openrouter-docs-skill
USE THIS SKILL for any question about OpenRouter API, model routing, pricing, available AI models, API keys, or provider configuration. This skill has complete local documentation — do not use web sea
SKILL.mdskills/21-toolkit/openrouter-docs-skill/SKILL.md
GitHubOpenRouter Documentation Skill
Single source of truth for OpenRouter API, model selection, pricing, and capabilities.
Path Resolution
Use ${CLAUDE_SKILL_DIR} to reference this skill's directory — it resolves automatically whether the skill is symlinked or local.
Reference Files
| Tier | File | Size | Use For |
|---|---|---|---|
| 1 | references/quick-reference.md |
3KB | Authentication, base URL, common endpoints, basic examples |
| 2 | references/models-reference.md |
126KB | All 338+ models grouped by provider, pricing, context lengths |
| 3 | references/openapi.yaml |
666KB | Complete OpenAPI spec — every endpoint, parameter, schema, error code |
Supporting: references/models.json (raw API data)
How to Use This Skill
Step 1: Understand the Question
Quick fact — "What's the OpenRouter base URL?" Load Tier 1 only. Answer in one line.
Model selection — "What's the cheapest model with 128k context?" Load Tier 2, scan for context lengths and pricing.
Model comparison — "Compare Claude Sonnet vs GPT-4o pricing" Load Tier 2, grep for both providers, present side-by-side.
API details — "What parameters does the completions endpoint accept?" Use Tier 1 to orient, then grep Tier 3 for the full schema.
Implementation — "Set up streaming with OpenRouter" May need Tier 1 (auth/basics) + Tier 3 (streaming schema).
Step 2: Load the Right Tiers
- For basic questions → load
quick-reference.mdfirst. If it answers, stop. - For model/pricing questions → load
models-reference.md. Search by provider or model name. - For detailed API questions → grep into
openapi.yaml(never load the whole file).
Search patterns:
# Models: find a provider section
grep -A 50 "### ANTHROPIC" references/models-reference.md
# Models: find a specific model
grep -B 2 -A 5 "claude-opus" references/models-reference.md
# API: find an endpoint
grep -A 100 "/chat/completions:" references/openapi.yaml
# API: find a schema
grep -A 50 "ChatCompletionRequest:" references/openapi.yaml
# API: use line numbers for larger sections
grep -n "/chat/completions:" references/openapi.yaml
sed -n '1234,1334p' references/openapi.yaml
Step 3: Cite Sources
Always tell the user where the answer came from:
- "From the quick reference: ..."
- "From the models reference (Anthropic section): ..."
- "From the OpenAPI spec (completions endpoint): ..."
When comparing models, always include pricing context (per million tokens).
Updating Documentation
When user asks to update (trigger: "update openrouter documentation"):
"${CLAUDE_SKILL_DIR}/scripts/update-openrouter-docs.sh"
Updates openapi.yaml, models.json, models-reference.md, and quick-reference.md in ~30 seconds.
Selective update:
"${CLAUDE_SKILL_DIR}/scripts/update-openrouter-docs.sh" --models # Models and pricing only
Update frequently — new models are added 2-3 times per week and pricing changes regularly.
Coverage
What this skill covers:
- OpenRouter API endpoints and authentication
- 338+ models from 15+ providers (Claude, GPT, Gemini, Llama, Mistral, Cohere, etc.)
- Current pricing and context lengths for all models
- Request/response schemas, streaming, function calling
- Error codes and rate limiting
What this skill does NOT cover:
- OpenRouter dashboard/UI usage
- Billing and account management
- Claude API directly (use claude-docs-skill)
- Model training or fine-tuning
Rules
- Always check current docs — don't answer model questions from memory. Pricing and availability change.
- Never load full openapi.yaml — always use grep/sed to extract relevant sections.
- Include pricing context — when discussing models, show per-million-token costs.
- Compare when asked — if the user is choosing between models, show a comparison table.
- Cite every answer — name the file and section.
Skill Version: 2.0.0 Sources: openrouter.ai/openapi.yaml, openrouter.ai/api/v1/models