distill
distill
Post-mission audit and distillation for crew operations. Two modes:
- capture — reads a session JSONL + mission report, classifies failure incidents using the 8-code taxonomy, outputs a structured audit JSON co-located with the report
- analyze — reads all audit JSONs in
reports/, aggregates trends, produces a findings report and GitHub issues with recommendations
When to Use
- capture: after every mission (pass or fail) — triggered by the poller as a follow-up mission
- analyze: weekly via cron, or manually to survey trends across missions
- Skip capture if
{report-name}-audit.jsonalready exists — idempotent guard
Prerequisites
gh auth status # needed for analyze mode
python3 --version # needed for JSONL parsing
ls ~/.claude/projects/ # session JSONL files must be accessible
More from fellowship-dev/dogfooded-skills
setup-harness
Scaffold the knowledge layer for a repo — ARCHITECTURE.md, QUALITY_SCORE.md, enhanced docs/code-structure.md, docs/code-guidelines.md, and FlowChad flow stubs. Gives agents a map, not a 1,000-page manual.
8hookshot
Read a repo's docs/ structure and generate Claude Code enforcement hooks — pre-edit reminders, skill-drift warnings, and markdown lint. Warning and guidance only; never auto-edits files.
7spec-plan
Interview the user relentlessly about a plan or design until reaching shared understanding — walk each branch of the decision tree, resolve dependencies one by one. Use when stress-testing a plan, getting grilled on a design, or before writing a PRD.
6build-prd
Guide collaborative PRD creation from feature requests — 7-step workflow with structured discussion, GitHub integration, and quality gates. Use when creating PRDs, writing requirements, or transforming feature requests into implementation-ready issues.
6maintenance
Infra-only health audit — LaunchAgent health, cron log errors, system health, secrets scan, label sync, stale branches, dependabot, booster-pack sync, memory consolidation, open issues. Never fixes autonomously.
6visual-evidence
Capture Playwright screenshots and GIF recordings for PR evidence — before/after comparisons, feature demos, interaction bugs. Runs headless in any CI/remote environment.
5