A branch of my-skills-maintenance: follow its shared rules (repo anchor
$REPO, propose-before-write, verify-then-report, GLOSSARY vocabulary, the
finishing proposal). The mandatory stray check does not apply to this
branch — it touches neither skills/ nor the store.
The update path for CODING_STANDARDS.md files already generated in other
repos. When setup-coding-standards' baseline gains a version, bring each
target file up to it: replace the baseline region above the file's marker with
the new baseline, keep everything below the marker untouched, show the diff,
then write. Creation is setup-coding-standards' job (it stops when a file
exists); this branch is that existing file's update path.
1. Collect target repos
Take the repo list from the user, or scan a directory they name (one level
deep, looking for CODING_STANDARDS.md). Repos without that file are
skipped — creating one is setup-coding-standards' job.
Done when the target list is fixed and shown to the user.
2. Classify each file by its stamp
The stamp is the marker line the file was generated with:
<!-- end of baseline v<N>; ... -->. The current version is the stamp inside
the source of truth:
grep -o 'baseline v[0-9]*' "$REPO/skills/self/setup-coding-standards/references/baseline.md" | tail -1
(Reading the version from baseline.md keeps one home for it; never hardcode
a version into this branch.)
Per file:
- No stamp — generated before stamps existed (legacy): not mechanically upgradeable. The first migration is a judgement merge — old baseline lines matched against the new sections by eye, with the user. Report it and stop for their call; never guess a merge.
- vN ≥ current — already current, skip.
- vN < current — upgrade candidate.
Done when every file has one class, shown to the user.
3. Compose the upgrade, show the diff, then write
For each candidate, the new file is exactly two parts:
baseline.md — verbatim, including its own stamp line
+ every line of the old file below its stamp line
Mechanically, in the target repo:
ML=$(grep -n 'end of baseline' CODING_STANDARDS.md | cut -d: -f1)
{ cat "$REPO/skills/self/setup-coding-standards/references/baseline.md"; \
tail -n +$((ML+1)) CODING_STANDARDS.md; } > CODING_STANDARDS.md.new
Verify before showing the user — two byte-identity checks:
# the region above the new stamp is baseline.md exactly (stamp included)
head -n $(grep -n 'end of baseline' CODING_STANDARDS.md.new | cut -d: -f1) \
CODING_STANDARDS.md.new | diff - "$REPO/skills/self/setup-coding-standards/references/baseline.md"
# the region below the new stamp equals the old file below its stamp
sed -n "$((ML+1)),\$p" CODING_STANDARDS.md | \
diff - <(sed -n "$(($(grep -n 'end of baseline' CODING_STANDARDS.md.new | cut -d: -f1)+1)),\$p" CODING_STANDARDS.md.new)
Only then show the diff (old file → .new) and wait for the user's go-ahead **per repo** before writing. These are their files; the "existing standards file is the human's to change" guardrail of setup-coding-standards` applies
here at update time. Declined candidates stay untouched.
Done when every candidate is written or explicitly declined, and each written file passes both byte-identity checks.
4. Report
List per repo: skipped (no file), current (stamp up to date), upgraded (old vN → new vM), legacy (no stamp, needs the judgement merge), declined. Committing is the user's call per repo — each commit belongs to its target repo, not this one; offer to make them if asked.