メニュー
Scan a codebase to detect technical debt (code, security, dependencies, tests, architecture). Orchestrate deterministic tools + LLM. Produce a `.debt-scan.json` file conforming to the schema.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Conventional Commits 1.0 enforcer. Two complementary modes: (1) AUTO-SUGGEST — when you say "commit", "commit my changes", "commit this", or invoke `/commit`, the skill inspects `git diff --staged`, proposes 1–3 conventional commit messages (type + optional scope + subject), and asks which one to use. You never have to remember the format. (2) VALIDATOR — every `git commit` passes through a PreToolUse hook that validates the first line of the message against the Conventional Commits regex. Non-conformant commits trigger a confirm prompt — they cannot slip through silently. Types allowed: feat, fix, refactor, perf, docs, test, chore, build, ci, style, revert. Optional scope in parentheses (lowercase kebab-case, ≤ 20 chars). Optional `!` and `BREAKING CHANGE:` footer. Use when: "commit", "commit my changes", "make a commit", "commit this", "validate commit", "/commit". Proactively invoke whenever the user says "commit" or runs `git commit` through the agent, and after a logical unit
Check for and apply AI-Native Dev Stack updates, non-destructively. Detects whether the stack repo is behind upstream, shows what changed, and fast-forwards the shared clone — without ever touching your personalized configs (CLAUDE.md, Mavis agent.md, config.sh). Use when: "upgrade the stack", "update ai-native-dev-stack", "is the stack up to date?", "get the latest rules/method". Proactively suggest at the start of a session if an update is available.
Detect architecture-level technical debt: coupling, cycles, boundary violations, layer leaks. Outputs findings conforming to the debt-finding schema.
CRUD operations on the Debt Registry. Manages the active list of accepted debts, decisions, and resolution status. Persists to the KG (Layer 0).
Generate preventive rules (linters, tests, CI checks) from detected findings. Stops the same pattern from re-appearing after a fix.
Challenge every finding and plan before validation. Reduce hallucinations, force proofs, verify completeness.
| name | debt-scan |
| description | Scan a codebase to detect technical debt (code, security, dependencies, tests, architecture). Orchestrate deterministic tools + LLM. Produce a `.debt-scan.json` file conforming to the schema. |
| license | MIT |
Run in order:
tools/scan_code.py (ruff / clippy / eslint / golangci-lint wrapper depending on language)tools/scan_security.py (trufflehog + osv-scanner + gitleaks)tools/scan_deps.py (dependency-cruiser + cargo audit / npm audit / pip-audit)Each tool produces structured JSON. No debt can be declared without going through a tool OR without explicit LLM proof.
For findings not covered by tools (architecture, semantic debt):
confidence + evidenceInvoke skills/critic/SKILL.md on the complete list.
Findings rejected by the Critic → excluded from the final report.
For each validated finding:
score = (impact × urgency × confidence) / effort_days
Compare to previous scans in .debt-history.json:
Explicitly list:
.debt-scan.json : validated findings + score + critic report.debt-history.json : updated with this scanreport.md : executive summarytools/run_all.sh : orchestrator