Description
Issue #38 (hydramancer.com onboarding portal) had its plan rejected 21 consecutive times by nebula between 2026-03-15 00:30 and 14:28 — nearly 14 hours of churning. The planner kept producing essentially the same plan, and nebula kept rejecting it for the same reason ("needs human sign-off on infrastructure decisions"), even though the plan pre-made all the decisions in a clear table.
This indicates a missing circuit breaker in the plan/review loop. After N rejections (e.g. 3), the system should either:
1. Escalate to a human for review instead of continuing autonomously
2. Mark the issue as blocked with a clear reason
3. Stop the loop entirely and flag it
Investigate:
- Is this loop still happening on other issues currently in planning/proposed/rejected status?
- What in nebula's review logic causes it to reject valid plans repeatedly?
- What in the planner causes it to re-submit essentially identical plans after rejection?
Fix: Add a circuit breaker that stops the plan/review cycle after 3 consecutive rejections and escalates to human review.
Nebula's reasoning: This is a high-priority bug — a missing circuit breaker allowed 14 hours of wasted compute and blocked issue #38 from progressing. The original title buried the root cause (missing circuit breaker) behind the symptom (21 rejections). Rewritten to lead with the architectural gap. Priority set to high because this affects the autonomous planning pipeline reliability and will recur on other issues until fixed.