Index of every page on this site: llms.txt
All skills
sandharbor avatar

/update-skills

@62fef7a
Run it yourself

Verify and update agent skills so their system descriptions stay accurate as the codebase evolves

  • 1 file
  • 4.1 KB
  • Updated 2 months ago
  • GitHub
Use this Skill: https://skilld.dev/gh/sandharbor/meadow/update-skills

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈28 tokens for metadata: the name and description. ≈1k when used: this file.

How this Skill uses context

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

This Skill turns off automatic use in Claude Code. Its description stays out of the automatic Skill list by default.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Update Skills

Skills encode two kinds of knowledge: why (purpose, goals, workflow steps) and how (file paths, command names, architecture details, code patterns). The "why" is stable, but the "how" drifts as the codebase evolves. This skill systematically verifies each skill's factual claims and updates them.

Step 1: Enumerate all skills

List every skill directory under .agents/skills/ (excluding this one):

ls -d .agents/skills/*/

Read each SKILL.md file to build the full list.

Step 2: For each skill, verify factual claims

Work through the skills one at a time. For each skill:

2a. Identify verifiable claims

Read the skill file and extract every factual claim about the system. These fall into several categories:

  • File paths — e.g. app/acceptance/e2e/tests/*.spec.ts, app/hosts/desktop/package.json
  • Directory structures — e.g. ASCII trees showing project layout
  • Command names and flags — e.g. ./app/acceptance/e2e/_module/scripts/slowcheck, ./app/acceptance/e2e/dev-e2e
  • Script behavior descriptions — e.g. "starts MinIO in Docker", "allocates free ports"
  • Code patterns — e.g. fixture APIs like checkpoint(message), function signatures
  • Architecture details — e.g. port numbers, API routes, file formats
  • Configuration — e.g. backing-store table names, Docker container names

2b. Check each claim against reality

For each claim:

  1. File/directory paths: Use Glob to confirm they exist. Check for files that were added, renamed, or removed.
  2. Commands and scripts: Read the script to confirm it still works as described. Check flags, arguments, and behavior.
  3. Code patterns: Read the referenced source files to confirm APIs, function signatures, and usage patterns still match.
  4. Architecture details: Read config files, server code, and test infrastructure to verify port numbers, route paths, table names, etc.
  5. Directory trees: Compare the claimed tree against what actually exists via Glob and ls.

2c. Classify each finding

For each discrepancy found, classify it:

  • Stale path: file was renamed or moved
  • Missing file: file was deleted
  • New file: relevant file exists but isn't mentioned
  • Changed behavior: script or code works differently than described
  • Changed API: function signature, route, or interface changed
  • Changed config: port, table name, container name, etc. changed

Step 3: Report findings before editing

Before making any changes, present a summary to the user for each skill:

Skill: <name>

  • [list of discrepancies found, or "No issues found"]

Ask the user to confirm before proceeding with edits. This prevents well-intentioned but unwanted changes.

Step 4: Update the skill files

For each confirmed discrepancy:

  1. Preserve the "why" — do not change the skill's purpose, workflow steps, or decision logic. Only update factual descriptions of the system.
  2. Update file paths to reflect renames/moves.
  3. Add new files to directory trees and key-file lists where relevant.
  4. Remove references to deleted files.
  5. Update behavior descriptions to match current script/code behavior.
  6. Update code examples to match current APIs.
  7. Update config values (ports, table names, routes) to match reality.

Use the Edit tool for surgical updates — do not rewrite entire skill files.

Step 5: Verify the updates

After editing, re-read each modified skill file and do a quick spot-check that the updates are accurate and that nothing was accidentally broken.

Then run:

./quickcheck

This confirms the skill file edits didn't break anything (skill files themselves aren't validated by check, but this ensures you didn't accidentally modify non-skill files).

Step 6: Summary

Print a final summary:

  • How many skills were checked
  • How many had discrepancies
  • What was updated (brief list per skill)
  • Any claims that couldn't be verified and may need manual review

Source: SKILL.md on GitHub

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No third-party reports yet.

Signed by skilld at 62fef7a. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub 3 weeks ago.

Activeupdated 2 months ago
disable-model-invocation
true

README badge

README badge for sandharbor/meadow/update-skills