# T-04: Discovery & AI Access Layer

## T-04: Discovery & AI Access Layer

<div id="bkmrk-w3c-dcat-v3-%28data-ca" style="border-left:4px solid #1a7f64;padding:14px 18px;margin:12px 0;
background:#f9fbfa;border-radius:0 8px 8px 0">**W3C DCAT v3 (Data Catalog Vocabulary)**  
Dataset catalogue standard. CTH publishes a DCAT v3 catalogue at `/api/catalogue` listing all available climate datasets, their access conditions, formats, and ODRL usage policies. Registered with Google Dataset Search.</div><div id="bkmrk-openapi-3.1full-api-" style="border-left:4px solid #1a6b8a;padding:14px 18px;margin:12px 0;
background:#f9fbfa;border-radius:0 8px 8px 0">**OpenAPI 3.1**  
Full API specification at `/api/openapi.json`. All endpoints documented with: security schemes (DID-based JWT), request/response schemas (JSON-LD), error codes, policy evaluation requirements. Enables direct agent integration without documentation scraping.</div><div id="bkmrk-llms.txtmachine-read" style="border-left:4px solid #b35c00;padding:14px 18px;margin:12px 0;
background:#f9fbfa;border-radius:0 8px 8px 0">**llms.txt**  
Machine-readable index at `https://wiki.cleantechhub.net/llms.txt` (and `/llms-full.txt`). Lists all framework pages with their purpose, so AI agents can navigate the wiki without embedding the full content. Updated automatically when framework pages change.</div><div id="bkmrk-json-ld-contextevery" style="border-left:4px solid #555;padding:14px 18px;margin:12px 0;
background:#f9fbfa;border-radius:0 8px 8px 0">**JSON-LD Context**  
Every framework concept (role, rule, credential type, standard) has a JSON-LD term at `https://vocab.cleantechhub.net/climate/`. Enables semantic interoperability with other climate data systems (e.g. UNTP registry, EU Commission linked data).</div>### Agent Access Pattern

Recommended pattern for an AI agent integrating with the framework:

1. Fetch `/llms.txt` to identify relevant framework sections
2. Read specific wiki pages via BookStack API (read-only, no credential required for public pages)
3. Request a delegated `cth:AgentCredential` from the delegating human's session
4. Call `POST /policy/evaluate` with the intended action
5. If `allow: true`, proceed with the action using the policy token
6. PROV-O record is auto-generated by the API gateway (no agent action required)