The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it reads and processes external data that could contain malicious instructions.\n
Ingestion points: Video transcripts read from libraries/[library-name]/transcripts/*.json (as described in roughcut_agent_prompt.md and direct_path.md) and user-provided scripts that are either pasted into chat or read from local files (as described in direct_path.md).\n
Boundary markers: Absent. The instructions do not define delimiters or provide specific warnings to the agent to disregard instructions embedded within transcripts or scripts.\n
Capability inventory: The agent has the ability to execute shell commands via local ruby scripts, copy files to the user's Desktop, and open system applications.\n
Sanitization: Absent. There is no evidence of filtering or validation for the content of transcripts or scripts before they are processed by the agent.\n- [COMMAND_EXECUTION]: The skill utilizes shell commands for its core functionality, including library management, metadata extraction, and file export. Specifically, it executes local ruby scripts, uses rg (ripgrep) for searching transcripts, and uses open -a to launch external video editing software.\n- [DATA_EXFILTRATION]: The skill includes functionality to copy exported XML files to the ~/Desktop/ directory. While this is a user-configurable convenience feature, it involves interacting with a sensitive user directory through shell commands (cp).

cut