ClawHub PR Maintainer

Use this skill for maintainer-facing ClawHub GitHub workflow, not for ordinary implementation work.

Start With Live GitHub State

• Use gh pr view or gh issue view against openclaw/clawhub; verify live state before commenting, labeling, closing, or recommending merge.
• For PRs, read title, body, author, labels, comments, files, commits, status checks, review state, and linked issues.
• Surface author identity briefly: GitHub name/login and account age when useful. Treat identity as triage signal, never as proof by itself.

Common read-only commands:

gh pr view <number> --repo openclaw/clawhub --json title,body,author,labels,comments,files,commits,statusCheckRollup,reviewDecision,url,additions,deletions,changedFiles
gh issue view <number> --repo openclaw/clawhub --json title,body,author,labels,comments,state,url
gh api users/<login> --jq '{login,name,created_at,type}'

Review Evidence Bar

• For bug fixes, require symptom evidence, a plausible root cause in the touched code path, and either a regression test or focused manual proof.
• For UI changes, require screenshots or video when the behavior is meaningfully visual. Use tests as supplemental evidence, not a substitute for visible proof.
• Do not merge or recommend merge based only on PR prose, AI rationale, or green CI when the changed behavior has not been exercised.
• For contributor-provided screenshots/videos/logs, inspect the artifact directly and state what it proves. Do not rerun proof:ui just to inspect existing evidence.

Structure PR Review Output

• Start every PR review with 1-3 plain sentences explaining what the change does and why it matters.
• Show size near the top as LOC: +x/-y (N files), using live PR stats or local diff stats.
• Then list findings first. If none, say No blocking findings or No findings.
• Always answer: affected ClawHub surface, bug or behavior being changed, evidence checked, and best-fix verdict.
• For bug/regression fixes, include a compact Provenance: line when a bounded history pass identifies it. Separate code author, PR author, merger/committer, current PR author, PR number, and date when those differ. If the blamed PR was merged by automation, identify the human trigger when practical; otherwise say trigger unknown.

Read Beyond The Diff

• For code-path bug, regression, or behavior changes, review the surrounding path, not just changed lines. Open the runtime entry point, owner module, one caller, one callee, adjacent tests, and sibling surfaces that should share the invariant.
• For docs/config/process-only changes, read the changed file, its linked or adjacent source of truth, and any route/workflow/template the change claims to affect. Do not require runtime caller/callee evidence when no runtime path exists.
• Compare against current origin/main behavior or current published docs when regression, compatibility, or user-visible docs accuracy matters.
• For dependency-backed behavior, read the upstream docs/source/types before judging API use, defaults, output shapes, errors, timeouts, memory behavior, or compatibility.
• Mention the main files or contracts read when the verdict depends on code-path, docs, config, or workflow evidence.
• If a required path is uninspected, keep reading or mark Remaining uncertainty; do not call the PR best, proof-sufficient, or merge-ready.

Best-Fix Review Loop

Every PR review must explicitly answer: "Is this the best fix, or only a plausible fix?"

Before verdict:

1. Reconstruct the bug, feature need, or behavior claim from the issue, PR, and proof.
2. For code-path changes, trace current behavior from entry point to failure or decision point.
3. For docs/config/process-only changes, trace the reader/operator workflow or automation path the change is meant to clarify.
4. Read touched files, relevant callers/callees for code changes, adjacent docs or tests, owner modules, and relevant source-of-truth docs.
5. Read sibling surfaces that should share the invariant or could be broken by a one-sided fix.
6. Compare against current origin/main and shipped behavior when relevant.
7. Identify at least one alternative fix location or shape, then reject it with evidence.

Review output must include:

• Best-fix verdict: best / acceptable mitigation / wrong layer / too narrow / too broad.
• Alternatives considered: 1-3 concrete alternatives and why rejected.
• Code read: compact list of main files/contracts checked.
• Remaining uncertainty: what was not proven.

Enforce Bug-Fix Evidence

• Never merge a bug-fix PR based only on issue text, PR text, or AI rationale.
• Before recommending merge for a bug fix, require:
  1. symptom evidence such as a repro, logs, failing test, or focused manual proof
  2. a verified root cause in code with file/line
  3. blame-backed provenance for regressions when traceable, or commit SHA/date when no PR is traceable
  4. a fix that touches the implicated code path
  5. a regression test when feasible, or explicit manual verification plus a reason no test was added
• If the claim is unsubstantiated or likely wrong, request evidence or changes instead of recommending merge.

Decide UI Proof Mode

Generate new visual evidence with the best proof runtime available in the current session. Use Crabbox through bun run proof:ui only when a Crabbox skill or working Crabbox capability is available. Otherwise ignore Crabbox and run the existing Playwright proof runtime against a real local ClawHub instance; missing Crabbox access is not a blocker.

• before-after: bug fixes, regressions, changed copy, changed layout, or any PR where main-vs-candidate comparison clarifies the change.
• feature: new page, new flow, new UI state, or behavior that cannot exist on origin/main.
• No generated proof: docs-only, backend-only, tests-only, metadata-only, or already-sufficient contributor evidence.

Write a temporary Playwright scenario under .artifacts/proof-scenarios/; do not infer manual clicks. Keep screenshots and videos in .artifacts/ until publishing. Never commit proof artifacts.

For the local fallback, start ClawHub with the relevant local Convex state and run the scenario through the local Playwright runner:

bun run proof:ui -- --runner local --mode feature \
  --scenario .artifacts/proof-scenarios/<name>.pw.ts \
  --candidate-url <local-clawhub-url>

For before/after proof, run the same scenario against an origin/main checkout and the candidate checkout, then pass both URLs with --baseline-url and --candidate-url. The runner accepts only localhost or loopback URLs and writes publishable baseline/ and candidate/ artifacts. Use the Codex app browser to inspect the running local instances and captured evidence.

Final Review Comment With Proof

Use proof-video for capture, editing, inspection and publication. Capture from the real running ClawHub instance and inspect every final image/video before uploading it. Lead with the successful candidate; label any failed baseline separately and state companion fixes and simulations.

Attach inspected media directly to the PR with native GitHub attachments:

gh pr comment <number> --repo openclaw/clawhub \
  --body-file .artifacts/proof/comment.md \
  --attach .artifacts/proof/after.mp4 \
  --attach '.artifacts/proof/before.png#Before' \
  --attach '.artifacts/proof/after.png#After'

Videos use bare uploaded URLs so GitHub renders players; do not add video alt text. Include exact refs, the tested flow, validation and limitations in the comment itself. Verify the final comment and return its direct link.

Never push proof assets or generated reports to any product repository branch, including qa-artifacts. The old proof:publish helper is retired. Do not leave only local paths or links to a source-tree directory as published evidence. If attachment upload fails, retain the local files and report the precise blocker instead of claiming they are attached.

ClawSweeper

ClawSweeper is the bot control plane for automated PR/issue review once ClawHub dispatch is configured. Until then, use this skill for manual maintainer review. If ClawSweeper has posted a review, read it as evidence but verify live PR state before acting.

Commenting And Labels

• Use literal multiline comment bodies or --body-file; never pass escaped \n strings.
• For issue comments and PR comments containing backticks or shell characters, prefer a single-quoted heredoc or --body-file over inline -b bodies.
• Do not wrap issue or PR refs like #123 in backticks when you want GitHub to auto-link them.
• Keep maintainer comments short: finding, evidence, requested action, and verification path.
• Use gh pr comment --body-file; add --attach for inspected media. Follow the proof-video skill for safe updates to an existing proof comment.
• Do not close more than five issues/PRs in one action without explicit confirmation and the exact target list.