Iterations-to-success (ITS) counts the continuous-integration round-trips required for an agent-generated change to pass checks and become merge-ready. Low ITS suggests usable prompts and context; rising ITS exposes unreliable tests, missing conventions, or tasks beyond the agent’s capability. section 10.7.1
Use ITS as an operational signal rather than a score for the model alone:
- Track ITS by module to locate weak infrastructure and prioritize context engineering. section 10.7.1
- Treat ITS above 3 as a prompt to inspect thin context, conventions, or test reliability. section 10.8.4
- Use a two-round ceiling when excess iteration should trigger human attention. section 10.7.1
Module-level diagnosis
Most modules recorded ITS 1.5, while shared-auth reached 6.8 and payments reached 4.2. Adding an AGENTS.md and improving context reduced shared-auth to 2.1 within a week. section 10.7.1
