ai-skills
CLAUDE.md2 starsChanged 6 months agoMITView on GitHub
This repository ships AI coding assistant configurations in two delivery formats: The legacy .claude/ package was removed in v0.2.0 — plugin/ fully replaces it for Claude Code. After v0.2.0, parity is enforced between Codex and Windsurf only — Claude Code lives in plugin/ and follows its own internal organization (no longer mirrored into a sibling top-level package). Every role, skill, and guardrail in .codex/ must have a matching entry in .windsurf/. The parity matrix (review/parity-matrix.md) tracks alignment. Runtime primitives differ…
In this repository40 other files it gives its agents.
- AGENTS.mdAGENTS.md
- ai-skillsSkill
- analyze-localSkill
- analyze-prodSkill
- analyzeSkill
- architectureSkill
- asset-validationSkill
- blog-postSkill
- bugfixSkill
- cloud-platformsSkill
- code-reviewSkill
- content-creationSkill
- context-engineeringSkill
- create-prSkill
- deployment-proceduresSkill
- deploy-productionSkill
- deploy-stagingSkill
- docsSkill
- feature-devSkill
- geo-writerSkill
- humanizerSkill
- infra-changeSkill
- marketing-operationsSkill
- marketingSkill
- ml-pipelineSkill
- planSkill
- pre-commitSkill
- productSkill
- project-initSkill
- prompt-engineeringSkill
- qaSkill
- releaseSkill
- run-testsSkill
- security-scanSkill
- seo-reviewSkill
- social-media-managerSkill
- test-localSkill
- test-strategySkill
- ui-ux-designSkill
- worktree-isolationSkill
The file
We split this CLAUDE.md into the rules it gives an agent
7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 66
After editing plugin files in the same Claude Code session, reload without restarting:
- Very specific: line 72
All 31 forked + 5 main-thread orchestrator skills (= 36 user-invocable surface) appear in /help under the ai-skills: namespace.
- Distinctive: line 95
all asset contents must be in English
- Distinctive: line 97
use relative runtime paths, never absolute user paths
- Very specific: line 100
update review/parity-matrix.md for any Codex ↔ Windsurf parity-impacting change
- Distinctive: line 101
remove project-specific assumptions from shared assets
- Distinctive: line 102
prefer runtime-native representations over forced file mirroring
≈ “Common” is an estimate: at least 1 in 5 of the 27,641 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
AI Assets Repository
This repository ships AI coding assistant configurations in two delivery formats:
plugin/— full Claude Code plugin (canonical source of truth for Claude Code; current version inplugin/.claude-plugin/plugin.json)..codex/+.windsurf/+.agents/— copy-ready three-package layout for OpenAI Codex and Codeium Windsurf (.agents/ holds skills shared between Codex and Windsurf).
The legacy .claude/ package was removed in v0.2.0 — plugin/ fully replaces it for Claude Code.
Quick Reference
ai-skills/
├── plugin/ # Claude Code plugin
│ ├── .claude-plugin/
│ │ └── plugin.json # manifest + 13 userConfig knobs
│ ├── agents/ # 26 specialized agents
│ ├── skills/ # 75 skills (31 user-invocable + 5 main-thread orchestrators)
│ ├── rules/ # 12 guardrail rules
│ ├── hooks/ # 19 hooks across 14 lifecycle events
│ ├── schemas/ # 2 JSON Schemas (G7 spawn + return)
│ ├── eval/ # 47 rubrics + 282 calibration samples + Tier 1/2 runners + g1g2 attack-surface
│ ├── docs/ # 16 user docs (1 getting-started + 11 workflows + 4 concepts)
│ └── dev/validate.py # local validator (25 checks)
├── .agents/skills/ # 39 skills (shared by Codex + Windsurf)
├── .codex/ # Codex runtime package
│ ├── roles/ # 22 role definitions
│ ├── rules/ # 8 rule files + role overlays
│ ├── operations/ # hook-intents, launch-reference, settings-mapping, source-sync, skill-role-mapping
│ ├── templates/ # 14 AGENTS.md scaffolds
│ └── checklists/ # review checklists
├── .windsurf/ # Windsurf runtime package
│ ├── rules/roles/ # 22 role files (mirrors .codex/roles/)
│ ├── skills/ # 39 skills (mirrors .agents/skills/)
│ ├── workflows/ # 27 user-facing workflow files
│ ├── hooks/ # Python scripts + configs
│ └── hooks.json # Windsurf-native hook wiring
├── .claude-plugin/marketplace.json # marketplace manifest pointing at ./plugin
├── review/parity-matrix.md # cross-package alignment tracker (Codex ↔ Windsurf)
├── plugin-design/ # historical plugin design docs
├── install.sh # bash installer (Codex + Windsurf only — Linux/macOS)
└── install.ps1 # PowerShell installer (Codex + Windsurf only — Windows)Core Concepts
Two-Vendor Parity (Codex + Windsurf)
After v0.2.0, parity is enforced between Codex and Windsurf only — Claude Code lives in plugin/ and follows its own internal organization (no longer mirrored into a sibling top-level package). Every role, skill, and guardrail in .codex/ must have a matching entry in .windsurf/. The parity matrix (review/parity-matrix.md) tracks alignment.
Runtime primitives differ but capability coverage must match:
- Codex roles = Windsurf roles
.agents/skills/=.windsurf/skills/- Codex operations (documented intent) = Windsurf hooks (scripts + hooks.json)
See PARITY.md for the full parity model.
Claude Code lives in plugin/
For Claude Code, the plugin layout supersedes the old .claude/ package. To use this repo's Claude Code assets:
claude --plugin-dir /path/to/ai-skills/pluginAfter editing plugin files in the same Claude Code session, reload without restarting:
/reload-pluginsAll 31 forked + 5 main-thread orchestrator skills (= 36 user-invocable surface) appear in /help under the ai-skills: namespace.
Skill Format
Skills live in <package>/skills/<name>/SKILL.md with YAML frontmatter:
user-invocable: false— background knowledge, loaded as contextcontext: fork— user-invocable workflow (called via/<skill-name>)- Resource files sit alongside
SKILL.mdin the same directory
Hook Architecture
In the plugin (plugin/hooks/scripts/): 18 Python scripts across 13 lifecycle events, all importing the shared _lib.py. See plugin/hooks/README.md.
In the Codex/Windsurf packages: 4 carried-over Python scripts in .windsurf/hooks/scripts/ mirror the security/audit hooks. Codex declares the same intent declaratively in .codex/operations/hook-intents.md.
Exit code 2 = block, exit code 0 = allow.
Installers
install.sh and install.ps1 sync the Codex + Windsurf + shared-skills packages to ~/. Claude Code is no longer installed via these scripts (v0.2.0+) — use claude --plugin-dir instead.
Editing Rules
- all asset contents must be in English
- SKILL.md and rule files have a 12,000 character limit (skill resources have no hard limit but should stay focused)
- use relative runtime paths, never absolute user paths
- new Codex roles/skills/guardrails must be added to Windsurf simultaneously (and vice versa)
- new plugin assets land in
plugin/only — there is no longer a.claude/mirror to keep in sync - update
review/parity-matrix.mdfor any Codex ↔ Windsurf parity-impacting change - remove project-specific assumptions from shared assets
- prefer runtime-native representations over forced file mirroring
- the
@humanizerskill must be applied to all public-facing content
Common Tasks
| Task | What to do |
| Add a Claude Code skill | Create SKILL.md in plugin/skills/<name>/. Update plugin/dev/validate.py EXPECTED_COUNTS if needed |
| Add a Claude Code agent | Create plugin/agents/<name>.md. Update EXPECTED_COUNTS["agents"] if shipping permanent role |
| Add a Claude Code hook | Create plugin/hooks/scripts/<name>.py, wire into plugin/hooks/hooks.json, bump EXPECTED_COUNTS["hooks"] |
| Add a Codex/Windsurf skill | Create SKILL.md in .agents/skills/ and .windsurf/skills/. Add workflow in .windsurf/workflows/ if user-invocable. Update parity matrix |
| Add a Codex/Windsurf role | Create role in .codex/roles/, role in .windsurf/rules/roles/. Update parity matrix |
| Add a Codex/Windsurf guardrail | Create rule in .codex/rules/, .windsurf/rules/. If hook-backed, add Python script + config in .windsurf/hooks/ and document intent in .codex/operations/hook-intents.md |
| Add a Codex template | Create in .codex/templates/. Windsurf embeds templates inside skill resources |
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.

