The Agent Skills Directory

[COMMAND_EXECUTION]: The script audiosnap/audiosnap-wrapper.sh contains a hardcoded absolute file path (/Users/han1/clawd/audiosnap/.build/release/audiosnap). This reveals the author's local username and will likely cause execution failures on other systems.
[COMMAND_EXECUTION]: The audiosnap/audiosnap-wrapper.sh tool utilizes AppleScript via osascript to execute commands inside the Terminal.app process. This is a deliberate technique to bypass macOS Transparency, Consent, and Control (TCC) permissions for Screen Recording by inheriting the permissions granted to the Terminal application.
[EXTERNAL_DOWNLOADS]: The audiosnap/podsnap.py script executes the yt-dlp utility with the --remote-components "ejs:github" flag. This configuration allows the tool to download and execute remote components or plugins from GitHub at runtime, which is an unverified external code loading mechanism.
[PROMPT_INJECTION]: The skill contains a vulnerability surface for indirect prompt injection via the podsnap.py transcription tool.
Ingestion points: Processes untrusted audio and video content from external URLs (YouTube, Bilibili, Xiaoyuzhou).
Boundary markers: None; transcribed text is returned as raw output to the agent.
Capability inventory: The skill possesses extensive capabilities, including system audio control, file system modification, and network access across multiple platforms.
Sanitization: No filtering or sanitization is performed on the transcribed content to prevent the agent from executing instructions found within the processed media.

ears