theAuth for AI Agents: Identity and Scoped Permissions
说明与用途
文章说明如何将多个 AI 代理使用的共享 API 密钥替换为具有受限权限的单独令牌,实现审计和单独撤销每个令牌的功能。
问题
A shared human API key used by multiple AI agents and scripts caused broad permissions, difficulty revoking individual agents, and a misleading audit trail.
结果
Agents operate with minimal blast radius, can be revoked individually, and actions are auditable per agent.
工作流
- Install Node 20+, TypeScript, and theauth package
- Create a SQLite database instance with agents enabled
- Seed an owner user row
- Create one agent per job using
theauth.agent.create - Call
authorize()before each agent action - Add constraints such as rate limits and argument patterns
- Check tokens in middleware to return 401/403
- Delegate to sub-agents with narrower scopes
- Read audit trail, rotate, and revoke tokens
证据
A shared human key has three problems, and they stack.
Each agent holds a token that maps to one row, one owner, and one permission list.
Call authorize() before every action
- 记录日期: 2026-10-10T23:11:04.136Z