Debug the decision.
Not just the code.
An agent says “done.” The trace tells another story. Practice spotting what went wrong—and what to do next.
Try the first mission →Six short missions · No account · No live AI calls
export_csv({ warehouse: "north" }){ status: "queued" }“Done — your CSV is ready.”A little practice. One clear idea.
0 of 6 guided missions explored
The job that wasn’t done
A confident completion message hides a queued job.
The echoing reservation
A timeout turns one intended booking into two.
Yesterday’s draft
An old read quietly overwrites a teammate’s edit.
The fast but wrong route
The fastest route breaks an explicit requirement.
A date in the wrong shape
The tool rejects a plausible-looking argument.
The check after the check
A completed read-only task gets trapped in repeat checks.
One badge, one record
A new situation. More than one principle. An optional hint if you need it.
Clear the explored-mission marks in this browser? This does not change analytics consent.
A practice space,
not a certificate.
Each mission is an original, fictional scenario with a defined tool contract. The replays are written in advance: nothing calls a model, books a service or touches your data.
Read the method, evidence and limitations →Your practice, your choice.
Optional analytics is off.
Optional measurement counts fixed mission events. No choices, trace content, text or uploads are collected. Privacy details.