claude-skills
CLAUDE.md1 starsChanged 43 days agoMITView on GitHub
This repository contains 327 custom slash commands (skills) for Claude Code organized in skills/ by category. Each skill is a markdown file with YAML frontmatter: external/ holds skills authored in other repos, in the Agent Skill format — a directory with SKILL.md plus optional references/ and workflows/. Currently 151 skills from 20 upstream repos (~39 MB), plus private ones in sources.local.txt. They are a different thing from the slash commands in skills/: Vendored copies are kept in sync with scripts/sync-external.sh:
In this repository3 other files it gives its agents.
- site/AGENTS.mdAGENTS.md
- site/CLAUDE.mdCLAUDE.md
- templateSkill
The file
We split this CLAUDE.md into the rules it gives an agent
7 rules. About 0 are common, 2 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
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 89
Add new sources as a line in external/sources.txt, then sync.
- Very specific: line 90
Keep each vendored LICENSE and UPSTREAM.md (provenance + pinned commit).
- Very specific: line 93
No LICENSE at all, or (A)GPL → external/sources.local.txt, vendored to external/.local/<name>/.
- Distinctive: line 109
Keep skills focused: one task per skill, clear steps
- Very specific: line 110
Every skill must have a description in frontmatter
- Very specific: line 111
Use $ARGUMENTS for user input
- Distinctive: line 113
Never hardcode tool-specific assumptions — skills should work across tech stacks
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 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 Skills Collection
This repository contains 327 custom slash commands (skills) for Claude Code organized in skills/ by category.
Project Structure
skills/ # All 327 skills organized by category
├── ai/ # 18 AI/LLM skills (agents, RAG, evals, guardrails, MCP)
├── api/ # 5 API development skills
├── accessibility/ # 2 accessibility skills
├── automation/ # 10 automation & integration skills
├── cloud/ # 10 cloud architecture skills (AWS, GCP, Azure)
├── code-quality/ # 10 code quality & review skills
├── content/ # 8 content & copywriting skills
├── data/ # 12 data & analytics skills
├── database/ # 6 database skills
├── debugging/ # 15 debugging & troubleshooting skills
├── devops/ # 12 devops & infrastructure skills
├── docs/ # 10 documentation skills
├── ecommerce/ # 8 e-commerce skills
├── finance/ # 10 finance & SaaS metrics skills
├── fullstack/ # 12 full-stack feature skills
├── git/ # 12 git & version control skills
├── i18n/ # 1 internationalization skill
├── legal/ # 8 legal & compliance skills
├── marketing/ # 2 marketing skills
├── meta/ # 4 meta-skills (router, pipelines, forge, health)
├── ml/ # 8 machine learning skills
├── mobile/ # 10 mobile development skills
├── networking/ # 19 networking & Cloudflare skills
├── observability/ # 8 observability & SRE skills
├── performance/ # 7 performance skills
├── product/ # 12 product management skills
├── sales/ # 21 sales & growth skills (lead gen, outreach, deals)
├── scaffold/ # 10 scaffolding skills
├── security/ # 11 security skills
├── testing/ # 13 testing skills
├── utils/ # 15 utility skills
└── web/ # 18 web & UI skills
pipelines/ # Composable multi-skill workflows (*.yaml)
external/ # Third-party Agent Skills vendored from upstream repos
├── sources.txt # Upstream manifest (name|repo|ref|subpath)
└── <skill>/ # Vendored copy: SKILL.md + references/ + workflows/
install.sh # Interactive installer (macOS/Linux/Git Bash)
install.ps1 # Windows PowerShell installerSkill Format
Each skill is a markdown file with YAML frontmatter:
---
description: One-line description shown in command palette
---
Prompt instructions for Claude.
$ARGUMENTS ← replaced with user inputExternal Skills (third-party)
external/ holds skills authored in other repos, in the Agent Skill format — a
directory with SKILL.md plus optional references/ and workflows/. Currently 151
skills from 20 upstream repos (~39 MB), plus private ones in sources.local.txt. They are a different thing from the slash
commands in skills/:
skills/ | external/ | |
| Format | one .md per skill, $ARGUMENTS | directory with SKILL.md |
| Installs to | ~/.claude/commands/ | ~/.claude/skills/ |
| Invoked as | /category--name | /name, or auto-triggered by description |
| Authored | here | upstream |
Vendored copies are kept in sync with scripts/sync-external.sh:
./scripts/sync-external.sh --check # which sources are behind upstream
./scripts/sync-external.sh # pull all sources forward (both manifests)
./scripts/sync-external.sh <name> # pull one source forward
./scripts/sync-external.sh --list # manifest entries + vendored commitsRules when touching external/:
- Never hand-edit a vendored directory — the next sync overwrites it. Fork upstream and repoint
external/sources.txtinstead. - Add new sources as a line in
external/sources.txt, then sync; don't copy files in by hand. - Keep each vendored
LICENSEandUPSTREAM.md(provenance + pinned commit). - Licensing decides which manifest an entry goes in, not preference:
- MIT / Apache-2.0 →
external/sources.txt, vendored toexternal/<name>/, committed and published. - No LICENSE at all, or (A)GPL →
external/sources.local.txt, vendored toexternal/.local/<name>/. Both are gitignored: usable locally, never republished. Private use is not distribution. - Moving an entry from the local manifest to the public one publishes it. Only do that if upstream's license changes to permit it.
sync-external.shwarns on a published entry with no license, and CI fails on it.
- MIT / Apache-2.0 →
- Installed names match upstream. Rename only for cause (e.g. a directory named
skill), and record why in the manifest comment. scripts/test-runner.shdeliberately ignoresexternal/— those files follow upstream's conventions, not this repo's..github/workflows/external-skills.ymlvalidates it instead.
See external/README.md for the full workflow.
Installation
Skills are installed by copying .md files into ~/.claude/commands/ (global) or .claude/commands/ (project-level). The installer renames files to category--skill-name.md format for flat directory compatibility.
External skills are installed alongside them as directories under ~/.claude/skills/ (global) or .claude/skills/ (project-level), keeping their upstream name.
When working on this repo
- Keep skills focused: one task per skill, clear steps
- Every skill must have a
descriptionin frontmatter - Use
$ARGUMENTSfor user input - Skills should detect project context (language, framework) automatically
- Never hardcode tool-specific assumptions — skills should work across tech stacks
Does 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.

