메뉴
Benchmark-gated skill evolution — consume bigpowers-benchmark report, propose plan-work change, edit skill via craft-skill, re-run benchmark, record ADR. Use when a skill underperforms on benchmark or stocktake finds systemic gap.
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
Self-review checklist for the coding agent to run before dispatching a reviewer. Checks CONVENTIONS.md compliance, Boy Scout Rule, test coverage, types, and SOLID. Produces a pass/fail checklist. Use before request-review, before committing, or when user asks for a code quality check.
Create new bigpowers skills with proper structure, progressive disclosure, and bundled resources. Use when user wants to create, write, or build a new skill for the bigpowers lifecycle.
Refine a rough idea into a clear, detailed specification through dialogue. Does not produce code. Use when user has a vague idea, wants to think through a feature before planning, or needs to turn "I want X" into a concrete spec.
Benchmark-gated skill evolution — consume bigpowers-benchmark report, propose plan-work change, edit skill via craft-skill, re-run benchmark, record ADR. Use when a skill underperforms on benchmark or stocktake finds systemic gap.
Interactive assumption-surfacing Q&A that stress-tests a plan through relentless questioning until every decision is resolved. Use when user wants to challenge a plan, validate decisions from conversation/context, or mentions "grill me". For doc-grounded variant, use grill-with-docs.
Block dangerous git commands (push, force push, reset --hard, clean, branch -D, checkout/restore .) and enforce Conventional Commits & Branch Protection before an AI agent runs them. Installs hook scripts for Claude Code, Cursor, Cursor CLI, and Gemini CLI; documents Google Antigravity Terminal deny lists. Use when the user wants git safety hooks, to block git push or destructive git in agents, or to mirror the same policy across AI coding tools.
| name | evolve-skill |
| model | opus |
| description | Benchmark-gated skill evolution — consume bigpowers-benchmark report, propose plan-work change, edit skill via craft-skill, re-run benchmark, record ADR. Use when a skill underperforms on benchmark or stocktake finds systemic gap. |
HARD GATE — No skill change ships without benchmark score ≥ pre-change baseline. Learning is measured and versioned — never implicit.
Regression gate — Run bash scripts/run-verification-gates.sh to catch mechanical regressions (compliance, sync pipeline, size budget) before spending time on benchmark evals. If golden suite fails, fix regressions first — they are pre-requisites for any capability improvement.
Establish baseline — Run run-benchmark <skill> --baseline. If no definition exists at specs/benchmarks/<skill>.yaml, create one following specs/benchmarks/SCHEMA.md first. Save report path in state.yaml. If specs/benchmarks/reports/BASELINE-<skill>.yaml already exists, skip this step.
Identify gap — Read the baseline report (specs/benchmarks/reports/BASELINE-<skill>.yaml). Find scenarios with result: FAIL or low pass_at_k. This is the measurable gap.
plan-work — Write a minimal change proposal targeting the failing scenarios. Include verify commands.
Edit via craft-skill / direct SKILL.md edit; run bash scripts/sync-skills.sh.
Re-run benchmark — run-benchmark <skill>. Compare new pass_at_k against baseline.
new pass_at_k < baseline) → revert the change and loop back to step 3.Record decision — Write specs/adr/NNNN-evolve-<skill>.md with before/after pass_at_k scores. Update session-state.
→ verify: grep -c 'run-benchmark\|pass_at_k\|BASELINE-' skills/evolve-skill/SKILL.md | awk '{if($1>=2) print "OK"; else print "FAIL"}'
See REFERENCE.md for ADR template.
## ADR-XXXX: Evolve <skill-name>
**Status:** Accepted
**Benchmark:** before X% / after Y%
**Change:** one-sentence summary
**Evidence:** path/to/benchmark-report.md
Benchmark repo: bigpowers-benchmark — resolve the local clone on this machine (e.g. gh repo clone danielvm-git/bigpowers-benchmark if absent); never assume a hard-coded absolute path.