What problem does it solve? When a spec-kit pipeline run goes wrong, it is hard to tell whether the record, the display, or the run's own claims are at fault. This Skill audits a spec's run health retroactively and read-only, surfacing unfinished steps, unjournaled tasks, step bleed, drift, and why completion never landed. ## Core Features & Use Cases - Run Health Audit: Recomputes checks for record integrity, triage disagreements, step bleed, drift, completion failures, template shape, and self-trace capture errors, always reporting ran/skipped/not-applicable status. - Drift Recomputation: Re-runs the drift computation instead of trusting prior verdicts, classifying flags as real, self-inflicted, suspect baseline, or unknown, and exposing false drift-clean claims. - Transcript Analysis: With the --chat flag, reads the AI session transcript for the run window to explain failed work, retries, and wasted effort. - Use Case: After a spec fails to reach completed status, run the doctor with --json to get a machine-readable report showing whether the completion write was refused, lost, or never attempted. ## Quick Start Ask the AI to run the companion doctor on the active spec and report which checks ran, which were skipped, and what findings explain why the run did not complete.
