sdlc-retro

# sdlc-retro The **Retrospective (Learn)** phase executor — LoopSmith's backward-learning loop, so a goal doesn't just ship, it *teaches*. Runs at the end of a goal, **after Review**. It is **advisory**: it proposes, writes only the audit-trail note freely, and **parks** any north-star or standing-rule change for your approval — it never rewrites your standing docs unattended (mirrors how `sdlc-kg`'s `maintain` only proposes). ## Executor resolution (host-aware) No `superpowers` / `code-review` companion covers retrospective learning, so this is **always LoopSmith's own**. It *complements* the Review phase's `superpowers:verification-before-completion` (which asks "is it correct / done?") — it does not duplicate it (this asks "what did we learn, what debt remains?"). Pure markdown discipline + `git` / `python3`, so it degrades gracefully on any host. ## Pre-flight — gather the evidence (repo-auto-detect, fail-open) Read what this specific repo offers; skip whatever's absent (never break the run): - **The original intent** — the goal's own text: the `.sdlc/goals/NNNN-*.md` file (local mode) or the issue body (`gh issue view "$goal"`, github mode). - **What shipped** — the diff for this goal's work (`git diff` / `git log` over its branch or commits). - **The journey** — `.sdlc/journey/<goal>.md` (local) or the issue timeline (github): the phase notes and 🔒 Critical Insights recorded as the goal ran. - **Standing context** — the repo `README`, `.sdlc/context/north-star.m...