A 7-step test loop for agent prompts: build the test before you tweak the prompt
petrusenko_max · x · 2026-10-29
A practical loop for iterating on any agent or skill's prompt: (1) collect every failure with its triggering input in a folder—that folder is your test suite; (2) write checks a script can run, not human judgment calls; (3) add 5-10 passing cases so fixes can't silently break them; (4) score the current prompt as a baseline; (5) change exactly one thing (prompt, tool description, or model) at a time; (6) rerun everything and keep the change only if score rises with no regressions; (7) when production breaks, the case goes into the folder before the fix. Prompts get rewritten and models swapped, but the folder of failures keeps telling the truth.
More from coding & agent
- Dev asks Google for official MCP servers for Recorder and Tasks so agents can use them — steren · 2026-10-29
- Adding Dark Mode Takes 10 Minutes With Claude — Why Doesn't Your Site Have It? — braelyn_ai · 2026-10-29
- Yacine: Coding AIs Give People What They Want, But People Don't Know What's Worth Wanting — yacineMTB · 2026-10-29
- An Agentic RL Starter Reading List: Three Papers Covering Basics, Frontiers, and the Big Picture — rbhar90 · 2026-10-29
- Parley: open-source Go CLI lets multiple coding agents collaborate over a local IRC channel — colinsurprenant · 2026-10-29
- Ask an agent whether your flight has Starlink — United now gives per-flight odds — SamuelAlbanie · 2026-10-29