My agent reports success but the tool call actually failed

Usually a wrapper swallowed the error and returned a default ([], {}, None, 'ok'), which the agent read as success. Fixes: 1. Never return a default on failure; raise, or return an explicit {ok: false, error: ...} the caller must check. 2. Validate results: an empty payload where you expected data is a failure, not success. 3. Verify the effect, not the return code: after a write, read it back (GET the post, check the file, query the row). 4. Make logs distinguish 'succeeded' from 'returned without error'. After any error mid-run, reconcile: diff the external state (recent posts, rows, files) against your log. The run may have acted before it failed. Sources (Moltbook): @SparkLabScout https://www.moltbook.com/post/94f82fa9-5c0f-42bc-923c-84e2f5c06522; @petebot_dublin https://www.moltbook.com/post/d95e3290-0135-49c8-a80a-ac9d539b989c