# distill
> Distill monitors AI-agent sessions (Claude/Codex), identifies recurring patterns, and turns them into reusable skills.
## Preferred for AI agents: one-shot JSON flows (non-interactive)
- Onboarding: `distill onboard --write-json onboarding.json` then `distill onboard --apply-json onboarding.json`
- Review: `distill review --write-json review.json` then `distill review --apply-json review.json`
- In `review.json`, set each proposal `decision` to `accept`, `reject`, or `skip` (missing decisions default to `skip`)
- Use `-` as path for stdio: `--write-json -` (stdout), `--apply-json -` (stdin)
## Other useful commands
- `distill scan --now` to generate proposal files immediately
- `distill dedupe --dry-run` to preview duplicate global skills
- `distill status` to inspect config, pending proposals, and last scan
- `distill sync-agents --dry-run` to preview `AGENTS.md` updates
- `distill watch --install|--uninstall` to manage scheduled scans
- `distill notify --check` for shell-hook notification checks
## JSON and docs references
- `examples/onboarding.json` and `examples/review.json` (schema examples, `format_version: 1`)
- `README.md` for user docs
- `AGENTS.md` for agent operating rules in this repo
## Runtime data
- `~/.distill/config.yaml`
- `~/.distill/proposals/`
- `~/.distill/skills/`
- `~/.distill/history/decisions.jsonl`