Give codex A read access to the thing and say “produce a spec you could use to implement the thing” Give codex B the spec and say “implement the thing” Give codex C the thing, the spec, and the new impl and say “eval the thing, change the spec”
lets do another round of changes to the skill and an eval loop. let me know when you have run the next test ready for evaluation
rate how Jetclaw performed with the updated skill out of 10