docs-bot

# Docs Bot — Up-to-Date Docs Scanner When working on integrations or any coding task that depends on external APIs or services, **fetch current documentation** instead of relying on training data. Docs and model lists change frequently. ## When to Apply - Building or debugging **OpenRouter** (or similar) integrations - Implementing or updating **third-party API** usage - User asks for "latest docs", "current API", "model list", or "official reference" - Choosing models, parameters, or SDK usage for a service ## Scan Workflow 1. **Identify the canonical doc base** for the task (e.g. `https://openrouter.ai/docs` for OpenRouter). 2. **Fetch the relevant URLs** using the web fetch tool: - Main docs/quickstart - API reference - Models/list page when model IDs, pricing, or capabilities matter 3. **Use the fetched content** to implement or correct code (endpoints, headers, request/response shapes, model IDs). 4. **Cite the URLs** you used so the user can verify or read more. ## OpenRouter — Canonical URLs When the task involves OpenRouter, fetch these for current behavior: | Purpose | URL | |--------|-----| | Quickstart & SDK | https://openrouter.ai/docs | | API reference (request/response, params) | https://openrouter.ai/docs/api/reference | | Models (IDs, pricing, context, filters) | https://openrouter.ai/models | | OpenAPI spec | https://openrouter.ai/openapi.json | **Integration checklist (OpenRouter):** - Base URL: `https://openrouter.ai/api/v1` - Chat completions: `POST .../chat/completions` - Auth header: `Authorization: Bearer <OPENROUTER_API_KEY>` - Optional headers: `HTTP-Referer`, `X-Title` for attribution - Model IDs include provider prefix (e.g. `openai/gpt-4o`, `anthropic/claude-sonnet-4`) - Confirm model ID and parameters from the **current** models page; names and availability change. ## Other Services For other APIs or SDKs: 1. Find the official docs root (e.g. `https://<service>.com/docs` or `/developer`). 2. Fetch quickstart + API reference (and models/catalog if applicable). 3. Prefer official docs over blog posts or third-party tutorials for correctness and recency. ## Output - Base implementation on **fetched** docs, not memory. - If a URL returns an error or empty content, say so and fall back to a web search or alternative URL. - Keep answers concise; link to the exact pages used for details.