SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsask-review

ask-review

General & other

View repository

Updated 10-16 04:32

About this skill

Request Code Review

Hand off your own queued work as a reviewable artifact. Default mode is a real GitHub PR on the current branch with a domain-aware self-review body. Markdown-only mode (no PR, just the text) fires only on explicit request.

When to use this skill

Trigger on any of these user phrases:

  • "open a PR / pull request for this"
  • "hand this off for review"
  • "this is ready for review"
  • "write me a self-review"
  • "clean up my commits and open a PR"
  • "give me a review doc / no PR, just the markdown"
  • "summarize this branch for the reviewer"
  • "request review on this branch"

Do NOT use this skill when:

  • The user wants to review someone else's PR for merge readiness → use do-review instead.
  • The user wants to triage feedback already received on a PR (comments, bot review, multi-reviewer streams) → use evaluate-code-review instead.
  • The repo has unrelated dirty mess across multiple branches/worktrees with no imminent review → use run-repo-cleanup.
  • The user only wants a single conventional commit with no review handoff → use the commit workflow directly.

The two output modes

Mode Trigger phrase What this skill produces
PR handoff (default) "open a PR", "ready for review", "hand this off", silence on format Clean branch pushed to origin, gh pr create executed, self-review body posted
Markdown review doc "just the text", "no PR", "give me a review doc", "write the review as markdown" Markdown file or inline output only — no git writes, no PR

If the user's phrasing is ambiguous, ask once. Default is not a license to skip the ask when intent is genuinely unclear.

Non-negotiable rules (load-bearing — do not skip)

  1. Commit dirty changes BEFORE opening the PR. Not after. A PR against a dirty tree is a broken handoff — the reviewer cannot tell what you meant to ship. Counters to every excuse live in references/rationalizations.md.
  2. Default mode is PR creation. Markdown-only mode fires only when the user explicitly asks for "review doc only" / "no PR" / "just the text". Do not infer markdown mode from vagueness.
  3. Never touch main / master / default branch directly. If the current branch is the default branch, create a new branch before committing anything.
  4. Review text is a self-review, not a changelog. Explain every change with a rationale, surface ≥2 weaknesses the author already knows about, ask ≥1 explicit question on something uncertain.
  5. PR body stays under 50,000 characters. GitHub's hard limit is 65,536; 50k leaves room for reviewer edits and quoted replies. If approaching it, split the PR.
  6. Domain drives the review lens. Backend, UI, and content-markdown PRs need different reviewer prompts and different self-review angles. Pick the domain before drafting.
  7. Multi-domain diffs fan out. When the diff spans 2+ domains with ≥5% lines each, dispatch one subagent per domain and combine outputs.
  8. Pass gh pr create --repo <owner>/<repo> explicitly every time. gh defaults are wrong often enough to bite.

Required workflow

1. Classify the mode

Read the user's request for an explicit markdown-only signal. If absent, default to PR handoff. State the mode out loud before proceeding — once declared, do not swap mid-workflow without re-stating.

2. Inspect current state (run in parallel)

git status --short
git branch --show-current
git log --oneline origin/main..HEAD 2>/dev/null || git log --oneline -5
git remote -v
gh pr list --repo <owner>/<repo> --head $(git branch --show-current) --json number,url

Capture: dirty? branch name? on default branch? unpushed count? PR already exists for this branch? origin URL (fork vs. upstream risk).

3. Handle dirty tree before anything else

If git status --short is empty, jump to Step 4.

If dirty:

  • If run-repo-cleanup is installed (~/.claude/skills/run-repo-cleanup/ or ~/.agents/skills/run-repo-cleanup/), invoke its diff-walk discipline for Phase 1 (dirty tree → conventional commits). Return here when done.
  • Otherwise, follow the inline tidy in references/handoff-mechanics.md — diff-walk the changes, group by domain, one conventional commit per concern. No git commit -am. No blind git add -A.
  • If the current branch is main / master / default, create a branch first (git switch -c <type>/<scope>-<slug>), then commit. Never commit dirty changes to the default branch.

Before moving on: git status --short must be empty AND the branch must be pushed to a remote (git push -u origin <branch> if no upstream).

4. Classify the diff's domain

Run git diff origin/main...HEAD --stat (or git diff --stat <base>..HEAD). Match the changed paths against this table:

Signal in path or diff Domain Reference
server/, api/, SQL migrations, *.controller.ts, routes/, auth/session, infra .tf/.yaml backend references/domains/backend.md
*.ts/*.tsx/*.js/*.jsx outside server/, hooks, client-side fetch, React/Vue/Svelte components frontend-ts-js references/domains/frontend-ts-js.md
server.ts with Server/Tool from @modelcontextprotocol/sdk, Zod tool schemas, SKILL.md authoring mcp-server references/domains/mcp-server.md
CSS, Tailwind classes, design tokens, a11y attributes, layout/spacing, component visuals ui-engineering references/domains/ui-engineering.md
bin/, scripts/*.sh, argparse/commander entrypoints, src/main.rs, flag parsing, exit codes cli-tool references/domains/cli-tool.md
*.md under content/, posts/, docs/, blog/docs prose, README bodies content-markdown references/domains/content-markdown.md
None of the above, or a clear mix that does not split cleanly generic references/domains/generic.md

5. Fan out if multi-domain

If the diff touches 2+ domains AND each domain has ≥5% of the changed lines, do not write a single review body alone:

  • For each domain cluster, dispatch a subagent via the Task tool. Brief each with BASE_SHA + HEAD_SHA, restrict it to the cluster's paths, and hand it the matching domain reference.
  • Each subagent returns: per-domain change summary, weaknesses to flag, 1-3 reviewer questions.
  • Combine outputs into one coherent body — one section per domain under an ## Areas heading, plus an overall summary.

Per-subagent prompt template and combination rules live in references/subagent-dispatch.md. Single-domain diffs skip this step.

6. Draft the review body

Follow the structure in references/review-text-template.md:

# <title matching commit style>

## Summary (2-4 sentences)
## Context / Why now
## The N items (per-item: files, rationale, verification)
## Files touched (aggregate table)
## Verification (what I ran, what I did NOT)
## Weaknesses and open questions
## Request to the reviewer
## Follow-ups (not in scope)

Hard requirements: under 50,000 chars; explain every change; surface ≥2 weaknesses the author knows about; ask ≥1 explicit question on something uncertain (this is the "please explain in depth" hook).

7. Ship it

PR handoff mode:

gh pr create \
  --repo <owner>/<repo> \           # Always explicit — gh defaults bite
  --base <target-branch> \
  --head <current-branch> \
  --title "<type>(<scope>): <subject>" \
  --body-file /tmp/pr-body.md

Verify immediately:

gh pr view <N> --repo <owner>/<repo> --json url,baseRefName,headRefName

URL must point to the intended repo. Base must be the intended target. If wrong, close the PR immediately and reopen on the correct target.

Markdown mode: write the review body to the path the user asked for (or inline it in the response). Do not push, do not run gh pr create, do not commit.

Rationalizations to counter (RED baseline)

The "commit dirty changes before the PR" rule is the bypass point. Agents under pressure will say:

Rationalization Counter
"The PR is urgent; I'll tidy commits in PR comments later." PR comments are not commits. Reviewers diff by commit. A broken commit history is a broken review.
"I'll squash on merge anyway." Squash-on-merge hides mistakes during review, not after. Reviewers need clean history now.
"Just this once — the changes are small." Rules don't bend for small. The small exception trains the large one.
"The working tree has WIP I don't want in the PR." Stash it, or move it to a second branch. Do not open the PR on a dirty tree hoping nobody notices.
"I'll open as Draft and fix it up." Draft PRs still get CI runs, bot reviews, and human skim. Open them clean or not at all.

Full table with verbatim agent capture lines and counters: references/rationalizations.md.

Output contract

Unless the user wants a different format, produce these artifacts in order — each at the step that produces it, not batched at the end:

  1. Mode declaration (after Step 1) — "PR handoff" or "markdown review doc".
  2. State inspection summary (after Step 2) — dirty? branch? unpushed? existing PR?
  3. Dirty-tree handling report (after Step 3) — what was committed and how, or "tree was already clean".
  4. Domain classification (after Step 4) — which domain reference(s) are in play.
  5. Fan-out decision (after Step 5) — single-domain or which subagents were dispatched.
  6. Review body draft (after Step 6) — visible before it is pushed anywhere.
  7. PR URL + verification output (after Step 7, PR mode), or markdown file path (markdown mode).

Do this, not that

Do this Not that
Check dirty tree first, commit, push, then open PR gh pr create against a dirty tree
Name the domain before drafting the body Generic body that ignores what the diff actually is
Fan out one subagent per domain when multi-domain Monolithic body that blurs domain concerns
Pass BASE_SHA + HEAD_SHA + diff to subagents, not session history Hand the subagent the full conversation context
Explain every change in the body "Various improvements" / "Refactoring" as the summary
Flag ≥2 weaknesses + ≥1 explicit reviewer question "LGTM, please merge" with no critique of your own work
Pass gh pr create --repo explicit every time Rely on gh defaults for the target repo
Push back with evidence when the reviewer is wrong Capitulate to clear the review and move on
Stay under 50,000 chars; split if larger Single 60k-char monolith

Guardrails and recovery

Do not: open a PR on a dirty tree; commit to default branch; assume gh defaults to the right repo; skip the dirty-tree handoff because "the PR is urgent"; collapse a multi-domain diff into a single voice; claim tests passed when only type-check ran; write in "Thanks for reviewing!" voice.

Recovery moves:

  • PR opened on the wrong repo — close immediately, reopen on the intended target. Always pass --repo.
  • Dirty tree committed to default branch accidentally — do not force-push. Create a branch pointing to current commit, reset default back, continue on the branch.
  • Subagent fan-out returned conflicting recommendations — surface the conflict in the body under "Weaknesses and open questions"; let the reviewer decide.
  • Body exceeds 50k chars — split the PR. Do not truncate.

Reference routing

Domain references (read the one that matches the diff):

  • references/domains/backend.md — server routes, API handlers, SQL, auth, infra
  • references/domains/frontend-ts-js.md — client-side TS/JS, state, data fetching, framework components
  • references/domains/mcp-server.md — @modelcontextprotocol/sdk imports, tool/resource registration, SKILL.md authoring
  • references/domains/ui-engineering.md — visual changes, CSS, design tokens, a11y, layout
  • references/domains/cli-tool.md — command entrypoints, flag parsing, help text, exit codes
  • references/domains/content-markdown.md — blog posts, docs, SKILL.md bodies, README prose
  • references/domains/generic.md — anything that does not fit above

Cross-cutting references (read by step):

  • references/handoff-mechanics.md — diff-walk commits, branching off default, pushing, opening the PR via gh (Step 3 + Step 7)
  • references/review-text-template.md — PR body structure, severity sections, weakness language (Step 6)
  • references/subagent-dispatch.md — per-domain subagent prompt template and output combination (Step 5)
  • references/rationalizations.md — counters when tempted to skip commits, push to default, or shortcut the body

Final checks

Before declaring done, confirm:

  • Mode declared and matches the user's request
  • git status --short empty on the author's branch
  • Branch pushed and tracks origin
  • PR URL returned (PR mode) or markdown path returned (doc mode)
  • Body ≤ 50,000 characters
  • Every change in the diff has an entry in the body
  • ≥2 weaknesses surfaced
  • ≥1 explicit reviewer question
  • Domain reference actually matches the diff
  • Multi-domain diffs fanned out to subagents (or explicitly justified single-voice)
  • gh pr view confirms PR is on the intended repo and base branch (PR mode)