# Austin Happy Hour Database - Agent Guidance

This file is for agents reading `https://austinhappyhourdatabase.com`.

## Purpose

Austin Happy Hour Database is a public directory of Austin happy-hour listings. The public website is intended to be discoverable by humans and agents for reading, citation, and navigation.

## Canonical Surfaces

- Public website: `https://austinhappyhourdatabase.com`
- Sitemap: `https://austinhappyhourdatabase.com/sitemap.xml`
- LLM index: `https://austinhappyhourdatabase.com/llms.txt`
- Agent rules: `https://austinhappyhourdatabase.com/.well-known/agent-rules`
- OpenAPI document: `https://austinhappyhourdatabase.com/api/v1/openapi.json`
- Public API base: `https://austinhappyhourdatabase.com/api/v1`
- Public MCP endpoint: `https://mcp.austinhappyhourdatabase.com/`
- MCP server card: `https://austinhappyhourdatabase.com/.well-known/mcp-server-card.json`

Use public listing URLs as canonical citations. Listing pages currently use `/happy-hour/<slug>` URLs.
The home page and public listing pages support markdown content negotiation when a request sends `Accept: text/markdown`.

## Allowed Public Use

- Read the public website and sitemap.
- Summarize or cite public listing pages.
- Link users to the relevant public listing page.
- Cache public page content for normal search, indexing, and retrieval workflows.
- Use the credential-free, read-only `/api/v1` resources for structured discovery.
- Use the public MCP endpoint for read-only listing discovery and citation.
- Use public page content as real-time agent input or retrieval context.

## Boundaries

- Do not treat `https://happyhourdatabase.com` as a public client API surface. It is the private CMS origin; the supported public API is `https://austinhappyhourdatabase.com/api/v1`.
- Do not call Payload CMS, admin, or API routes unless you have separate authorization from the site owner.
- Do not attempt to authenticate to `/admin` or infer private CMS data.
- Do not use the public MCP endpoint for writes. It is read-only and exposes no Payload admin or review state.
- Do not use public site content for AI model training or fine-tuning.
- Verify time-sensitive venue details with the venue source before presenting them as current fact.

## Crawl Guidance

- Prefer the sitemap over broad crawling.
- Keep request rates modest. One request per second per client is a reasonable default unless separately authorized.
- Respect `robots.txt` and the boundaries in this file.
- The robots content-signal policy is `search=yes`, `ai-input=yes`, `ai-train=no`.

## Public API Guidance

- Start with `GET /api/v1/meta` to record `apiVersion`, `dataVersion`, and `listingCount`.
- Use `GET /api/v1/happy-hours` for filtered discovery, `/map` for lightweight geocoded points, `/facets` for counts, and `/{idOrSlug}` for one listing.
- Follow `page.nextCursor` with the exact original query. Restart without a cursor after `409 CURSOR_STALE`.
- Send `If-None-Match` with a prior ETag to avoid retransferring unchanged data.
- Times are 24-hour Austin local time (`America/Chicago`). Starts are inclusive, ends are exclusive, and `endsNextDay` makes overnight windows explicit.
- `generatedAt` and listing `revision` identify projected content. They do not claim that a venue's offer was recently re-verified.
- No API key is required. The public API permits cross-origin reads and supports `GET` and `HEAD` only.

## Data Notes

Happy-hour menus, prices, days, and times can change. Treat the public site as a directory and discovery aid, not as a guarantee that a venue's current offer is unchanged.