claude-agent-sdklisted

# Claude Agent SDK ## When to activate Building a Python or TypeScript application that uses Claude Code capabilities programmatically; deploying Claude as an autonomous agent inside a product; writing code that drives the `claude` CLI in non-interactive mode; scripting agentic workflows that need tool calls, retries, and context management handled automatically. ## When NOT to use Using Claude Code interactively in the terminal — that is the default experience, not an SDK use case; building a simple chatbot or single-turn Q&A interface (use the Messages API directly); when Anthropic Managed Agents is a better fit (hosted infrastructure, automatic scaling, built-in memory persistence). ## Instructions **What the Agent SDK is:** Same tool loop, context management, and agent capabilities as interactive Claude Code — packaged as a library you embed in your own application. You control the infrastructure; Anthropic provides the model and agent loop. **SDK vs alternatives — choose the right layer:** | Need | Use | |---|---| | Embed agentic Claude in your app, own the infra | Agent SDK | | Agentic Claude hosted by Anthropic, hands-off ops | Managed Agents | | Single-turn responses, no tool loop needed | Messages API | | Interactive terminal workflow | Claude Code CLI | **Installation:** Python: ```bash pip install claude-code-sdk ``` TypeScript: ```bash npm install @anthropic-ai/claude-code ``` **`--bare` flag via options:** Skips `CLAUDE.md` loading and MCP server discov