Skill Refresh

Process (fully autonomous)

Step 1: Detect context

If argument is a platform name (n8n, ghl, azure), use it directly. If no argument, scan todos/ for pending release monitor todos and auto-detect platform. If a todo exists, read it for: version, matched keywords, release notes link.

Step 2: Fetch latest docs

n8n: Try Context7 first (/n8n-io/n8n-docs), fall back to web search on docs.n8n.io. GHL: Web search on marketplace.gohighlevel.com/docs/ for the matched keywords. Azure: Web search on learn.microsoft.com + azure.microsoft.com/updates for matched keywords.

Focus queries on the keywords from the todo (e.g., "breaking", "deprecated", "task runner"). If no todo, query the 3 key validation areas per platform:

Step 3: Extract claims from skills

Read each target skill's SKILL.md + all reference files. Extract testable claims (version numbers, API endpoints, auth patterns, syntax rules, behavioral statements). Build a checklist: claim → source line → current value → doc value → match?

Step 4: Identify gaps

Compare extracted claims against fetched docs. Categorize each gap:

• CRITICAL: Wrong information that would cause errors (wrong API, wrong auth method, wrong syntax)
• IMPORTANT: Missing new feature or deprecated feature not flagged
• MINOR: Outdated version number that still works but isn't latest

Step 5: Apply fixes (autonomous)

For each gap found:

1. Read the file containing the stale claim
2. Edit with the corrected information
3. Add source URL as a comment or reference where applicable

Do NOT ask for approval at this stage. Apply all fixes.

Step 6: Present summary and ask for commit

Present a single summary table:

## Skill Refresh Summary — [platform] [date]

### Changes applied:

| File | Line | Before | After | Severity | Source |
|------|------|--------|-------|----------|--------|
| ... | ... | ... | ... | CRITICAL | [url] |

### Files modified: [count]
### Claims verified: [total] — [pass] OK, [fail] corrected, [skip] unchanged

Commit these changes? (y/n)

Wait for user approval. If approved:

• git add only the modified skill files
• Commit with message: fix(<platform>): skill refresh — [count] claims corrected
• If a todo triggered this, mark the todo as completed

Step 7: Update todo (if applicable)

If triggered by a release monitor todo:

• Mark todo status as completed
• Add completion note with date and count of changes

Key rules

• Never ask questions during steps 1-5. The skill is fully autonomous until the commit approval.
• Never skip the summary. Even if zero gaps found, show the verification count.
• Always show before/after for every change so the user can review.
• Prefer Context7 over web search — more structured, less noise.
• Do not modify files outside the skill directories — this skill only touches .claude/skills/ and todos/.
• If a claim cannot be verified (doc source unavailable), mark it as UNVERIFIED in the summary, don't guess.