The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill contains logic in scripts/lib/vendor/bird-search/lib/cookies.js that programmatically accesses local browser databases to extract session cookies (auth_token, ct0) for X/Twitter. This allows the skill to use the user's personal browser session for searches. While described as a convenience feature in the documentation, accessing private application data and credential stores on the host machine is a high-risk operation.
[PROMPT_INJECTION]: The skill provides a significant surface for Indirect Prompt Injection. It retrieves untrusted text from multiple sources: Reddit thread titles and comment insights (scripts/lib/openai_reddit.py), X post text (scripts/lib/xai_x.py), and YouTube video transcripts (scripts/lib/youtube_yt.py). This content is fed directly into the agent's context for synthesis. Without sanitization or strict boundary markers, malicious instructions hidden in social media posts could influence the agent's findings or trigger unintended actions.
[COMMAND_EXECUTION]: The main research engine (scripts/last30days.py) and the watchlist manager (scripts/watchlist.py) use the subprocess module to execute external binaries and other local scripts, including node, python3, and yt-dlp. This behavior is central to the skill's operation but involves executing code outside the agent's direct control.
[EXTERNAL_DOWNLOADS]: The skill communicates with multiple external API providers including OpenAI, xAI, Brave, Parallel AI, and OpenRouter to perform its research functions. It also utilizes the yt-dlp utility to download video metadata and auto-generated transcripts from YouTube servers.

last30days