GAP-TOOLS-037: Fetch Content Processing

Description

CC's WebFetchTool requires a prompt parameter that processes and summarizes fetched web content before returning it. Babysitter's fetch tool returns raw HTTP responses. Different content processing models.

Current State

Babysitter's fetch tool params: url, timeout, raw. Makes an HTTP GET request and returns { status, statusText, body }. Body is truncated at 50KB unless raw: true. No content processing, summarization, or extraction.

CC's WebFetchTool params: url (required), prompt (required -- describes how to process the fetched content). CC fetches the URL, then applies the prompt to extract or summarize relevant information before returning it. This reduces token usage by filtering irrelevant content.

Target State

Add a prompt parameter to the fetch agentic tool:

an LLM call (or simpler extraction logic) using the prompt as instructions. Example: prompt: "Extract all API endpoint URLs" would fetch a page and return only the extracted URLs.

• **prompt**: Optional string. When provided, fetched content is processed through

When prompt is omitted, current raw-response behavior is preserved.

Implementation options: 1. **LLM-based**: Route fetched content + prompt through a lightweight model call. Expensive but matches CC's behavior exactly. 2. **Extraction-based**: Use the prompt as a pattern/instruction for rule-based extraction (regex, CSS selectors, text filters). Cheaper but less flexible. 3. **Hybrid**: Try extraction first, fall back to LLM for complex prompts.

Dependencies

• None for raw param addition.
• LLM-based processing depends on model access within agentic tool execution context.

Key Files

Recommendation

Phase 3. Low priority -- babysitter's raw fetch is often sufficient since the LLM processing the response can extract what it needs. The prompt param matters more in token-constrained scenarios.