ce-retrolisted
Install: claude install-skill relusion/vg-sdlc-claude-plugins
# Retro
**Invocation input:** Plan to report on: $ARGUMENTS
Aggregate what the pipeline already recorded into a read-only retrospective. This
is the toolset's **observability** surface: every other skill produces rich
per-run artifacts (specs, `verification.md`, `code-review.md`, auto-build run
reports) and appends to a lightweight metrics stream; `retro` reads them across
runs and reports descriptive signals. It **mutates no plan artifact** and renders
no verdict — the numbers are signals over the model's own (fallible) outputs, not
ground truth.
## Runtime Inputs
- **Plan slug (optional):** which plan to report on. If absent, read
`docs/plans/plans.json` and ask (or report across all plans).
- **Loaded (read-only):** `docs/plans/<slug>/.metrics.jsonl` (if present), each
`specs/<id>/ce-spec.md` (TC tags), `specs/<id>/tasks.json` (task counts),
`specs/<id>/verification.md`, `code-review.md` + each `specs/<id>/review-summary.json`
(review findings by lens × severity × confidence + the per-feature `suppressed`
count) + `review-learnings.md` (dismissal `RL-N` entries — for the
recurring-dismissal / promote-to-`review-policy.md` signal), `plan.json` and/or
`feature-plan.md` (Final Complexity per feature — for the complexity-drift
signal), `docs/plans/express-log.jsonl` (accepted express-patch activity, when
reporting repo-wide), and `auto-build/<date>-run.md`.
## Execution Contract
0. **Session write lease (structural, first act).** `python3 "${CLAUDE_SKI