web_search

# web_search — look it up through the engine `sc search` is the one web search verb every shell has, on every harness. It posts your query to the engine API with your own bearer token; the host calls Tavily with the instance's API key and returns the results. The key never reaches a shell, and a sandboxed shell needs no network egress. ## When to search - A fact that changes: a library's current API, a release note, a CLI flag, a version's known bug, an error message you cannot explain from the code. - Before guessing at an external service's protocol or a package's behaviour. - Never for anything the repo, its catalogue, or your own memory already answers. ## The verb ```bash sc search "<query>" # 5 results + a short synthesized answer sc search "<query>" --max 10 # 1..20 results sc search "<query>" --depth advanced # deeper crawl, slower sc search "<query>" --json # raw payload: answer, results[] (title, url, snippet, score) ``` Output is a numbered list: title, URL, snippet. Treat snippets as leads, not proof — open the URL that matters (`curl -sL <url>` or your harness's fetch tool) before you rely on it, and cite the URL in what you write. ## When it fails Every failure names its cause and carries no secret: | Message starts with | Meaning | Do | |---|---|---| | `web search is not configured` | no key on this instance | Tell the FnB: set it in the GUI → **Scripts → Web Search**. Do not work around it with a ...