claude-skills
CLAUDE.md1 starsChanged 4 months agoApache-2.0View on GitHub
A collection of custom Claude Code skills — prompt-driven instruction sets that Claude Code loads on demand to handle specific tasks. Skills are plain Markdown files with YAML frontmatter. Python tooling (requirements.txt, .venv) exists for validation and pre-commit hooks; there is no application build system or test suite. Each skill lives under the skills/ directory: SKILL.md must begin with YAML frontmatter: The description field is the most critical part of any skill. Claude Code reads it to decide whether to…
- Reads credentials
- Installs packages
In this repository6 other files it gives its agents.
- github-actions-security-reviewSkill
- github-doc-syncSkill
- hand-offSkill
- pr-comment-triageSkill
- skill-validatorSkill
- suno-songwriterSkill
The file
We split this CLAUDE.md into the rules it gives an agent
9 rules. About 0 are common, 4 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 43
When adding a new skill, also add a matching entry to marketplace.json and a .claude-plugin/plugin.json in the skill directory, then …
- Very specific: line 58
Create skills/<skill-name>/SKILL.md with frontmatter (name, description) and the full skill instructions.
- Very specific: line 60
If the skill references large lookup tables or tag catalogs, put them under skills/<skill-name>/references/ and instruct the skill to read …
- Very specific: line 61
Add a .claude-plugin/plugin.json in the skill directory and a matching plugin entry in .claude-plugin/marketplace.json so the skill is …
- Very specific: line 62
Run the skill validator to confirm structure before committing: skills/skill-validator/scripts/validate-skills.py
- Distinctive: line 69
Character counts: when a downstream tool has a character limit, compute counts — do not estimate.
- Distinctive: line 78
All changes require review from @Fyzel (enforced via CODEOWNERS).
- Distinctive: line 80
All commits must be GPG-signed.
- Distinctive: line 134
Run all hooks manually:
≈ “Common” is an estimate: at least 1 in 5 of the 28,029 CLAUDE.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
CLAUDE.md
This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
What This Repository Is
A collection of custom Claude Code skills — prompt-driven instruction sets that Claude Code loads on demand to handle specific tasks. Skills are plain Markdown files with YAML frontmatter. Python tooling (requirements.txt, .venv) exists for validation and pre-commit hooks; there is no application build system or test suite.
Skill File Structure
Each skill lives under the skills/ directory:
skills/
<skill-name>/
SKILL.md # The skill itself (required)
references/ # Supporting reference material (optional)
<reference>.mdSKILL.md must begin with YAML frontmatter:
---
name: <kebab-case-identifier>
description: <natural-language trigger description>
---The description field is the most critical part of any skill. Claude Code reads it to decide whether to auto-invoke the skill in a given conversation. It should:
- Cover the canonical trigger phrases ("create a song", "write lyrics")
- Cover fuzzy/adjacent triggers ("turning a poem into a song", "requesting a chorus")
- Explicitly note when to trigger even without the primary keyword ("even when the user does not say the word 'Suno'")
Plugin Marketplace
This repo doubles as a Claude Code plugin marketplace, so each skill is installable individually as a plugin.
.claude-plugin/marketplace.json(repo root) is the catalog. It lists one plugin entry per skill, each with a relativesourcepointing at the skill directory (e.g."./skills/suno-songwriter").- Each
skills/<name>/.claude-plugin/plugin.jsonis that skill's plugin manifest. Because the skill'sSKILL.mdsits at the plugin root (no nestedskills/dir, noskillsmanifest field), Claude Code auto-loads it as a single-skill plugin; the invocation name comes from theSKILL.mdfrontmattername. - Users install with
/plugin marketplace add Fyzel/claude-skillsthen/plugin install <skill-name>@claude-skills.
When adding a new skill, also add a matching entry to marketplace.json and a .claude-plugin/plugin.json in the skill directory, then validate both:
claude plugin validate . # marketplace.json
claude plugin validate ./skills/<name> --strict # the skill's plugin manifestThe .skill publishing pipeline (below) excludes .claude-plugin/ from the archive, so the plugin manifest does not leak into the Agent Skill releases.
Reference Files
Large reference files are too big to read in full. Skills that use them must instruct Claude to scan section headers first, then read only the relevant entries for the current task. When editing reference files, preserve the heading structure — it is the navigation surface.
Adding a New Skill
- Create
skills/<skill-name>/SKILL.mdwith frontmatter (name,description) and the full skill instructions. - Install the skill by placing it at
~/.claude/skills/<skill-name>/(personal, all projects) or.claude/skills/<skill-name>/(project-scoped). Skills are file-based — no settings.json registration required. - If the skill references large lookup tables or tag catalogs, put them under
skills/<skill-name>/references/and instruct the skill to read them selectively by section. - Add a
.claude-plugin/plugin.jsonin the skill directory and a matching plugin entry in.claude-plugin/marketplace.jsonso the skill is installable from the marketplace (see Plugin Marketplace). - Run the skill validator to confirm structure before committing:
skills/skill-validator/scripts/validate-skills.py
Skill Authoring Conventions
- Two-phase structure works well for creative/generation skills: an interview phase to gather the brief, then a composition/output phase. Skip the interview if the user already supplied sufficient context.
- Hard limits vs. advisory targets: distinguish between limits that cause silent failures in downstream tools (hard — non-negotiable) and optimal ranges that improve quality (advisory — state them, but don't enforce them as gates).
- Output as fenced code blocks: anything the user needs to copy-paste verbatim should be in its own fenced code block with nothing extra inside.
- Character counts: when a downstream tool has a character limit, compute counts — do not estimate.
Branch Flow and CI/CD
This repo uses a three-branch promotion pipeline: dev → test → main.
- Push to
devautomatically opens a PR promotingdev→test. - Push to
testautomatically opens a PR promotingtest→main. - Merge to
maintriggers skill packaging/publishing and wiki publishing. - All changes require review from @Fyzel (enforced via CODEOWNERS).
All commits must be GPG-signed. Configure git signing before committing:
git config commit.gpgsign trueBranch naming conventions (all cut from dev, PR back into dev):
feat/*— new skills or featuresdoc/*— documentation-only updates
Promotion PRs require two GitHub repo settings configured:
- Secret
PROMOTE_TOKEN— a PAT withreposcope (used to open PRs across branches;GITHUB_TOKENcannot open PRs targeting protected branches in some configurations). - Variable
PROMOTION_ASSIGNEE— GitHub username to assign promotion PRs to.
Skill Publishing
On every merge to main, two workflows run:
publish-skills.yml — discovers skill directories under skills/*/, zips each one into <skill-name>.skill (flat archive — no wrapping directory), and creates a GitHub Release tagged <skill-name>-v<count+1>, where count is the number of existing matching tags for that skill. Releases are idempotent: an existing tag is skipped, not an error.
The .skill archive contains exactly the skills/<skill-name>/ directory contents at the zip root (flat — no wrapping directory):
SKILL.md
references/
<reference>.mdpublish-wiki.yml — builds the GitHub repo wiki from skills. The Home.md page is a generated index table (skill name + description from frontmatter). Each skill gets its own wiki page: the SKILL.md content with frontmatter stripped, followed by any files under references/ appended verbatim. The wiki is only updated when content changes.
Python Tooling
Python tooling is managed via .venv and requirements.txt:
python -m venv .venv
pip install -r requirements.txtPackages: pylint, bandit, pre-commit, PyYAML.
Pre-commit Hooks
Pre-commit is configured with five hooks (.pre-commit-config.yaml). All run against local tools that must be installed:
- actionlint — lints GitHub Actions workflow files (requires
actionlint) - package-skills — validates all
SKILL.mdfiles on any commit touchingskills/(requires.venv) - trivy-secrets — secret detection scan of the whole repo (requires
trivy) - pylint — lints any Python files (requires
pylint) - bandit — security scan of any Python files (requires
bandit -r -ll)
Install hooks after cloning:
pre-commit installRun all hooks manually:
pre-commit run --all-filesDoes it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

