The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the @anthropic-ai/claude-code package from the global NPM registry.
[PRIVILEGE_ESCALATION]: The instructions encourage using the --dangerously-skip-permissions flag, which explicitly bypasses interactive safety prompts for file writes and command execution. It also details the --add-dir flag, which expands the tool's access scope beyond the current working directory.
[COMMAND_EXECUTION]: The skill provides patterns for programmatic execution of shell commands and delegation of complex coding tasks to an agentic CLI, including automated pipelines using exec().
[INDIRECT_PROMPT_INJECTION]: The skill describes a tool that processes untrusted project data (code, files, metadata) and has significant capabilities to act on that data, creating a surface for indirect injection attacks.
Ingestion points: Reads project files (src/, TODO.md) and processes piped content (cat src/utils.ts | claude).
Boundary markers: None identified in the provided execution patterns to delimit untrusted project content from instructions.
Capability inventory: The tool possesses the ability to read files, edit code, and execute shell commands (e.g., via Bash tool mentioned in --allowedTools).
Sanitization: No explicit sanitization or input filtering is mentioned in the delegating instructions.

claude-code