The Agent Skills Directory

[DATA_EXFILTRATION]: The skill accesses sensitive browser session data by reading from "~/.config/yt-dlp/cookies.txt". While used for authenticated access to social media platforms (Instagram, TikTok, X), this exposes the user's active login sessions to the agent's execution environment.
[COMMAND_EXECUTION]: The skill frequently executes shell commands and custom scripts (scripts/fetch_youtube.sh, scripts/fetch_twitter.sh). It invokes several binary tools including yt-dlp, ffmpeg, whisper, and defuddle to process downloaded content and extract metadata.
[EXTERNAL_DOWNLOADS]: The skill relies on several external dependencies and APIs. It requires yt-dlp, openai-whisper, ffmpeg, and defuddle. It also performs network requests to external services including the X/Twitter v2 API, Groq transcription API, and various user-provided URLs.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection.
Ingestion points: Untrusted data enters the agent context via scripts/fetch_youtube.sh (transcripts), scripts/fetch_twitter.sh (tweets), and the defuddle tool (web articles).
Boundary markers: The instructions do not define strict delimiters or "ignore instructions" warnings when processing the raw extracted text.
Capability inventory: The skill has access to shell execution (yt-dlp, ffmpeg), network operations (curl), and file system writes (saving extractions to .md files).
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from external sources before it is analyzed by the AI.

eat