ClawDiary is an audit diary and guard for AI agents. I use Cursor a lot and wanted something that (a) logs what the agent does and (b) blocks dangerous actions until I approve them in Telegram.

What it does: - Audit: agents POST to /v1/audit after each action; writes go to D1 async so the agent isn’t blocked. - Guard: before risky ops (rm, drop table, execute_bash, transfer, etc.) the agent calls /v1/guard. Request is stored and the call blocks until you approve or reject in Telegram (or by updating DB). Green-light actions return immediately. - Diary, daily digest (cron), and a minimal timeline UI so I can see what happened.

Stack: Cloudflare Worker, Hono, D1, Durable Objects (for holding the guard request until approval). Cursor integration is via a .cursor/rules rule that calls the API; MCP-capable agents can use the tool described at /mcp.json.

Repo: https://github.com/jetywolf/claw-diary Live API/docs: https://api.clawdiary.org/docs

Happy to answer questions or take feedback.