first-table-nz · git:20260719.1748a98 · 2026-07-19 · sha256 c061894f60bea57f
first-table-nz git:20260719.1748a98A
Immutable. This exact content never changes and is served at /api/v1/blob/c061894f60bea57f.
--- name: first-table-nz description: "Query First Table NZ restaurants, 50%-off early-table deals, city/suburb discovery, cuisine search, and availability slots. Read-only; no booking, login, or payment." license: MIT compatibility: "Requires Python 3.10+ and network access for live data" metadata: thecolab.category: "events-and-media" thecolab.source_owner: "First Table" thecolab.source_type: "commercial" thecolab.auth: "none" thecolab.access_mode: "html-readonly" thecolab.data_class: "public" thecolab.writes: "false" thecolab.browser: "false" thecolab.risk: "low" thecolab.cache_ttl: "24h" thecolab.schema_version: "1" thecolab.skill_type: "html-readonly" thecolab.pack: "nz-commercial-web" thecolab.source_url: "https://www.firsttable.co.nz" thecolab.allowed_domains: "stellate.firsttable.net,www.firsttable.co.nz" thecolab.last_verified: "2026-07-19" thecolab.health: "healthy" thecolab.maintainer: "@adam91holt" --- # First Table NZ ## Goal Query live First Table NZ restaurant and availability information through a deterministic Python CLI using public website data and unauthenticated GraphQL calls observed from the frontend. ## Use this when - A user asks which First Table restaurants are available in a NZ city or suburb - A user wants to search First Table restaurants by cuisine, suburb, or name - A user wants restaurant detail, ratings, session types, diner limits, or booking-fee snapshots - A workflow needs machine-readable First Table NZ restaurant or availability data ## Do not use this for - Booking, checkout, payment, account signup, login, favourites, cancellations, or member actions - First Table countries outside New Zealand unless the CLI is explicitly extended and verified for that host - Claiming guaranteed table availability without rechecking live data immediately before use - Redistributing First Table data as a bulk dataset ## Preferred workflow 1. Run `scripts/cli.py` with the narrowest subcommand that answers the task 2. Use `city` to inspect city metadata, suburbs, and tags before a targeted search 3. Use `search` for restaurant discovery and `detail` when an exact First Table restaurant id is known 4. Use `availability` for read-only slot checks; never attempt booking mutations 5. Use `--json` for agent chaining, comparisons, or structured reports 6. Mention that data is a live, unofficial public snapshot from First Table NZ ## CLI Run with: ```bash python3 skills/first-table-nz/scripts/cli.py <command> [flags] ``` ### Commands - `cities [--json]` — list discoverable NZ First Table city/region pages - `city [city-slug] [--json]` — inspect a city/region page, suburbs, tags, sessions, and counts - `search [query] [--city slug] [--session breakfast|lunch|dinner|dinner2|lasttable] [--date YYYY-MM-DD] [--people N] [--limit N] [--json]` — find restaurants in a city - `detail <restaurant-id> [--json]` — fetch restaurant detail by First Table id - `availability <id...> [--date YYYY-MM-DD] [--people N] [--available-only] [--limit N] [--json]` — fetch read-only public slots for one or more restaurant ids Examples: ```bash python3 skills/first-table-nz/scripts/cli.py city auckland --json python3 skills/first-table-nz/scripts/cli.py search sushi --city auckland --limit 5 python3 skills/first-table-nz/scripts/cli.py detail 1142 --json python3 skills/first-table-nz/scripts/cli.py availability 1142,1698 --date 2026-05-25 --people 2 --available-only --json ``` ## Resources - CLI entrypoint: `scripts/cli.py` - Live smoke test: `scripts/smoke_test.py` - API and stability notes: `references/api-notes.md` ## Notes - No API key, username, password, account cookie, or browser automation is required for the read-only commands - The CLI uses Python stdlib only - The upstream GraphQL and frontend SSR shapes are unofficial and can change; verify with live smoke tests before large workflows - Availability responses can include cached or stale status fields from the upstream service; treat them as source metadata, not an agent-side guarantee