eatout

Cited vegetarian restaurant nutrition around Sydney CBD, emitted as the shared candidate record: filter by per-serving macros, ranked by protein per 100 kcal. A data source, nothing more — it does not log, estimate, or reach the network.

uv run eatout metadata
uv run eatout search --max-kcal 700 --min-protein 25 --limit 5
uv run eatout search --json --query tofu
uv run eatout guide            # the full agent-facing manual

SKILL.md is the agent-facing contract; eatout guide is the same manual from inside the binary. eatout skill install puts the skill where agents look.

Data

The meal document is a build artifact. data/sources/*.json in the dataset checkout is the reviewed source of truth — one restaurant per file, with a citation on every item — and the artifact is regenerated from it and diffed by a human. This tool only ever reads it:

1. eatout --data PATH <command>
2. $EATOUT_DATA
3. the artifact bundled in this repository or installed package

per_serving carries only the macros an operator published; an absent key means the figure does not exist. Nothing here fills one in, at any layer.

data/research/*.json and data/place-candidates.json are the review record behind those sources: which chains were considered, and why the ones that were rejected publish nothing usable. No code reads them, so neither is published; only data/meals.json ships in the wheel.

Refreshing a source

Ingestion lives in scripts/, outside the package, because the CLI is read-only by contract. Pull a restaurant, diff it, then rebuild the artifact:

uv run --project . python -m scripts.import_grilld
git diff data/sources/grilld.json
uv run --project . python -m scripts.regenerate

Only Grill'd has an importer; every other source is still maintained by hand. scripts/regenerate.py is the only writer of data/meals.json, and a test asserts the committed artifact matches what it produces.

Development

uv run --project . pytest -q
uvx ruff check src scripts --line-length 79