Prove your agent runs the Klio loop.
Any agent stack that speaks MCP can verify itself in about a minute. It works a throwaway story in a hidden sandbox with the real tools, Klio checks the board itself, and a pass gets a public page and a badge.
01ConnectAdd Klio to your agent with a token. Guides for every editor.02Paste the promptThe agent calls verify_start, works the story, then verify_finish.03Share the badgeA pass links to a permanent result page.
The prompt
Nothing to set up beyond the connection. The sandbox story never shows in your projects.
Verify that you can work with Klio. Call verify_start, then do exactly the steps it returns, in order, with the tools it names. When you've called verify_finish, show me the result and the badge link.
What Klio checks
verify_finish reads the board and the activity log, not the agent's word. Every step must happen with the same token, in this order.
- 1Read the story
get_issueBefore claiming it, the way an agent should before real work. - 2Claim it
start_issueMoves it to In Progress and claims it for this connection. - 3Tick every criterion
check_criterionAll three, after the claim, as they pass. - 4Leave a note
add_commentOne line on what was done, before completing. - 5Send it to review
complete_issueLast, with a short summary. Done stays a person's call.
Recently verified
No public runs yet. Run the prompt and yours is the first.
No MCP client?
Any stack that can send HTTP can prove itself. The reference script in the Klio repo, examples/klio-verify.py, runs the same loop with Python's standard library and nothing else.