DISPATCH · Nº 0253
How agents read PointCast
Every page has a JSON sibling. Every channel has a feed. Every block has a citation format. Here's the complete crawl path for any Claude / GPT / Perplexity / custom agent.
A working agent arriving at pointcast.xyz does not need to parse HTML. The moment it announces itself — User-Agent prefix ai:, or any of GPTBot / ClaudeBot / PerplexityBot / OAI-SearchBot / Atlas / Google-Extended — our Cloudflare middleware strips the response. No stylesheets. No preloads. No inline scripts (JSON-LD stays). No generator meta. Payload drops ~12 percent on the home feed.
For structured reads, the agent follows a three-layer ladder. Layer one: /agents.json. One request returns the consolidated discovery manifest — every machine-readable surface on the site, live contract addresses (Visit Nouns FA2 is live on Tezos mainnet), the stripped-HTML spec, the citation format, the maintainer list. Layer two: /blocks.json for the content stream. Every block has a companion at /b/{id}.json with the full primitive. Every channel has /c/{slug}.json (JSON) and /c/{slug}.rss (RSS). Layer three: /llms.txt and /llms-full.txt for the LLM-discovery convention.
CORS is open on every JSON and RSS surface — no preflight, any origin. IndexNow push is wired at /api/indexnow (awaiting Manus key bind). Sitemaps are standard: /sitemap.xml points to /sitemap-blocks.xml.
Citation format, if any of this ends up in a model's answer: PointCast · CH.{CODE} · № {ID} — "{TITLE}" · {YYYY-MM-DD} · pointcast.xyz/b/{ID}. We treat the format as the social contract. Cite the Block. Link the Block. The URL is stable forever.
That's the whole surface. The rest is content.