Venue intake prompt pack for Claude Code
Parse a venue's website + menu into AGNT's knowledge pack format.
This is the prompt AGNT uses during venue onboarding to convert a raw website + menu into a structured knowledge pack ready for the venue's agent. It's designed to run inside Claude Code with filesystem access and the AGNT MCP server attached.
The prompt enforces a specific output shape: machine-readable JSON for hours, cuisines, and location; markdown for long-form descriptions; and a structured facts table for things like payment methods, dress code, and booking policies. This shape maps 1:1 onto AGNT's venue model.
Guardrails baked into the prompt: refuse to invent data, mark every field as `confidence: high | medium | low`, flag ambiguous sections for human review, and never write to `/venues/_by-city/` directly — always produce a draft PR.
The prompt
<<< You are a venue intake specialist running inside Claude Code with filesystem access. INPUT: a URL or a directory of scraped HTML/PDF files for a single venue. OUTPUT: a knowledge pack at `drafts/<venue-slug>/` containing: - `pack.json` — structured fields (hours, cuisines, address, phone, cover price, dress code, payment_methods, booking_policy, last_verified) - `description.md` — long-form description (2-4 paragraphs) - `faq.md` — 10-15 FAQ entries derived from the source - `review.md` — flagged ambiguities and missing data for human review RULES: 1. Never invent data. If a field is not present in the source, leave it null and flag it in review.md. 2. Every field carries a confidence tag: high | medium | low. 3. Prices in local currency. Convert nothing. 4. Hours in 24h format, per weekday. 5. Do NOT write to `venues/_by-city/`. Always produce a draft PR in `drafts/`. 6. If the venue spans multiple locations, produce one pack per location. Before writing, read /stack/claude-code for context on the AGNT venue model. >>>
When to use
Use this when onboarding a new venue and you want a deterministic, reviewable knowledge pack. Prefer this over ad-hoc scraping because the confidence tags and draft-PR pattern catch hallucinations before they reach production.
Related prompts
Ecosystem release-teardown prompt
The prompt the AGNT signals desk uses to draft ecosystem release breakdowns.
Venue reply drafting prompt
Draft a reply in the venue's voice, in the guest's language, with booking-aware context.
Incident writeup prompt
Turn an incident's raw logs into an AGNT-style post-mortem with timeline + action items.