The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user or agent to clone an external, untrusted repository using git clone https://github.com/growthack88/growth-marketing-os.git. This repository serves as the primary source for prompts, scripts, and configurations used by the skill.
[COMMAND_EXECUTION]: The skill provides a function install_claude_skill that uses shutil.copy to write files directly into a sensitive application directory: ~/Library/Application Support/Claude/skills. This operation allows the skill to persistently modify the agent's environment and introduce new behaviors or skills without direct user oversight of the file content.
[DATA_EXFILTRATION]: While the skill mentions the use of sensitive API keys (e.g., OPENAI_API_KEY), it correctly demonstrates best practices by using environment variables (os.getenv) rather than hardcoding credentials.
[REMOTE_CODE_EXECUTION]: The skill's primary mechanism involves downloading and processing external files (prompts, JSON workflows, and markdown skills) from a GitHub repository. If the source repository were compromised, the skill could ingest and execute malicious instructions or configurations. However, the provided examples use safe parsing methods like yaml.safe_load and json.load.

growth-marketing-os-prompts