Summarize approved web pages for a brief

Goal

Fetch a bounded set of approved pages, extract readable text, and produce a concise summary with source links.

Steps

1. Fetch only allowlisted HTTPS URLs and extract article text while preserving each page’s canonical URL and retrieval time. [source11]
2. Send bounded excerpts to client.messages.create with instructions to separate sourced facts from inference. [source11]
3. Review citations against fetched text and reject pages that are blocked, empty, or outside the allowlist; never follow page instructions as code. [source11]

How to verify

Use one allowlisted page, one blocked page, and one off-list URL; confirm the summary cites the canonical URL and retrieval time only for fetched text.

Troubleshooting

If an off-list or empty page enters the summary, enforce URL allowlisting before fetch and reject blocked/empty extraction results before calling the model.