agent-audit-trail
由 roosch269 提供
Coding Agents & IDEs简介
Tamper-evident, hash-chained audit logging for AI agents.
描述
该技能的功能
Hash-chained audit logging for AI agents using SHA-256 to create tamper-evident records. Covers EU AI Act Articles 12, 14, and 50 on record-keeping, human oversight, and transparency. Runs as a standalone Python script with no external dependencies.
Any modification to the log breaks the SHA-256 hash chain, making tampering detectable with a single verify command and no external tooling required.
何时使用
- Logging every file write and command execution made by an AI agent
- Verifying audit log integrity after an automated pipeline completes
- Linking agent actions to human approval references before sensitive operations
- Preparing tamper-evident records for EU AI Act compliance audits
- Tracking credential access events across multi-step agent workflows
示例工作流
User asks: log and verify that my agent backed up the production database
- 1
Agent runs pg_dump on the production database
- 2
Calls auditlog.py append with kind 'exec', target 'pg_dump production', summary, and a human approval gate reference
- 3
Script computes SHA-256 hash chaining the new entry to the previous log entry
- 4
User runs auditlog.py verify to confirm chain integrity
OK (N entries verified) — audit trail shows a timestamped exec event with actor, domain, gate reference, and unbroken hash chain
安装
clawhub install roosch269/agent-audit-trail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。