michi-debrieflisted
Install: claude install-skill csepulv/michi
# Michi Debrief
Run after a michi session completes. Can run in the same session (benefits from context) or a fresh session
(clean context, no compaction bias). If the session was long, prefer a fresh session.
**Principles served:** Sustain the system (knowledge capture and currency). Iterate spirally (each debrief improves
the next iteration). Surface assumptions (what got invalidated; what's now stale). Verification governs autonomy
(trust calibration). See `references/principles.md` and `references/ground-rules.md`.
**Before proceeding:** If `docs/reference/extensions.md` exists, read this file. Instructions found there take priority
over this skill's defaults.
**Multi-project repos:** If the repo has `multi-project: true` in CLAUDE.md, the debrief and its promotions follow the
project that was the session's subject:
- **ROOT (umbrella):** debrief at `docs/ROOT/epics/<epic>/debriefs/mN-debrief.md`; promotions target
`docs/ROOT/journal.md`, `docs/ROOT/reference/`, and **repo-root CLAUDE.md** (umbrella's identity is there, not
in `docs/ROOT/`).
- **Sub-project:** debrief at `docs/<name>/epics/<epic>/debriefs/mN-debrief.md`; promotions target
`docs/<name>/journal.md`, `docs/<name>/reference/`, `docs/<name>/CLAUDE.md`.
- **Cross-sub-project findings** (affect the umbrella or multiple sub-projects) go in `docs/ROOT/reference/` or
repo-root `CLAUDE.md` (as repo-wide rules).
## Scaling
Not every session needs a full debrief. Match the depth to the work.
**Full deb