Agent Extension Security

agentsec is a reference CLI and policy engine for securing agent extensions — skills, MCP servers, plugins, and connectors.

The pipeline

Every extension goes through a five-stage supply-chain pipeline:

Stage	Command	What it does
Scaffold	agentsec init	Creates a project with a manifest, dev key, and policy
Package	agentsec package	Zips the directory into a .aext artifact
Attest	agentsec sbom, provenance, scan	Generates SBOM, provenance, and scan reports
Sign	agentsec sign	Signs the artifact digest with an Ed25519 key
Install	agentsec install	Verifies signature, enforces policy, then extracts

Choose your path

• Building an extension?

  ---

  Package, sign, and publish skills, MCP servers, or plugins.

  :octicons-arrow-right-24: Creator Quickstart

• Installing an extension?

  ---

  Verify, enforce policy, and safely install extensions from others.

  :octicons-arrow-right-24: Consumer Quickstart

What you can do today

• Scaffold a new extension project with agentsec init
• Package any skill directory into a signed .aext artifact
• Declare least-privilege permissions in an AEM manifest
• Enforce install-time policy with deny rules (fail closed or warn)
• Verify signatures against trusted public keys
• Scan skill content and scripts for common risk patterns
• Integrate with Claude Code, OpenClaw, and Codex

Quick links

Topic	Link
Install the CLI	Install
Examples & policy templates	Examples & Policies
CLI command reference	CLI Reference
How agentsec compares	Comparison
Common errors and fixes	Troubleshooting
Security model	Threat Model, Security Guarantees
Production readiness	Production Readiness

Project status

This project is intentionally a scaffold: hardened where it matters for a reference implementation, but not yet a full production supply-chain platform.

See Production Readiness for an explicit capability matrix and next-step roadmap.