---

name: url-gen description: | Generate clean, human-facing URLs for fetch tools when external evidence is needed. version: 1.0.0 category: research tags:

• urls
• fetch
• discovery when_to_use:
• You need external evidence and must provide URLs for fetch tools
• You are forming a list of sources to retrieve with fetch_url_contents author: kdcube created: 2026-01-16

---

URL Generation

Overview

Generate URLs that are relevant and likely accessible for fetch tools. Prefer human-facing pages and avoid deep, speculative paths.

When to Use This Skill

• You need external evidence and must provide URLs for fetch tools.
• You are forming a list of sources to retrieve with fetch_url_contents.

Rules

1. Relevance

   • Only suggest URLs clearly relevant to the task.
   • Do not invent deep paths if you are unsure they exist.

2. Prefer human-facing pages

   • Prefer normal, human-facing pages over programmatic endpoints.
   • If multiple paths lead to the same info, prefer the one without segments like api, v1, v2, json, rest, graphql.
   • Example: prefer https://openai.com/pricing over https://openai.com/api/pricing.

3. Avoid machine-only endpoints unless requested

   • Do not suggest /api/, .json, .xml, or /graphql unless the user explicitly asks for APIs or raw data.

Hard Rule for fetch_context

• Generated URLs do NOT exist in context, so you MUST NOT put them in fetch_context.path.
• Put generated URLs directly into tool_call.params (e.g., {"urls": ["https://..."]}).
• fetch_context is only for reusing existing strings from context.