Guides and prompts
Copyable prompts for finding, reading, connecting, and reporting on site content.
Each guide starts from an outcome, not a Tool name. Copy the prompt, replace the brackets, and keep the completion criterion with it.
Find and read the strongest pages
Use this when the answer should come from Colin’s published work rather than from the model’s memory.
Use the lateano.com MCP server only for site claims.
Goal: answer [question].
Search: [topic or phrase], across all fields.
Read: the strongest [two] matching pages in Markdown.
Return: a concise answer with each source title and canonical URL. Label any inference.
If more pages matched than the Tool returned, say so.The agent normally calls search_content, then get_page_markdown for the
selected routes. It may call get_page_info when relationships affect the
answer.
Done: every claim about the site is grounded in a page the agent read, and every cited URL came from a Tool result.
Explore one subject across the site
Use this when you want the site’s own vocabulary before choosing search words.
Map what lateano.com publishes about [subject].
First inspect the site's tags and choose the closest real tag slug; do not invent one.
Search that tag across every collection.
For each result, report its collection, title, path, and one-sentence relevance.
Read the best page, inspect its relationships, then choose and read one returned
related path. Explain how the two pages connect.The agent calls list_tags, hands one returned slug to search_content, and
reads the best result with get_page_markdown. It then calls get_page_info
and reads one returned relationship.
Done: the search uses a slug returned by the server, accounts for every returned collection, and grounds the connection in two pages it read.
Find documentation or another standing page
Use this when the page may sit outside the searchable collections.
Find the lateano.com page that explains [site capability or convention].
List the pages outside the searchable collections, select the closest description,
read that page in Markdown, and answer from it. Return the page title and path.The agent calls list_pages, then get_page_markdown. This is the branch for
developer documentation, About, and index pages; search_content cannot return
them.
Done: the selected path came from list_pages, and the answer comes from
the Markdown at that path.
Preview and submit feedback
Use this when something is missing, broken, confusing, inaccurate, or worth reporting for another reason.
Prepare feedback for lateano.com about [route]: [message].
Choose one kind: missing, broken, confusing, inaccurate, or other.
Choose severity: high, medium, or low.
Show the exact route, message, kind, and severity before changing anything.
Wait for my approval. After approval, generate a new UUID, submit once, and return
the recorded route and receipt time. Reuse the UUID only if that same call may not
have arrived.The preview is the boundary. submit_feedback runs only after approval, and a
changed message gets a new idempotency key.
Done: the user approved the exact fields, the Tool returned receivedAt,
and the agent reported what was recorded without promising a reply.
The Tool reference carries every argument, result shape, handoff, and limit used by these guides.