agent-passport
由 markneville 提供
Web & Frontend Development简介
OAuth for the agentic era — consent-gating for ALL sensitive agent actions including purchases, emails, file.
描述
该技能的功能
Agent Passport adds a consent layer between AI agents and sensitive actions. Users grant time-limited mandates that define what an agent can do, to whom, and under what constraints. Covers shell commands, emails, file writes, API calls, purchases, and public posts as distinct action categories with separate allowlists and rate limits.
Replaces all-or-nothing agent permissions with granular, expiring mandates scoped by action type, target, and rate limit rather than requiring a trust decision on the entire agent.
何时使用
- Limiting agent email sending to a specific company domain for 30 days
- Setting a per-transaction spending cap on automated purchases
- Freezing all agent actions instantly with the kill switch
- Reviewing an audit trail of every command an agent ran
- Granting temporary file write access to a specific folder path
示例工作流
User asks: Allow my agent to send emails to my team at company.com
- 1
Run mandate-ledger.sh create-from-template email-team company.com
- 2
Confirm the mandate is created with the correct domain allowlist and daily rate limit
- 3
Before sending any email, call check-action with category 'communication' and the recipient address
- 4
On authorized: true, send the email and call log-action with the mandate ID and a human-readable description
Agent can send emails to *@company.com within the configured rate limit; every send is recorded in the audit trail
前置要求
- AGENT_PASSPORT_LEDGER_DIR environment variable
- AGENT_PASSPORT_LICENSE_KEY environment variable (Pro tier only, for real-time threat definition updates from api.agentpassportai.com)
安装
clawhub install markneville/agent-passport运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。