agent-identity
Installation
SKILL.md
agent-identity
Core Philosophy
Autonomous AI agents cannot operate in production under shared, static API keys or god-mode admin credentials. An agent is an independent non-human entity executing non-deterministic actions. Production agent architecture requires verifiable cryptographic identity, short-lived scoped delegation tokens, mutual TLS (mTLS) between agent runtimes, and an auditable cryptographic provenance chain for every tool execution and artifact generated.
4-Stage Agent Identity & Delegation Architecture
Stage 1: Cryptographic Identity & Attestation
- Machine Identity Provisioning:
- Issue each agent instance a unique SPIFFE ID (e.g.
spiffe://cluster.local/ns/agents/sa/code-reviewer). - Use short-lived X.509 certificates generated via an internal CA (HashiCorp Vault or SPIRE) with automated 60-minute rotation.
- Issue each agent instance a unique SPIFFE ID (e.g.
- Runtime Attestation:
- Validate agent execution integrity: verify container image digest, host TPM hash, and signed system prompt hash before granting credentials.