The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill includes logic in scripts/lib/setup_wizard.py to automatically install the yt-dlp binary using Homebrew (brew install yt-dlp) during its first-run setup. This is a well-known, legitimate tool used for YouTube data extraction.
[COMMAND_EXECUTION]: The skill frequently uses the subprocess module to invoke system tools and its own scripts. Notable executions include running node for X/Twitter searches, yt-dlp for YouTube transcripts, and the security CLI on macOS for Keychain interactions. It implements safety measures for user-provided configurations, such as regex validation for SSH host aliases used in traffic routing.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from the public internet (Reddit, X, YouTube transcripts). It explicitly identifies this risk in its instructions and uses specific boundary markers like <untrusted_content> tags, along with safety notices, to prevent the agent from following instructions embedded in the scraped data.
[CREDENTIALS_SAFE]: The skill handles sensitive authentication data by reading browser cookies and macOS Keychain items. These operations are restricted to its stated purpose of authenticating with services like X/Twitter and Truth Social. The codebase includes logic to warn users if secret files have insecure permissions (e.g., advising chmod 600).

last30days