Claude API Builder

Use when code imports Anthropic SDKs or the user asks to build with the Claude API, Agent SDK, tool use, streaming, files, or batches.

Security Scan

Instruction-only skill focused on Anthropic/Claude API guidance; no credentials, installs, or surprising I/O are requested, though there are minor provenance/name inconsistencies to be aware of.

Purpose & Capability

The skill's name, description, and SKILL.md consistently target Anthropic/Claude API guidance. No binaries, env vars, or config paths are requested, which is proportionate for an advisory skill. Minor inconsistency: a bundled file is named agents/openai.yaml despite the skill being Claude-focused, and provenance references a source module (src/skills/bundled/claudeApi.ts) even though no code file is included; these are likely bookkeeping/naming issues rather than functional red flags.

Instruction Scope

SKILL.md stays on-topic: it describes when to invoke the skill, what inputs to expect, and what outputs to produce. It does not instruct the agent to read system files, environment variables, or transmit data to arbitrary endpoints. It does encourage consulting official docs (i.e., 'browse live docs'), which may imply use of the agent's web/browse tool but is not an instruction to exfiltrate data.

Install Mechanism

There is no install spec and no code files to install. Instruction-only skills have the lowest install risk and nothing will be written to disk by the skill itself.

Credentials

The skill declares no required environment variables, credentials, or config paths. This is appropriate for a documentation/advisory skill. Note: real usage implementing Anthropic APIs will require the user's Anthropic API key, but the skill itself does not request or require it in its metadata or instructions.

Persistence & Privilege

The skill does not force persistence (always: false) and does not request elevated platform privileges. Autonomous invocation is allowed by default (disable-model-invocation: false), which is normal; nothing in the skill claims or requires permanent/privileged presence.

Assessment

This is an instruction-only helper for building with Anthropic/Claude APIs: it doesn't request credentials or install code and is internally consistent for that purpose. Two minor oddities to note before installing: a bundled file is named agents/openai.yaml (name mismatch) and the provenance text references a source module even though no source file is included — these look like metadata/naming issues, not active threats. If you plan to implement running code based on the guidance, you'll need to supply your Anthropic API key at runtime; only provide such keys when you trust the runtime and the agent's tooling. Also consider whether you want the agent to browse the web for live docs (it may require enabling a browsing tool); review platform policies for granting browsing or outbound network abilities to agent-invoked skills.

Use this skill when the task is specifically about Anthropic or Claude API integration.

Trigger Signals

• Imports like anthropic, @anthropic-ai/sdk, or Agent SDK packages
• Requests about streaming, tool use, prompt caching, file uploads, or Claude batch jobs
• Questions about choosing Anthropic API patterns for an app

Workflow

1. Detect the project language and SDK surface.
2. Choose the smallest fitting pattern: simple messages, streaming, tools, batches, or files.
3. Prefer official docs and SDK-native examples.
4. Keep model names, error handling, and retries explicit.
5. Explain when to browse live docs for fast-moving API details.

Guardrails

• Do not pivot to non-Anthropic SDKs unless the user asks.
• Prefer primary documentation over memory for model/API specifics.
• Separate one-shot prompts from stateful agent workflows.

Example Requests

• Add Anthropic streaming support to this TypeScript app.
• Help me implement tool use with the Python Anthropic SDK.

Inputs

• Project language
• Anthropic API feature needed
• Performance or UX constraints

Outputs

• Recommended API pattern
• Implementation guidance
• SDK-specific caveats

Success Criteria

• The guidance matches Anthropic APIs specifically.
• The selected pattern fits the use case and language.
• Fast-moving API details are called out clearly.

Non-Goals

• General LLM vendor-neutral advice
• Non-Anthropic SDK design by default

Source Provenance

Derived from src/skills/bundled/claudeApi.ts and the bundled Claude API references.