The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests YouTube transcripts through various methods including yt-dlp and the Apify API. Since this data originates from an external, user-controlled source (YouTube video captions), it presents a surface for indirect prompt injection where malicious instructions within a transcript could attempt to influence the agent's content generation or subsequent actions. No explicit boundary markers or sanitization logic for the LLM prompt interpolation were identified in the provided templates.
[COMMAND_EXECUTION]: The skill invokes several shell commands and Node.js scripts to perform its pipeline steps. This includes using yt-dlp for video metadata extraction and custom scripts in the scripts/ directory. Several of these scripts, such as scripts/fetch-transcript.js and scripts/post-thread.js, utilize dynamic path resolution via environment variables to locate and require Node.js modules, which is a common but sensitive pattern for resolving local dependencies.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to fetch transcripts from Apify's API and YouTube's servers. Additionally, the flyer generation scripts utilize Playwright, a tool that typically manages the download and execution of browser binaries (Chromium) to render HTML templates into images.
[DATA_EXFILTRATION]: The skill is designed to transmit generated content (blog posts, tweets, LinkedIn posts) and related metadata to external platforms including Supabase and Twitter. These operations are core features of the skill's content distribution pipeline and occur based on user configuration and triggers.
[CREDENTIALS_UNSAFE]: The skill manages several sensitive credentials, including the Supabase Service Role Key, Apify API Token, and Twitter API keys. The instructions and scripts facilitate passing these secrets to external services via environment variables in shell commands and HTTP headers.

content-cascade