— Skill

Claude API skill

LLM integration answers rot in chat memory — this skill forces Claude to read current API/SDK guidance before it invents a model id, streaming shape, or cache flag.

What changes

Loads Anthropic’s Claude API / SDK reference skill (models, pricing, streaming, tools, MCP, agents, caching, token counting, migration) so answers track the skill’s live guides instead of stale recall.

When

Use Anthropic’s claude-api skill whenever the work is Claude / Anthropic API or SDK shaped — model choice, pricing, streaming, tool use, MCP, agents, prompt caching, token counting, or SDK upgrades — and wrong params cost real money or break production.

Good fits:

• Picking a current model id and request params for a new feature.
• Migrating SDK major versions or model families with the bundled guides.
• Cost-optimize / eval / hillclimb workflows that the skill routes into shared guides.
• Debugging streaming, tool calls, refusals, or token limits with the reference — not folklore.

Skip it for document editing (claude-pdf / docx / xlsx / pptx), MCP server scaffolding (claude-mcp-builder), or pure UI taste (frontend-design-taste).

How to start

Enable the claude-api plugin/skill from anthropics/skills. Attach before coding or answering Claude/Anthropic API questions; never answer model/pricing/limit questions from memory.

Attach the claude-api skill. Do NOT answer Anthropic/Claude API questions from memory.
Language: [Python | TypeScript | Java]. Goal: [implement | migrate | cost-optimize | build-eval | other].
Context: [file paths or paste]. Constraints: [budget / latency / tools / MCP].
Read the matching skill guides first. Quote current model ids and params from the skill.
If a subcommand fits (migrate, upgrade, cost-optimize, build-eval, hillclimb), run that workflow.
No invented beta headers or pricing.