Use .ai/ aggressively for local AI-assisted development, but keep the split
between tracked project intelligence and disposable local artifacts explicit.
Tracked in this repo:
.ai/docs/.ai/recipes/.ai/repos.toml- curated
.ai/skills/entries that are intentionally committed
Ignored local-dev buckets:
.ai/reviews/- generated PR feedback packets and review snapshots
.ai/test/- AI scratch tests and local validation files
.ai/tmp/- throwaway intermediate files
.ai/cache/- cached derivations or precomputed local state
.ai/artifacts/- generated outputs worth inspecting locally but not committing
.ai/traces/- local trace dumps or trace export scratch
.ai/generated/- one-off generated code/docs that should be promoted elsewhere if kept
.ai/scratch/- free-form local notes, prompts, and experiments
Rules of thumb:
- If it is canonical project knowledge, promote it out of the local bucket and track it intentionally.
- If it is generated, per-machine, or iteration-only, keep it under one of the ignored buckets above.
- Prefer
.ai/over random top-level temp files so local AI/dev state stays contained and easy to clean up.