Maintain

Full Maintenance Pass

Run each phase in order. Report results after each phase. Continue even if one phase has no changes.

Phase 1: Pull local sources

If skill-maintain CLI is available:

skill-maintain sources

If CLI is not available, check for tracked repos in .skill-maintainer/config.json under tracked_repos. For each repo path that exists:

1. Record the current HEAD SHA: git -C rev-parse HEAD
2. Pull: git -C pull --ff-only
3. Compare new HEAD to old. Report CHANGED (with git -C log --oneline ..) or UPTODATE.

If no config file exists, skip this phase and note "no .skill-maintainer/config.json found -- skip source pull (run /skill-maintainer:init-maintenance to set up)".

Phase 2: Check upstream docs

If skill-maintain CLI is available:

skill-maintain upstream

If CLI is not available, check .skill-maintainer/config.json for llms_full_url and upstream_urls. Fetch the llms-full.txt URL using WebFetch, split by Source: delimiters, and compare each watched page's content against stored hashes in .skill-maintainer/state/upstream_hashes.json.

If no config file exists, skip and note "no config -- skip upstream check".

Phase 3: Quality report

If skill-maintain CLI is available:

skill-maintain quality

If CLI is not available, perform the checks manually. For every SKILL.md found in the repo:

Spec compliance

• Frontmatter has name and description fields
• name is kebab-case, matches directory name
• No disallowed frontmatter fields (allowed: name, description, license, allowed-tools, metadata, compatibility)

Token budget

• Count total chars in the skill directory (.md files only), divide by 4
• Warn if over 4,000 tokens, critical if over 8,000

Body size

• SKILL.md under 500 lines

Freshness

• metadata.last_verified present and within 30 days of today

Description quality

• Description contains a WHAT verb (handles, generates, validates, designs, checks, runs, creates, builds, manages, monitors, tracks, reports)
• Description contains a WHEN trigger phrase ("use when", "when user", "when the user", "invoke with")

Output a table with one row per skill: name, valid, tokens, lines, days since verified, description quality.

Phase 4: Review and propose updates

After all three phases:

1. Read references/best_practices.md (bundled with this plugin) or .skill-maintainer/best_practices.md (if present in the repo)
2. Review change details from Phases 1-3
3. Determine whether best_practices.md needs updates based on:

• New or changed upstream doc pages (Phase 2) that affect skill authoring rules
• New patterns or conventions from pulled repo changes (Phase 1)
• Quality report findings that suggest missing or outdated checklist items (Phase 3)

1. If updates needed: list each proposed change with rationale. Wait for user approval before writing.
2. If no updates needed: report "best_practices.md is current -- no changes needed"

Rules

• Never auto-write to best_practices.md -- always show proposed changes and wait for approval
• Run all phases even if one reports no changes
• If a phase fails, report the error and continue with remaining phases
• After finishing, summarize: repos pulled, upstream pages checked, quality issues found, best practices edits (if any)

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: fblissjr
• Source: fblissjr/fb-claude-skills
• License: Apache-2.0

Install and usage instructions live in the source repository linked above.