Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
직접 명령은 검토 Prompt를 거치지 않습니다. 실행하기 전에 소스를 확인하세요.
npx skills add https://github.com/dartsim/dart --skill dart-review-pr명령은 한 줄로 유지됩니다. 복사하기 전에 가로로 스크롤해 전체 내용을 확인하세요.
로컬 사본을 원하시나요? SkillsMP에서 현재 제공할 수 있는 파일을 다운로드하세요.
SOC 직업 분류 기준
SKILL.md 표시 중
| name | dart-review-pr |
| description | DART Review PR: review a PR or address review feedback |
Use this skill in Codex to run the DART dart-review-pr workflow. The editable
workflow source lives in .claude/commands/; this file is its generated adapter
in the shared .agents/skills/ catalog.
/dart-review-pr <arguments>$dart-review-pr <arguments>Treat the text after the skill name as $ARGUMENTS. When the workflow
references $1, $2, etc., map those to the positional values supplied by the
user.
Review or respond to PR: $ARGUMENTS
@AGENTS.md @docs/onboarding/code-style.md @docs/onboarding/ai-reviews.md @docs/onboarding/ai-tools.md @docs/ai/verification.md
Pick the sub-workflow from mode= in $ARGUMENTS, defaulting to review.
For candidate=<id>, read its candidate.json at the path printed by
review-gate prepare. Verify the supplied base/head/tree and inspect
git diff <merge_base> <head> with surrounding code. Work from the immutable
candidate in an isolated read-only checkout; do not accidentally review dirty
files or a later HEAD. No PR needs to exist. The parent supplies objective,
acceptance criteria, factual gates, prior findings, and author-session IDs.
For a PR number, obtain its current head/base and complete diff with
gh pr view and gh pr diff, then follow the same coverage policy. A PR review
without a prepared local candidate is useful feedback, not publication evidence.
Apply the assigned scope from docs/onboarding/ai-reviews.md: correctness
covers the complete PR diff and acceptance evidence; contracts independently
traces consumers, sibling cases, and negative cases and records the required
input/consumer matrix for exclusions, parsers, or validators. Challenge test
oracles against actual requirements. A non-substantive assessment must prove
unchanged behavior under the owner's strict baseline rules. Missing evidence
or unobserved effective reviewer settings makes the report incomplete.
Use a distinct non-author session for each substantive scope. Check code style,
tests, docs, and focused commits. For 3D claims,
require the dart-verify-sim text oracle plus assessed visual/debug evidence,
or a justified replacement. Report every surviving finding as a coherent batch, including
repair regressions and earlier findings whose disposition is unsupported.
Stay read-only. For a local candidate return the final JSON report defined in
docs/onboarding/ai-tools.md for the parent to import with review-gate record.
Include observed session/model/effort, coverage, completion, findings with stable
IDs and concrete evidence, and verified dispositions. Do not mutate the evidence
store yourself. A clean verdict requires complete coverage for the current stage under the
review owner; explicitly retain pending hosted acceptance checks.
Use the paginated review/CI inspection commands and the single Review-Fix Loop
Workflow in docs/onboarding/ai-reviews.md. Collect the completed batch, verify
claims, and repair the underlying defect family. That owner defines trigger
ownership, current-head completion, the two-round strategy checkpoint, false
positive dispositions, blockers, and readiness; do not restart a per-comment
fix/push/review loop here.
For published PRs, prefer a new follow-up commit so reviewers can inspect each round; amend or force-push only after explicit maintainer/user approval and only when the user requests it or a clear reason exists (removing sensitive content, repairing branch history).
Run the relevant local gates, including pixi run lint before every commit.
Merge the latest base, validate and pass the independent local review gate
before each approved push, and apply the owner's remote
divergence recovery if the head moved. Reuse existing explicit authority for
this PR, action, and scope; ask only where it is missing. No inline bot replies.
Monitor CI (gh pr checks $1); readiness and merge remain separately gated and
require approval for the corresponding external mutation.