hatch3r-context-health

To prevent task drift during long agentic coding sessions, assess context health continuously using a five-check degradation model that monitors task recall, progress clarity, and error patterns to enforce corrective actions at Green/Yellow/Orange/Red thresholds.

Context poisoning in agent orchestration refers to outdated file content, stale assumptions, contradictory context, and accumulated errors causing confident-but-wrong outputs. It is detected by flagging these specific degradation signals during long multi-turn development sessions.

Handle repeated tool failures and stale file knowledge in a coding agent by applying corrective action routing, which refreshes context through re-reading and re-planning, delegates work to a sub-agent with a handoff, or checkpoints and stops with a progress postback.

Use a context health assessment for ambiguous or irreversible coding tasks when you experience many conversation turns, repeated tool failures, or stale file knowledge where delegation to a sub-agent or a complete session restart is needed to prevent wrong decisions.

Yes, you can use model-aware profiles from configuration to enforce corrective actions based on Green, Yellow, Orange, and Red context degradation thresholds, ensuring long development sessions remain accurate and on track.

The best way to recover from silent drift in multi-turn codebase interactions is to evaluate scope alignment and file currency, then apply corrective actions like context refreshing or agent handoff to prevent incorrect decisions and maintain codebase awareness.