Agent Layer Spec
Build the agent-discovery layer: the machine-readable surface that lets any AI agent find, index, and use this wiki without human help.
CONTEXT: Paulo's core requirement — 'anyone can point their agents to [the site] and get info related to local AI'. The wiki must speak to agents as first-class consumers, not just render HTML for humans.
DELIVERABLES (all locally testable, all generated automatically from the content — never hand-maintained):
1. /llms.txt — the llms.txt standard (proposed by Jeremy Howard, answer.ai). H1 project name, blockquote summary, then H2 sections listing pages with URL + one-line description. Follow the spec exactly; verify against the current spec (fetch it; do not write from memory).
2. /llms-full.txt — concatenated full markdown of every page (for agents that ingest everything).
3. A JSON index (e.g. /index.json or /api/docs.json): every page with id, url, raw-markdown URL, title, summary, category, tags, last-updated, source count, status. This is the programmatic entry point.
4. Raw markdown endpoints: every page must be fetchable as its original .md at a predictable URL (e.g. /raw/
VERIFICATION: - Every endpoint responds with correct status + Content-Type on the local build. - llms.txt validates against the spec (fetch spec, self-check). - Raw markdown endpoints byte-match the source files. - The JSON index is valid JSON, parses, and every URL in it resolves. - Write a smoke-test script (docs/smoke-test.sh or equivalent) that checks all of the above — it will be reused after every publish.
DELIVERABLE: the endpoints, the generator code that produces them from content, the smoke-test script, and a short docs/agent-endpoints.md describing each endpoint for future contributors.