunks (formatting-only, renames without need, generated junk). 4) Tests: Which existing tests cover this? Which new tests were added? Are any tests weak, flaky, or testing implementation details instead of behavior? 5) Correctness: Walk the happy path and two failure paths. Cite file:line for risks. 6) API/compat: Any breaking public API, schema, flag, or migration? Say how callers must change. 7) Ops: Logging, metrics, feature flags, rollbacks—missing anything for a safe ship? 8) Secrets/safety: Secrets, PII, destructive defaults, or widened permissions? Output format - Verdict: Ready / Needs changes / Do not merge - Blocking issues (must fix): numbered list - Non-blocking nits: numbered list - Suggested commit/PR title + 3-bullet summary for humans - Exact next commands I should run (tests, linters) — do not run destructive commands Rules - Prefer evidence over vibes. Quote paths and line ranges. - If you lack the repo or test output, say what you still need instead of guessing. - Do not claim this matches any external benchmark score.