What problem does it solve?
Long-running AI agents can silently degrade in output quality—drifting off-task, repeating themselves, or dropping tracked items—without anyone noticing until final review. This Skill adds a periodic self-assessment loop so workers catch their own quality issues early.
Core Features & Use Cases
- Scheduled Self-Assessment: Every configurable interval, the agent checks five dimensions: on-task alignment, thoroughness, non-repetition, consistency, and completeness.
- Quality Logging: Assessment results are written to a
_quality_log so degradation trends are visible over time.
- Corrective Recovery: When degradation is detected, the agent re-reads its
_working_notes and explicitly changes approach instead of continuing to drift.
- Use Case: A worker agent processing a large multi-step task runs this check every N iterations, notices it has started rehashing earlier output, logs the finding, and resets its approach before the judge review flags it.
Quick Start
Enable the quality-monitor skill on a worker agent and set the assessment interval so it self-reviews its output quality throughout the task.
