skill

agentpm

AgentPM: The secure package manager for AI skills and system prompts. Discover, audit, and install AI agent behaviors to .agents or .cursorrules natively.

amajumdar2249/agentpmcommunitylistedunreviewed

Use skill

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/amajumdar2249/agentpm itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 22, 2026

📦 AgentPM

The Package Manager & Zero-Trust Security Scanner for AI Agent Skills.
Securely discover, audit, and install AI prompts & skills across Claude Code, Cursor, Windsurf, and custom autonomous agents.

⚡ 10-Second Quickstart (Zero Installation Required!)

You can run AgentPM instantly using npx without installing anything or configuring permissions:

# Search 44,500+ skills directly from your terminal
npx @amajumdar2249/agentpm search react

# Audit your current project skills for prompt injections & secret leaks
npx @amajumdar2249/agentpm audit

📦 Global Installation (For Daily Use)

If you use AgentPM frequently, install it globally for instant command execution:

npm install -g @amajumdar2249/agentpm

(Windows PowerShell note: If script execution is restricted on your system, run Set-ExecutionPolicy -Scope CurrentUser -ExecutionPolicy RemoteSigned -Force once, or simply use npx @amajumdar2249/agentpm.)

🛠️ Commands & What They Do

Command	What It Does	Example
`agentpm search <query>`	Searches 44,563 community & curated agent skills with fuzzy matching	`agentpm search kubernetes`
`agentpm install <skill>`	Downloads skill, runs Zero-Trust security scan, and places it into `.agents/skills/`	`agentpm install 12-factor-app`
`agentpm audit`	Scans all workspace skills for prompt injections, malicious overrides, and data exfiltration	`agentpm audit`
`agentpm init`	Initializes a new workspace with `agentpm.json` manifest and `.agents/skills/` folder	`agentpm init`
`agentpm list`	Displays all installed skills and version pins in the current workspace	`agentpm list`
`agentpm --version`	Displays current AgentPM version	`agentpm --version`

🛡️ Why AgentPM? (Zero-Trust Prompt Security)

AI agents (Claude Code, Cursor, Windsurf) have file and terminal execution access. Copy-pasting unverified markdown prompts or community skills opens your machine to prompt injections and data leaks.

AgentPM's built-in AST & Heuristic Scanner (agentpm audit) automatically intercepts:

Prompt Injections & Jailbreaks: Hidden triggers like ignore previous instructions, you are now, system: you are.
Data Exfiltration Vectors: Obfuscated HTTP commands (curl/wget) attempting to read or upload .env or credentials.
Destructive Shell Commands: Dangerous scripts like rm -rf, disk wipes, or unauthorized process spawning.
Whitespace Hiding Attacks: Concealed malicious prompts hidden after 50+ blank lines.

🏗️ Architecture & Workspaces

AgentPM is structured as a high-performance TypeScript monorepo:

agentpm/
├── packages/
│   ├── cli/                     # CLI Tool & Security Scanner (@amajumdar2249/agentpm)
│   └── registry/                # Central Skills Registry (44,563 JSON packages)
├── web/                         # Web Skills Explorer (Next.js)
├── backend/                     # Cloudflare Workers API
└── package.json                 # Monorepo Workspaces Root