/update-workflows

Audits pack-test.yml — the workflow most prone to silent drift — against the current repo state and proposes fixes. Other workflows (release-publish.yml, docs-deploy.yml, docs-preview.yml) are generic and not covered here; inspect them manually if you've touched them.

Proposes all changes and waits for user approval before writing anything.

Steps

Run each step in order.

Step 1 — Gather current repo state

Run all of these in parallel:

# Vault-local skills shipped in the template
ls template/.claude/skills/

# Top-level template subdirs that pack-test `ls`-checks
ls template/
ls template/scripts/qmd/ 2>/dev/null || true

# Global skills list and bin names
grep -n 'GLOBAL_SKILLS' bin/create.js
jq -r '.bin | keys[]' package.json
jq -r '.version' package.json

Also read:

• .github/workflows/pack-test.yml — the workflow being audited
• .claude/skills/pack-test/SKILL.md — coupled checklist/counts
• A representative global skill (e.g. template/.claude/skills/brain-ingest/SKILL.md) — to verify grep_check patterns still match

Step 2 — Audit

For each item below, mark OK or DRIFT. Record every DRIFT with the exact pack-test line that's wrong and what it should be.

Vault-local skill coverage (pack-test.yml check ".claude/skills/<name>" lines):

• Every dir in template/.claude/skills/ has a matching check line.
• No check line points at a dir that no longer exists.

Global skill coverage (pack-test.yml check "~/.claude/skills/<name>/SKILL.md" + .csb-version lines):

• Every name in GLOBAL_SKILLS (from bin/create.js) has both a SKILL.md check and a .csb-version check.
• No extra names appear that aren't in GLOBAL_SKILLS.
• Corresponding grep_check / absent_check loops iterate the same set.

grep_check patterns:

• Each pattern (claude-second-brain path, claude-second-brain qmd, etc.) still appears in at least one shipped global skill under template/.claude/skills/.

absent_check patterns:

• Each pattern (__QMD_PATH__, __CSB_CONFIG__, INDEX_PATH=\$QMD_INDEX) still doesn't appear in any shipped file. (A stale absent_check for a token that no longer exists anywhere is harmless but noisy — optional to clean up.)

Bin names:

• Every bin name referenced in pack-test.yml (e.g. claude-second-brain, csb) exists as a key in package.json's bin object.

pack-test/SKILL.md coupling:

• The expected-skills comment lists the same set as template/.claude/skills/.
• The ls and grep commands in Step 3 reference current paths.
• The checklist table's "N subdirs" count matches the actual ls template/.claude/skills/ | wc -l.
• The cleanup rm -rf line in Step 4 targets the current GLOBAL_SKILLS set.

Step 3 — Propose changes

Present a consolidated proposal:

## Workflow audit proposal

### .github/workflows/pack-test.yml
- DRIFT: <specific line> — <what's wrong> → <proposed fix>
- ...

### .claude/skills/pack-test/SKILL.md
- DRIFT: <specific line/section> — <what's wrong> → <proposed fix>
- ...

If everything is OK, report "pack-test is in sync — nothing to update" and stop.

Stop here and wait for user approval before applying any edits. If the user wants to skip items, drop them from the list before continuing.

Step 4 — Apply approved edits

Use Edit on:

• .github/workflows/pack-test.yml
• .claude/skills/pack-test/SKILL.md

Do not edit other workflows, other skills, or source files. If the audit surfaced drift elsewhere (e.g. a grep_check pattern that's truly gone from every skill), flag it in the report — don't fix it here.

Step 5 — Report

Tell the user:

• Files edited and what was changed on each.
• Anything flagged but not fixed (and why).

What can go wrong

• No drift found — Step 2 produces all OK; report and stop without touching anything.
• Ambiguous fix — if the audit finds drift but multiple corrections are plausible (e.g. a grep_check pattern that could be renamed or deleted), ask the user rather than guessing.
• Other workflow changes on this branch — not in scope; mention them in the Step 5 report so the user knows to review manually.