The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill provides instructions to download installation scripts directly from claude.ai, which is the official domain for the Claude service.
Evidence: curl -fsSL https://claude.ai/install.sh | bash and irm https://claude.ai/install.ps1 | iex.
[REMOTE_CODE_EXECUTION]: The installation process involves piping a remote script to a shell interpreter (bash/PowerShell). While this pattern is often associated with risk, in this context, it is the documented and official method for installing the developer's CLI tool.
[COMMAND_EXECUTION]: The skill facilitates the execution of the claude CLI tool using the bash tool with pty:true. It allows for one-shot prompts that can interact with the local file system and git environment.
[INDIRECT_PROMPT_INJECTION]: The skill operates by reading project files and a CLAUDE.md file, which creates an attack surface for indirect prompt injection if those files contain malicious instructions intended to influence the agent's behavior.
Ingestion points: Project root directory, src/ directory, and the CLAUDE.md configuration file.
Boundary markers: Uses the -p flag for one-shot prompts to scope individual tasks.
Capability inventory: The CLI tool has the capability to edit files, execute shell commands, and automate git workflows (commit/push).
Sanitization: The skill documentation does not specify sanitization for project file content, relying on the internal safety filters of the Claude Code CLI tool.

claude-code