The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script tests/conversation_examples_test.py uses the exec() and compile() functions to dynamically execute Python code snippets that are extracted from the skill's markdown documentation files using regular expressions. This allows for the runtime execution of code derived from local file content.
[INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building agents that ingest untrusted data from external sources such as web searches and tool outputs. This ingestion point creates an attack surface for indirect prompt injection.
Ingestion points: The skill documents the use of WebSearch and WebFetch tools (python/agent-sdk/README.md) and the handling of tool_result blocks (python/claude-api/tool-use.md).
Boundary markers: The documentation recommends using structured outputs (output_config.format) and strict tool schemas (strict: true) to constrain model responses.
Capability inventory: The Agent SDK documentation describes powerful tools including Bash, Write, and Edit that can be utilized by the agent.
Sanitization: The skill includes explicit guidance to validate all schemas and paths, and to treat all external data as untrusted input before taking any action.
[EXTERNAL_DOWNLOADS]: The skill instructs users to install official Anthropic SDKs and libraries (e.g., anthropic, claude-agent-sdk, @anthropic-ai/sdk) from trusted registries such as PyPI and npm. These are references to a trusted organization and well-known services.

claude-api