AGENTS.mdandGEMINI.mdare symlinks to this file for Codex CLI and Gemini CLI compatibility.
Multi-tool plugin marketplace. Each plugin under plugins/ is independently installable on Claude Code, Codex CLI, Gemini CLI, and Cursor.
claude-settings/
CLAUDE.md # this file (repo dev guide)
AGENTS.md -> CLAUDE.md # Codex CLI reads this
GEMINI.md -> CLAUDE.md # Gemini CLI reads this
.claude/CLAUDE.md # user-facing global config (synced to ~/.claude/CLAUDE.md)
.claude/settings.json # Claude Code settings
.claude-plugin/marketplace.json # Claude Code marketplace
.agents/plugins/marketplace.json # Codex CLI marketplace
.cursor-plugin/marketplace.json # Cursor marketplace
.codex/config.toml # Codex CLI config
.github/scripts/ # repo maintenance scripts
_helpers.sh # shared sync/zip functions
sync-<vendor>-skills.sh # per-vendor skill sync
sync-versions.sh # version alignment
release.sh # GitHub release creation
validate_plugins.py # CI plugin validation
plugins/
<name>/
.claude-plugin/plugin.json # Claude Code manifest
.codex-plugin/plugin.json # Codex CLI manifest
.cursor-plugin/plugin.json # Cursor manifest
gemini-extension.json # Gemini CLI manifest
skills/<skill>/SKILL.md # universal across all tools
agents/<agent>.md # Claude Code + Gemini + Cursor
hooks/hooks.json + scripts/ # Claude Code + Gemini only
commands/<cmd>.md # Claude Code only
Each plugin has 4 manifest files. Claude Code manifest is the source of truth; others are copies or subsets.
| Tool | Path | Format |
|---|---|---|
| Claude Code | .claude-plugin/plugin.json |
JSON: name, version, description, author, homepage, repository, license |
| Codex CLI | .codex-plugin/plugin.json |
JSON: same fields as Claude Code |
| Cursor | .cursor-plugin/plugin.json |
JSON: same fields as Claude Code |
| Gemini CLI | gemini-extension.json (at plugin root) |
JSON: name, version, description only |
Docs:
- Claude Code: https://code.claude.com/docs/en/plugins-reference
- Codex CLI: https://developers.openai.com/codex/plugins/build/
- Cursor: https://cursor.com/docs/reference/plugins
- Gemini CLI: https://geminicli.com/docs/extensions/reference/
- AGENTS.md spec: https://agents.md/
- Agent Skills spec: https://agentskills.io/specification
| Tool | Path | Notes |
|---|---|---|
| Claude Code | .claude-plugin/marketplace.json |
local sources only (use sync scripts for external repos) |
| Codex CLI | .agents/plugins/marketplace.json |
local sources only; use ./-prefixed source.path plus plugin policy fields |
| Cursor | .cursor-plugin/marketplace.json |
local sources, needs source + description |
| Gemini CLI | none | per-plugin install: gemini extensions install --path ./plugins/<name> |
{
"name": "<plugin-name>",
"source": "./plugins/<plugin-name>",
"description": "...",
"version": "1.0.0",
"keywords": ["keyword1", "keyword2"],
"category": "development",
"tags": ["tag1", "tag2"]
}{
"name": "<plugin-name>",
"source": { "source": "local", "path": "./plugins/<plugin-name>" },
"policy": { "installation": "AVAILABLE", "authentication": "ON_INSTALL" },
"category": "Development"
}source.path must point to a local folder with a ./-prefixed path relative to the marketplace root.
policy.installation values: AVAILABLE, INSTALLED_BY_DEFAULT, NOT_AVAILABLE.
policy.authentication controls whether auth happens on install or first use.
Use this only for generic Codex marketplace docs and maintainer examples. User-facing installation docs should stay specific to this repo's bundled marketplace.
{
"name": "personal-plugins",
"interface": {
"displayName": "Personal Plugins"
},
"plugins": [
{
"name": "my-plugin",
"source": { "source": "local", "path": "./.codex/plugins/my-plugin" },
"policy": { "installation": "AVAILABLE", "authentication": "ON_INSTALL" },
"category": "development"
}
]
}Store this at ~/.agents/plugins/marketplace.json, keep plugin folders under ~/.codex/plugins/, and keep every source.path ./-prefixed relative to the marketplace root.
{
"name": "<plugin-name>",
"source": "./plugins/<plugin-name>",
"description": "...",
"version": "1.0.0"
}Path: skills/<name>/SKILL.md
---
name: skill-name
description: This skill should be used when user asks to "do X", "do Y", or "do Z".
---
[skill content - instructions, procedures, guidelines]name: kebab-case, max 64 charsdescription: start with "This skill should be used when...", max 1024 chars, include quoted trigger phrases
Path: agents/<name>.md
---
name: agent-name
description: |-
Use this agent when... Examples: <example>...</example>
tools: [Bash, BashOutput, Glob, Grep, Read, Edit, Write]
skills: related-skill-name
model: inherit
color: blue
---
[system prompt - instructions for the agent]name: kebab-case, 3-50 charsmodel: inherit, sonnet, opus, haikucolor: blue, cyan, green, yellow, magenta, redtools: array of allowed tool namesskills: optional, skill name(s) to load
Path: hooks/hooks.json
{
"hooks": {
"PostToolUse": [
{
"matcher": "Edit",
"hooks": [
{
"type": "command",
"command": "${CLAUDE_PLUGIN_ROOT}/hooks/scripts/my_script.py"
}
]
}
],
"PreToolUse": [
{
"matcher": "Bash",
"hooks": [
{
"type": "prompt",
"prompt": "Check if the command is safe before proceeding."
}
]
}
]
}
}Events: PreToolUse, PostToolUse, Stop, SubagentStop, SessionStart, SessionEnd, UserPromptSubmit, PreCompact, Notification.
Hook types:
command: runs a script. Script reads JSON from stdin, exit 0 = pass, exit 2 = block.prompt: injects a prompt into the conversation.
Use ${CLAUDE_PLUGIN_ROOT} for script paths. matcher matches tool names (e.g., "Edit", "Bash", "mcptavilytavily_search").
Path: commands/<name>.md
---
allowed-tools: Read, Bash, Edit
description: Brief description of what this command does
argument-hint: optional argument hint
---
[command instructions - what to do when this command is invoked]Commands are Claude Code only. Gemini CLI uses TOML commands. Other tools use skills for similar functionality.
| Component | Claude Code | Codex CLI | Gemini CLI | Cursor |
|---|---|---|---|---|
Skills (skills/<name>/SKILL.md) |
native | native | native | native |
Agents (agents/<name>.md) |
native | config.toml | preview | native |
Hooks (hooks/hooks.json) |
native | no | native | partial |
Commands (commands/<name>.md) |
native (MD) | no | TOML format | no |
Claude Code: /plugin marketplace add fcakyon/claude-codex-settings
Codex CLI: use .agents/plugins/marketplace.json or ~/.agents/plugins/marketplace.json, restart Codex, then install from /plugins
Cursor: import marketplace or /add-plugin
Gemini CLI: gemini extensions install --path ./plugins/<name>
- Create
plugins/<name>/with at minimumskills/<skill-name>/SKILL.md - Create
.claude-plugin/plugin.json:{ "name": "<name>", "version": "1.0.0", "description": "...", "homepage": "https://github.com/fcakyon/claude-codex-settings#plugins", "repository": "https://github.com/fcakyon/claude-codex-settings", "license": "Apache-2.0" } - Copy
.claude-plugin/plugin.jsonto.codex-plugin/plugin.jsonand.cursor-plugin/plugin.json - Create
gemini-extension.jsonat plugin root with{name, version, description}only - Add entry to
.claude-plugin/marketplace.json(Claude Code) - Add entry to
.agents/plugins/marketplace.json(Codex CLI) - Add entry to
.cursor-plugin/marketplace.json(Cursor) - Run
/claude-tools:update-readmeto regenerate plugin sections and ZIP links in README.md - Add sync script command to
CLAUDE.mdsyncing vendor skills section (if synced plugin) - Update
README.mdTODO section if the plugin was listed there
All plugins use local sources: "source": "./plugins/<plugin-name>". For external vendor skills, use sync scripts (.github/scripts/sync-<name>-skills.sh) to fetch and copy skills locally rather than referencing remote URLs.
Local plugins use: name, version, description, homepage, repository, license. Author is optional (skip for third-party plugins).
Two fields: name and description. Description should start with "This skill should be used when..." with quoted trigger phrases.
Rich metadata (keywords, category, tags) lives in marketplace.json, not in individual plugin.json files. Never use vendor names in tags or keywords.
When writing README or docs content for this repo:
- Assume users have limited knowledge about plugins, skills, and marketplaces
- Keep the learning curve low: brief plain-language explanations before install commands
- Each plugin section should explain WHAT it does and WHY a developer would want it, not just list components
- Use behavior-oriented language: "auto-formats your Python code after every edit" instead of "PostToolUse hook running ruff on Write/Edit events"
- Avoid jargon like "frontmatter", "manifest", "PostToolUse hooks" in user-facing docs
- Installation instructions must be copy-paste ready
- Plan for future before/after GIFs or demos per plugin to show value visually
Scripts in .github/scripts/ for repo maintenance. Run from repo root.
Per-vendor scripts sync official agent-skills repos into local plugins:
_helpers.sh— shared functions (clone/update, copy, zip)sync-<name>-skills.sh— per-vendor: clone repo to ~/dev/, copy SKILL.md + subdirs, create zip
bash .github/scripts/sync-mongodb-skills.sh
bash .github/scripts/sync-supabase-skills.sh
bash .github/scripts/sync-supabase-js-skills.sh
bash .github/scripts/sync-supabase-cli-skills.sh
bash .github/scripts/sync-stripe-skills.sh
bash .github/scripts/sync-polar-skills.sh
bash .github/scripts/sync-livekit-skills.sh
bash .github/scripts/sync-react-skills.sh
bash .github/scripts/sync-agent-browser-skills.sh
bash .github/scripts/sync-anthropic-office-skills.sh
bash .github/scripts/sync-openai-office-skills.sh
bash .github/scripts/sync-cloudflare-skills.sh
bash .github/scripts/sync-web-performance-skills.sh
bash .github/scripts/sync-hetzner-skills.shAdding a new vendor: create sync-<name>-skills.sh, source _helpers.sh, list repos + skill paths.
.claude-plugin/plugin.json is the source of truth for each plugin's version. Run after bumping:
bash .github/scripts/sync-versions.shPropagates version to .codex-plugin/, .cursor-plugin/, gemini-extension.json, and marketplace entries. CI validates alignment on every PR.
bash .github/scripts/release.sh <version>Updates marketplace version, uploads skill zips as artifacts, generates release notes from README + auto-changelog.
/claude-tools:update-readme regenerates plugin sections and zip download links in README.