What problem does it solve? Long AI coding sessions suffer from "Context Rot" — quality degrades as the agent repeats failed debugging attempts, goes in circles, or loses track of decisions, and abrupt session termination can wipe out all progress. ## Core Features & Use Cases - 3-Strike Rule: After 3 failed debugging attempts on the same issue, the agent stops, documents what was tried, and recommends a fresh session instead of thrashing. - Circular Detection: Recognizes when the same approach is being retried and forces a fundamentally different strategy or a pause. - Auto-Save Protocol: Writes a state snapshot to .gsd/STATE.md immediately when a warning threshold is hit, before recommending /pause, so state survives hard session termination. - Use Case: During a long refactoring session, the agent fails to fix a test three times. The skill triggers a state dump capturing attempts, hypothesis, and next steps, then the user resumes later with /resume and full context. ## Quick Start Ask the agent to monitor this session for repeated failures and save a state dump to .gsd/STATE.md if debugging stalls.