Args:
    url: The URL to fetch.
    mode: "markdown" returns clean readable content (default).
          "raw" also includes the raw HTML alongside the markdown.

Returns a JSON object with the extracted content plus `confidence`,
`method` (which tier/strategy produced it), and `tier` — always
check `confidence` before trusting the result for anything important.
On failure, `failure_reason` explains why (never a silent empty result).

Args:
    url: The URL to extract structured data from.
    json_schema: A JSON Schema (as a JSON string) describing the fields
        to extract, e.g. '{"type":"object","properties":{"price":
        {"type":"number"},"title":{"type":"string"}}}'.
    instructions: Optional extra guidance for the extraction model.

Requires ANTHROPIC_API_KEY to be set in the environment — this tool
will return a diagnosable failure (not a crash) if it isn't.

Args:
    recipe_id: e.g. "ats_greenhouse", "ats_lever", "ats_ashby".
    slug: The platform-specific identifier (e.g. a Greenhouse board
        token — the part of boards.greenhouse.io/<slug>).
    entity_name: A human-readable name to attach to results (e.g. the
        company name), since the raw API responses often don't include it.