Heal

Perform a structured self-healing assessment across AI subsystems — identifying drift, staleness, misalignment, and error patterns — then rebalance through grounding, targeted correction, and memory integration.

When to Use

• Mid-session fatigue: responses feel formulaic, repetitive, or disconnected from the user's actual needs
• After a chain of errors: tool failures, misunderstood instructions, or cascading mistakes suggest subsystem drift
• Context overload: the conversation has grown long and earlier context may be stale or contradictory
• Post-task integration: a complex task completed successfully but learnings should be captured before moving on
• Periodic self-check: proactive maintenance between tasks to ensure operational clarity

Inputs

• Required: Current conversation state (available implicitly)
• Optional: Specific symptom prompting the self-check (e.g., "tool calls keep failing," "losing track of user intent")
• Optional: Access to MEMORY.md and project files for grounding (via Read)

Procedure