The Agent Skills Directory

[SAFE]: The skill functions as a documentation library for interacting with well-known AI CLI tools such as Claude, Gemini, and OpenAI Codex. No malicious code or hidden commands were found.
[SAFE]: Includes explicit security instructions for handling untrusted data, recommending the use of shell heredocs and temporary files to prevent command injection when passing prompts to subprocesses.
[INDIRECT_PROMPT_INJECTION]: The skill manages a surface for indirect injection as it facilitates sending external data to AI agents with tool-use capabilities.
Ingestion points: Prompts are passed to the {prompt} placeholder in various CLI command templates (e.g., in references/claude.md).
Boundary markers: The SKILL.md file instructs on using heredocs to isolate the prompt content from the shell command.
Capability inventory: The documented CLIs are capable of file system access and tool execution.
Sanitization: The skill mitigates risks by instructing users to avoid shell metacharacters and to use restricted modes like --mode ask or --mode plan to disable destructive tools.
[SAFE]: All external documentation links refer to well-known service providers or official technology project pages.

invoke-agent