菜单
Print a debt-ops health summary from the metrics log, covering registration rate, hook feedback action rate, ADR creation, and AI-authored share. Use when the user asks for "debt-ops metrics", "debt health", "registry stats", or invokes $metrics. Read-only, never writes the log.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Write or refresh a "Tech debt operations" section in the project's AGENTS.md so the team shares one source of truth for debt-ops disciplines. Run ONLY when the user explicitly asks to set up, install, or initialize debt-ops disciplines — never auto-invoke. Idempotent; only the managed section changes, other sections are untouched.
Register a deferred decision in the debt registry. Trigger by judgment, not a marker scan, whenever a future reader would ask "why this way?": an unmade decision, stub, loosened type, bypassed check, swallowed error, a default picked "for now", or a TODO/FIXME/HACK/XXX marker. Trigger immediately whenever you defer work, or when the user invokes /debt-ops:add. Over-register freely; the developer drops with "drop A", "drop A,C", or "drop all".
Write or refresh the `## Tech debt operations` section in CLAUDE.md so a team shares one source of truth for debt-ops disciplines and cached quality commands. Idempotent. Only the managed section changes; other sections are untouched. Invoked explicitly via /debt-ops:init (solo users get the same content from the SessionStart inject).
Print a debt-ops health summary from the metrics log, covering registration rate, hook feedback action rate, ADR creation, and AI-authored share. Use when the user asks for "debt-ops metrics", "debt health", "registry stats", or invokes /debt-ops:metrics. Read-only, never writes the log.
Audit the debt registry, rank survivors by churn × Fowler quadrant, surface a top-N list, then walk paydown on user follow-up. Use when the user asks to review debt, see what to pay down, work through entries, or invokes /debt-ops:review. Stale entries drop with `drop A,B,C`.
Register a deferred decision in the tech-debt registry. Trigger by judgment, not a marker scan, whenever a future reader would ask "why this way?": an unmade decision, stub, loosened type, bypassed check, swallowed error, a default picked "for now", or a TODO/FIXME/HACK/XXX marker. Trigger immediately whenever you defer work, or when the user asks to log/register tech debt. Over-register freely; the developer drops with "drop A", "drop A,C", or "drop all".
| name | metrics |
| description | Print a debt-ops health summary from the metrics log, covering registration rate, hook feedback action rate, ADR creation, and AI-authored share. Use when the user asks for "debt-ops metrics", "debt health", "registry stats", or invokes $metrics. Read-only, never writes the log. |
Read the hidden metrics log and tell the user whether v1's tripwires are tripping.
TOPLEVEL=$(git rev-parse --show-toplevel)
REPO_HASH=$(printf '%s' "$TOPLEVEL" | shasum | cut -c1-12)
# Hooks and skills share one deterministic cache base (ADR 0011). Override the
# base with DEBT_OPS_CACHE; default is ~/.cache/debt-ops.
CACHE_DIR="${DEBT_OPS_CACHE:-$HOME/.cache/debt-ops}/cache/$REPO_HASH"
LOG="$CACHE_DIR/metrics.jsonl"
if [ -f "$LOG" ]; then
tail -n 500 "$LOG"
else
echo "MISSING: no metrics.jsonl found for repo hash $REPO_HASH"
fi
If the file is missing or empty, tell the user the hooks haven't fired yet in this repo and stop.
One JSON object per line, three event shapes:
{"event":"edit","file":"...","registry_count":N,"ts":"..."} — every agent edit{"event":"feedback","file":"...","result":"pass|fail","ts":"..."} — every quality-check fire{"event":"session","registry_count":N,"adr_count":M,"ai_authored_count":K,"ts":"..."} — start of each sessionTimestamps are ISO-8601 UTC.
Filter to the last 7 days. Then compute:
event:edit and event:session.registry_count minus first (across either edit or session events). >0 means Discipline 1 is firing.adr_count minus first (session events only).ai_authored_count / registry_count, when registry_count>0).count(result:pass) / count(event:feedback).result:fail, look at the next feedback event for the same file. Count those that flipped to pass. Divide by total fails. Below 50% means the agent isn't reliably acting on hook output — the architectural alarm bell.If there are fewer than 5 sessions in the window, say "need more data" and skip the verdict.
One screen. No padding. Use → and ↑/↓ for trends. Example shape:
debt-ops metrics — last 7 days
─────────────────────────────────
edits : 142 (8 sessions, ~18 edits/session)
registry : +3
adrs : +1
ai-authored : 50% → 60% ↑
feedback ran : 89 times
pass rate : 88%
fail → pass rate: 80% (8/10)
verdict: ok
End with one judgment line: