메뉴
Coordinate a model-diverse local code review, automatically repair actionable findings with sequential workers, and re-review to a final verdict. Use for /review or when launched as an SC ensemble reviewer.
기본적으로 소스를 먼저 확인하는 Prompt가 선택됩니다. 직접 명령으로 전환하거나 로컬 사본을 다운로드할 수도 있습니다.
설치 여부를 결정하기 전에 SKILL.md와 SkillsMP에 표시된 보조 파일을 읽어 보세요.
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/HazAT/pi-config --skill review명령은 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 | review |
| description | Coordinate a model-diverse local code review, automatically repair actionable findings with sequential workers, and re-review to a final verdict. Use for /review or when launched as an SC ensemble reviewer. |
Review for correctness, security, regressions, and acceptance-criteria gaps. Delegated reviewers remain read-only. In coordinator mode, do not edit source directly; coordinate repair workers when the review requires changes.
Derive the coordinator files before continuing:
if [ -n "${PI_SESSION_FILE:-}" ]; then
SESSION_FILE="$PI_SESSION_FILE"
elif [ -n "${CLAUDE_CODE_SESSION_ID:-}" ]; then
SESSION_FILE="$HOME/.claude/projects/$(printf '%s' "${CLAUDE_PROJECT_DIR:-$PWD}" | tr '/.' '--')/${CLAUDE_CODE_SESSION_ID}.jsonl"
else
echo "Error: /review requires a persistent session file (PI_SESSION_FILE or CLAUDE_CODE_SESSION_ID)." >&2
exit 1
fi
test -f "$SESSION_FILE" || { echo >&2; 1; }
PLAN_FILE=
TODOS_FILE=
REVIEW_FILE=
SESSION_FILE. Read PLAN_FILE or TODOS_FILE when each exists; they are optional.~/.pi/agent/skills/superconductor/SKILL.md and follow its live orchestration preflight.Launch one model-diverse SC reviewer ensemble with three independent read-only Pi Chat UI sessions and unique attempt-scoped labels:
| Role | Pi model | Reasoning | Owns | Avoids |
|---|---|---|---|---|
| lead | openai-codex/gpt-5.6-sol | high | broad correctness, acceptance criteria, cross-report synthesis | style-only polish |
| grok | openrouter/x-ai/grok-4.6 | configured default | adversarial correctness, security, trust boundaries | general maintainability polish |
| kimi | openrouter/moonshotai/kimi-k3 | configured default | regressions, cross-file behavior, frontend/accessibility, focused tests | duplicating broad security review |
At preflight, verify all three exact model IDs under the enabled pi provider with sc chat providers --json. Fail clearly rather than substituting a provider or model. Current sc team run cannot select a UI or model per role, so do not use it for this ensemble. Launch each role with sc layout run tabs, --provider pi, --ui chat, its exact --model, and --reasoning high only for the lead. The explicit Chat UI is required because SC's omitted/auto UI currently resolves delegated Pi launches to terminal mode rather than the configured Pi chat experience.
Give each role:
Start all three sessions without waiting between launches. Capture every returned stable target, wait for all three, and read each target's final report. Check target/provider errors and malformed or incomplete output; do not infer success from idle state.
After collecting the grok and kimi reports, send them to the same lead target for a synthesis turn. Require the lead to deduplicate claims, preserve disagreements, and return one proposed verdict with concrete findings. Wait for and read that same lead target again; do not launch a replacement lead.
Reviewers should not all run the same broad suite. Let the kimi role run focused tests when useful; the coordinator runs one definitive verification set after collecting reports.
Independently verify the lead's synthesis against the diff and specialist reports. A finding blocks approval only when it is concrete, introduced by the review range, and either P0/P1 or a P2 required by the selected acceptance criteria. P3 never blocks approval.
Write the current attempt to the supplied review file before any repair. Set APPROVED only when no blocking finding remains.
NEEDS CHANGES automaticallyWhen blocking findings remain:
NEEDS CHANGES and stop rather than letting a worker absorb user-owned work into a commit.REVIEW-FIX-<round>-<n>.~/.pi/agent/skills/worker/SKILL.md.Run at most two repair rounds. Stop early and report NEEDS CHANGES when a worker blocks, verification fails, a safety decision requires the user, or blocking findings remain after round two. Do not hide or downgrade unresolved findings to force approval.
Always end with a normal assistant response containing:
none;Do not end after only writing or rereading the review file.
The coordinator owns every handover-file write. Do not create a captain protocol, custom state store, worktree, push, PR, merge, or cleanup action.
When the launch prompt identifies this session as an ensemble reviewer:
APPROVED or NEEDS CHANGES, followed by findings and evidence. Do not call sc team report; these model-pinned Chat UI reviewers are collected from their stable targets.# Code Review
**Verdict:** APPROVED | NEEDS CHANGES
**Session:** `/absolute/path/to/session.jsonl`
**Plan:** `/absolute/path/to/session.plan.md` | not provided
**Todos:** `/absolute/path/to/session.todos.md` | not provided
**Reviewed commits:** [base..HEAD]
**Repair rounds:** 0 | 1 | 2
## Summary
[Concise evidence-based assessment]
## Verification
- `[command]` — [result]
## Findings
### [P1] [title]
- **File:** `path/to/file:line`
- **Issue:** [specific problem]
- **Impact:** [observable consequence]
- **Suggested fix:** [concrete direction]
- **State:** open | addressed by `<sha>`
## Repair Commits
- `REVIEW-FIX-1-1` — `<sha>` — [summary and verification]
## Independent Reports
- **lead (`openai-codex/gpt-5.6-sol`, high):** [synthesized verdict and evidence]
- **grok (`openrouter/x-ai/grok-4.6`):** [verdict and evidence]
- **kimi (`openrouter/moonshotai/kimi-k3`):** [verdict and evidence]
## Residual Risks
- None | [nonblocking risk]
Omit empty findings or repair sections. Preserve concise attempt and repair evidence when replacing the file with the final verdict.