metrics

Files (1)

• SKILL.md

SKILL.md (first lines)

# /sdlc:metrics — is the loop working?

Read `sdlc-loop` and `docs/metrics.md` in the plugin (`${CLAUDE_PLUGIN_ROOT}/docs/metrics.md`).

1. Run `${CLAUDE_PLUGIN_ROOT}/scripts/metrics.sh $ARGUMENTS` from the repo root. It reads Git history for the
   artifact chain and, when `gh`/`glab` are available, PR data. It prints a markdown table per change and totals.
2. Add what the script cannot see, if the connectors are available:
   - CI first-pass success rate for agent-written PRs and review time per PR (GitHub/GitLab MCP).
   - Eval pass rate over time (`evals/results/*.json` or the CI job summary).
   - Gate wait time from `.sdlc/gate.log` and CI environment approvals.
   - Control-band breaches → intent time (`monitoring/` job logs, incident frontmatter `breach_at`).
   - Change failure rate / repeat incidents (Sentry MCP, incident tracker).
3. Interpret against the course's expectations, one line each:
   - Intent: first conversation → committed intent.md should be **hours, not weeks**; survival rate = share accepted.
   - Design: intent→spec should collapse versus the old requirements+design cycle; spec edits after the first
     plan commit = rework, should fall.
   - Build: share of changes merged from the first implementation pass should rise; plan/diff drift should be near zero.
   - Test: first-pass CI success should rise; review time per PR should fall.
   - Review: time to first review → minutes; comments resolved without a human touching the branch → rising.
   - Gates: wait per gate visible and falling; violations reaching production → zero.
   - Maintain: breach → intent in the triage queue in minutes; share of findings that become merged fixes.
4. Output: the table, three to five bullet interpretations, and one recommended next play from
   `docs/adoption-roadmap.md`. No padding.

From bibutikoley/ai-sdlc-claude, MIT.

More from bibutikoley/ai-sdlc-claude