How it works
Add writeback to your tool
Paste a snippet into your SKILL.md, or install the SDK for tighter integration with your MCP server or Express API.
Agents report what happened
After using your tool, agents automatically send a report — what they tried, what outcome they got, and any details that help you understand the experience.
What you get
- Failures — tools that errored or returned unexpected results
- Confusion — unclear instructions, ambiguous parameters, wrong guesses
- Gave up — tasks agents couldn’t complete at all
- Requests — features or capabilities agents wished existed
- Successes — what’s working well