What problem does it solve? When an agent or developer keeps retrying materially equivalent changes under the same hypothesis without gaining new evidence, work stalls in a failure loop. This Skill interrupts that cycle, reconstructs what each attempt actually established, and returns a decision-ready recovery state instead of another equivalent mutation. ## Core Features & Use Cases - Stagnation Detection: Classifies work as stalled only when attempts are materially equivalent, the hypothesis is unchanged, and no decision-relevant evidence is emerging, never based on attempt count alone. - Attempt-Evidence Reconstruction: Rebuilds each attempt as hypothesis, action, result, evidence gained, and remaining mutation, keeping observed facts, reported results, inference, and unknowns distinct. - Recovery Decision Model: Returns one of four states: Not stalled, Blocked, Diagnostic next (one discriminating checkpoint), or Diversify (handoff to structural candidate search via explore-decision-space). - Use Case: Two nil-guard edits to a handler fail with the same stack trace. The Skill pauses a third guard, reconstructs both attempts under the unchanged handler-input hypothesis, and recommends a single read-only checkpoint at the decoder entry with outcome-dependent next decisions. ## Quick Start Ask the agent to pause the repeated attempts, reconstruct what each attempt established, and tell you the recovery state and next checkpoint without making any changes.
