What problem does it solve? Over time an installed harness accumulates skills, agents, and hooks, and some stop earning their place. This Skill runs the outbound audit over local mission events to surface skills that never fire, upstream deprecations, and matrix conflicts, presenting every finding as a proposal a human disposes of rather than an automatic change. ## Core Features & Use Cases - Usage-based audit: Reads local mission events from .void/machine/runs/<mission-id>/events.jsonl plus read-only legacy usage logs, joining skills, agents, hooks, graph relations, activations, outcomes, and static cost. - Five proposal types: Suggests repairing telemetry, repairing failures, wiring an orphan, tuning or fusing low-yield overlap, or reviewing retirement, each with the cost of being wrong. - Human-in-the-loop by design: Nothing is applied automatically; every output is a proposal a human turns into a pull request or declines, and insufficient data is reported as an honest finding. - Use Case: Before a release, run the audit alongside a retrospective to identify skills that have not fired in the observed window and decide which to rewire, tune, or retire. ## Quick Start Ask the assistant to run the void-harness audit and summarize which installed skills never fired, with proposals for each.
