my-skills-maintenance

my-skills maintenance — one entry for the aggregation repo

The single entry for every maintenance flow of this aggregation repo (flow vocabulary per GLOSSARY: vendor sync, consolidation, patch record, distribution chain, pending update). One discoverable skill, six branches — each branch's procedure lives in a disclosed reference loaded only when that branch fires. This skill replaces consolidate-strays and my-skills-vendor-sync (retired at the map #18 migration).

Shared rules — every branch

1. Locate the repo anchor ($REPO) — never assume the working directory is the repo. Read the repo anchor ~/.agents/.my-skills-repo (one line: the aggregation repo root, written on a previous run, ADR-0009). Verify it: the directory exists, git -C "<path>" remote get-url origin ends with /my-skills (the repo's identity — a fork or a moved clone passes, any other repo fails; the scripts re-verify at runtime, so a stale anchor fails loudly, never silently), and it owns scripts/vendor-sync.mjs. Found → bind $REPO and run every command from it — node scripts/... below means "$REPO"/scripts/.... Missing or stale → ask the user for the new location (or offer to re-clone from GitHub) and write their answer back into ~/.agents/.my-skills-repo (one line).
2. Propose before you write: every git write, gh issue op, and store change is proposed to the user and waits for explicit go-ahead (AGENTS.md safety valves). The default run is read-only.
3. Verify, then report: script tests (node --test "$REPO"/scripts/*.test.mjs) and a fresh dry-run are the verification levers; git status --porcelain clean before any issue op (docs/agents/issue-tracker.md).
4. Editing skill content? Load writing-for-agents first (the reference for drafting skill content), then run the branch.
5. Vocabulary: name concepts with the GLOSSARY terms, never drifted synonyms.
6. The finishing proposal: every branch ends by proposing the same quartet — commit → push → issue close / distribution — as one proposal, and stops for the user's go-ahead. Nothing remote or destructive happens without it.

Always run first — the stray check

Every run starts with the mandatory stray dry-run (map #25) — except the standards-update branch, which touches neither skills/ nor the store:

node "$REPO/scripts/consolidate-strays.mjs"

Read the whole report (default is a read-only dry-run): every store skill classifies as current, new stray, or modified stray.

• Any stray → load references/stray-recovery.md and settle it (adopt, recover, or explicitly decline) before doing anything else — nothing proceeds past a stray, because the next distribution would clobber it.
• Zero strays → proceed to the requested branch.

Pick a branch

The request is about…	Load
resolving a pending update — sync, three-way merge patched files, bump pins	references/vendor-sync.md
consolidating or recovering stray skills	references/stray-recovery.md
recording a patch deviation	references/patch-record.md
checking freshness / reading or closing a pending-update issue	references/freshness.md
distributing to the canonical store (or retiring a skill)	references/distribute.md
rolling a baseline upgrade into consumer CODING_STANDARDS.md files	references/standards-update.md

If the request spans branches (e.g. "sync and then distribute"), run the branches in order — vendor-sync → distribute — never in parallel.

Done

When the branch's completion criterion is met, make the finishing proposal (shared rule 6) and stop. Nothing remote or destructive happens without it.