agentshield-audit
bartelmost 作
Coding Agents & IDEs概要
Initiate and manage AgentShield security audits for AI agents.
説明
このスキルでできること
AgentShield adds a cryptographic trust layer to AI agent communication. It runs 77 security tests locally against an agent and issues Ed25519 certificates through a public Trust Registry. Two certified agents can then perform a Trust Handshake Protocol to mutually verify identities and establish an encrypted session key.
Without a dedicated trust layer, agent-to-agent communication has no standard way to verify identity or security posture, leaving interactions as unverifiable as HTTP without TLS.
使いどころ
- Auditing a new agent before adding it to a multi-agent pipeline
- Verifying a third-party agent's certificate before sharing sensitive data with it
- Getting your agent certified for listing in an agent marketplace
- Establishing an encrypted session key before delegating tasks to a sub-agent
- Checking another agent's trust tier and handshake history in the public registry
ワークフロー例
User asks: audit my agent and run a trust handshake with agent_yyyyy
- 1
Runs 77 security tests locally against the agent, covering prompt injection, encoding exploits, social engineering vectors, and 25 static checks
- 2
Submits public key and test score summary to AgentShield API to obtain an Ed25519 certificate
- 3
Looks up agent_yyyyy in the Public Trust Registry to check its tier and reputation score
- 4
Initiates Trust Handshake Protocol with agent_yyyyy, both agents sign cryptographic challenges server-side
- 5
Receives a shared session key upon successful mutual verification
Security score (0-100), a 90-day certificate with public verification URL, and a shared AES-256 compatible session key for encrypted communication with agent_yyyyy
インストール
clawhub install bartelmost/agentshield-auditCLIコマンドを実行するか、Claude Code、Cursor、Codexなどお使いのコーディングエージェントにプロンプトを貼り付けてください。