fetching-from-the-web

What problem does it solve? When an AI agent fetches web content, it often gets summarized text instead of exact bytes, hits a 403 from a publisher, or runs into a sandbox egress block — and then wastes effort retrying or routing around policy boundaries. This Skill defines the correct behavior for each of these situations. ## Core Features & Use Cases - Exact-byte retrieval: When the precise content matters, fetch with curl into the scratchpad and read from disk instead of relying on a summarizing fetch tool. - 403 handling: Do not retry or try sibling URLs; attribute search snippets to the publisher and mark them for re-verification. - Egress block handling: Treat EGRESS_BLOCKED as a domain-wide policy boundary — never route around it with alternate runners or CI probes, and mark gaps as needing a human or an unblocked environment. - Use Case: While researching documentation, your WebFetch call returns a 403. Instead of retrying mirrors, you attribute the snippet to the publisher and flag it for later verification. ## Quick Start Before running any WebFetch or curl command, load this Skill and follow its rules for exact-byte fetches, 403 responses, and egress blocks.