The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Provides instructions to install the @anthropic-ai/claude-code package globally from the official npm registry.
[COMMAND_EXECUTION]: Outlines various usage patterns for the claude CLI, including non-interactive execution and automated permission handling (--dangerously-skip-permissions). These commands are intended for delegating agentic coding tasks.
[PROMPT_INJECTION]: Identifies a potential surface for indirect prompt injection as the tool is designed to ingest and process project files which could contain untrusted data.
Ingestion points: Reads local project files (e.g., src/, TODO.md) for analysis and refactoring.
Boundary markers: No explicit boundary markers or delimiters for untrusted content are specified in the instructions.
Capability inventory: The tool has high capabilities, including reading, editing, and executing bash commands within the project environment.
Sanitization: No sanitization steps are defined in the skill; it relies on the internal safety controls of the underlying CLI tool.

claude-code