Agents Reflection Skills

Meta-skills that let AI coding agents configure themselves

No more editing config files. Just tell your agent what you need.

Русский • 中文 • Português

---

Installation

Via Skills CLI (recommended):

claude install-skill CodeAlive-AI/agents-reflection-skills

Via plugin marketplace:

/plugin marketplace add https://github.com/CodeAlive-AI/agents-reflection-skills.git
/plugin install agents-reflection-skills@agents-reflection-skills
# Restart Claude Code for changes to take effect

---

What's Included

This plugin provides 7 skills that work across Claude Code, Codex CLI, Cursor, VS Code, Gemini CLI, and more:

Skill	What It Does
mcp-management	Install and manage MCP servers across 10+ coding agents
hooks-management	Manage hooks and automation for Claude Code and Codex CLI
settings-management	Configure settings for Claude Code (JSON) and Codex CLI (TOML)
subagents-management	Create and manage subagents across Claude Code and Codex CLI
skills-management	Organize, discover, and share skills for coding agents
plugins-management	Package and publish your own plugins
optimizing-claude-code	Audit repos and optimize CLAUDE.md for agent work

Lightweight: All 7 skill descriptions combined use less than 500 tokens in your context window.

---

Use Cases (skills)

mcp-management

Install and manage MCP servers across 10+ coding agents

Install to All Your Agents at Once

"Install the Postgres MCP server to Claude Code, Cursor, and VS Code"

Uses add-mcp to install MCP servers to 10+ coding agents (Claude Code, Cursor, VS Code, Claude Desktop, Gemini CLI, Codex, Goose, GitHub Copilot CLI, OpenCode, Zed) with a single command. Handles config format differences (JSON, YAML, TOML) automatically.

Connect to Your Database

"Connect Claude to my PostgreSQL database"

Installs the database MCP server, configures your connection string. Now you can query your data conversationally.

Install GitHub Integration

"Connect Claude to GitHub so it can create PRs and manage issues"

Installs the official GitHub MCP server. Claude can now create branches, PRs, and work with issues directly.

Multi-Agent Sync

"Make sure all my coding agents have the same MCP servers"

Configures MCP servers consistently across Claude Code, Cursor, VS Code, Gemini CLI, and other agents, handling each agent's config format and paths.

---

hooks-management

Manage hooks and automation for Claude Code and Codex CLI

Auto-Format Code

"Run Prettier on TypeScript files after every edit"

Adds a PostToolUse hook. Every file Claude touches gets formatted automatically — no more style inconsistencies.

Auto-Run Tests

"Run pytest whenever Claude edits Python files"

Adds a PostToolUse hook for *.py files. Instant feedback on whether changes broke anything.

Log All Commands

"Log every command Claude runs to an audit file"

Adds a PreToolUse hook that appends commands to ~/.claude/command-log.txt. Full audit trail.

Block Dangerous Commands

"Add hook that blocks global rm -rf commands"

Adds a PreToolUse hook that rejects destructive commands before they run. Protects against accidental data loss.

Require Confirmation

"Add hook that asks user permission for commands containing 'db reset'"

Adds a PreToolUse hook that pauses and requests confirmation for database resets or other sensitive operations.

---

settings-management

Configure settings for Claude Code (JSON) and Codex CLI (TOML)

Enable Sandbox Mode

"Turn on sandbox mode so Claude can work without asking permission for every command"

Enables native sandboxing — reduces permission prompts by 84% while keeping your system safe.

Block Access to Secrets

"Block Claude from reading .env files and anything in /secrets"

Adds deny rules to permissions. Sensitive files stay protected even if Claude tries to access them.

Switch Models

"Use Opus model for this project"

Updates settings to use claude-opus-4-5 — better reasoning for complex architectural decisions.

Share Team Configuration

"Set up project settings so everyone gets the same Claude config"

Creates .claude/settings.json in project scope. Commit once, entire team gets identical setup.

---

subagents-management

Create and manage subagents across Claude Code and Codex CLI

Create a Code Reviewer

"Create a reviewer subagent that can only read files, uses Opus for quality"

Creates a custom subagent with tools: Read, Grep, Glob and model: opus. Thorough, insightful code reviews.

Create a Test Runner

"Create a subagent that runs tests and reports failures"

Creates a specialized agent for running test suites with limited tool access for safety.

---

skills-management

Organize, discover, and share skills for coding agents

List Available Skills

"Show me all my installed skills"

Lists skills from user and project scopes with their triggers and descriptions.

Move Skills Between Scopes

"Move this skill to my user scope so I can use it everywhere"

Moves skills between project and user scopes for broader or narrower availability.

---

plugins-management

Package and publish your own plugins

Create a Plugin

"Create a new plugin with my custom skills"

Scaffolds a new plugin structure with manifest, README, and skill directories.

Publish to GitHub

"Publish my plugin to GitHub"

Packages your plugin and creates a GitHub release for others to install.

---

optimizing-claude-code

Audit repos and optimize CLAUDE.md for agent work

Audit Project Readiness

"Check if this repo is set up well for Claude Code"

Runs a comprehensive audit analyzing CLAUDE.md files, settings, MCP configs, and project structure. Returns prioritized recommendations (P0/P1/P2) with specific suggestions for improvement.

Optimize CLAUDE.md

"Review my CLAUDE.md and suggest improvements"

Evaluates memory file quality: structure, conciseness, @import validation, essential sections (commands, architecture, conventions). Provides concrete edits to make your project more agent-friendly.

---

Requirements

• Claude Code CLI or Codex CLI
• Python 3.x (for skill scripts)
• gh CLI (optional, for plugin publishing features)

---

Structure

agents-reflection-skills/
├── .claude-plugin/
│   └── marketplace.json         (marketplace catalog)
├── plugins/
│   └── agents-reflection-skills/
│       ├── .claude-plugin/
│       │   └── plugin.json
│       ├── skills/
│       │   ├── mcp-management/
│       │   ├── hooks-management/
│       │   ├── settings-management/
│       │   ├── subagents-management/
│       │   ├── skills-management/
│       │   ├── plugins-management/
│       │   └── optimizing-claude-code/
│       ├── LICENSE
│       └── README.md
├── CLAUDE.md
├── LICENSE
└── README.md

---

Learn More

• MCP Servers Guide
• Hooks Documentation
• Custom Subagents
• Sandboxing

---

Changelog

v2.0.1 — Bug fixes and skill management improvements

• Fixed shutil.rmtree crash on symlinked skills (installed via npx skills add)
• Added multi-agent deletion: --agent and --all-agents flags for delete_skill.py
• Added npx skills remove, npx skills list, and update workflow guidance
• Renamed plugin directory to agents-reflection-skills

v2.0.0 — Universal agent support

• Renamed all skills: dropped claude- prefix, manager to management
• MCP management now supports 10+ agents via npx add-mcp integration
• Added Codex CLI references (TOML config, AGENTS.md, Starlark rules)
• 42-agent registry for cross-agent skill operations
• Remote skill assessment framework for evaluating ecosystem skills
• Standardized reference file naming (claude-*, codex-*)

v1.1.1 — Hooks improvements

• Script-first guidance for complex hooks
• Recommended script location: ~/.claude/hooks/

v1.1.0 — Multi-agent support v1.0.0 — Initial release

License

MIT — see LICENSE

---

_{Built with Claude Code}