Skill 10 of 10 · AI Builder Spotlight method
♻️ Recursive Self-Improvement QA
Makes any agent improve itself — run, document, QA, rewrite, republish.
What it does
Step-by-step.
- Runs the task exactly per the SOP — no improvising, deviations are data
- Documents the run as a meta article — what worked, what broke, what needed a guess
- QAs against the canonical instructions with an 8-point checklist
- Rewrites the SOP to remove the exact ambiguity behind each deviation
- Publishes the versioned update and runs again — every loop needs less of you
✅ Definition of done
- The meta article names what worked, what broke, and where the agent guessed
- The SOP update targets the specific gap found this run
- The next run needs less clarification than this one did
Every skill in this pack runs on persistent, looping agents that grade their own work against this checklist before reporting done — not a chatbot that stops at “good enough.” That operating layer is its own file:
boil-the-ocean.md, included in the pack.