What problem does it solve? Long agent sessions silently degrade as the context window fills with stale file dumps, abandoned approaches, and failed corrections, causing the model to drop earlier constraints without any error. This Skill provides a discipline for treating context as a managed budget so quality does not rot invisibly across turns. ## Core Features & Use Cases - Four context moves: Clear between unrelated tasks, reset after two failed corrections, compact long coherent threads with a named focus, and delegate heavy multi-file investigation to fresh-context subagents that return only conclusions. - Anti-context-rot persistence: Routes durable state (plans, checkpoints, ADRs, provider state) to the filesystem so work survives any session reset. - Signal-driven action table: Maps symptoms like forgotten constraints, correction loops, and widening searches to the correct remediation move. - Use Case: During a long debugging session the agent starts re-asking settled questions and contradicting earlier decisions; apply the two-correction reset, compact with a focus, and push further log reading into a subagent to restore reasoning quality. ## Quick Start Ask the agent to review the current session for context-rot signals and apply the appropriate clear, compact, or subagent delegation move before continuing the task.
