YouTube Archiver

Archive YouTube playlists into markdown notes with metadata, transcripts, AI summaries, and tags. Use when a user asks to import/sync YouTube playlists, arch...

This skill appears to do what it says, but it legitimately needs sensitive access in two places: (1) browser cookie access to fetch private playlists (on macOS the terminal may need Full Disk Access to read Chrome cookies) and (2) optional API keys for summaries/tags (the skill will read the env var name you place in .config.json). Before installing or running: review the scripts locally (they are included), prefer exporting cookies to a cookies.txt file rather than granting broad Full Disk Access when possible, run a dry-run first, and only provide API env var names that you intend the skill to use (avoid reusing env var names holding unrelated secrets). If you want to avoid remote providers, set provider to "none" or use a local provider such as Ollama. If you are uncertain, run the scripts in an isolated environment (container or VM) and inspect network activity during an initial run.

Type: OpenClaw Skill Name: youtube-archiver Version: 1.0.0 The skill is classified as **benign**. The skill's purpose is to archive YouTube playlists, which inherently involves interacting with YouTube (potentially requiring authentication via browser cookies) and processing video data. The code uses `yt-dlp` for video metadata and transcript extraction, and `urllib.request` for interacting with various LLM providers for summarization and tagging. Key observations supporting this classification: 1. **Transparency and Stated Purpose:** Both `SKILL.md` and `README.md` clearly describe the skill's functionality, requirements (e.g., Python, `yt-dlp`), and potential system permissions needed (e.g., macOS Full Disk Access for browser cookies). The instructions for the AI agent in `SKILL.md` are direct and align with the stated purpose, showing no signs of prompt injection attempts to mislead the agent or exfiltrate data. 2. **Secure Command Execution:** The `yt_utils.py` script uses `subprocess.run` to execute external commands (`yt-dlp`, `summarize`). Crucially, these commands are consistently passed as a list of arguments (e.g., `["yt-dlp", "--cookies-from-browser", browser_name]`), which prevents shell injection vulnerabilities. 3. **API Key Handling:** LLM API keys are retrieved from environment variables (`os.environ.get`), which is a secure practice that avoids hardcoding sensitive credentials. 4. **Path and File Handling:** User-controlled output directories are processed with `Path(...).expanduser().resolve()`, and video titles used in filenames are sanitized with `sanitize_filename` to mitigate path traversal risks for file creation. 5. **No Malicious Intent:** There is no evidence of intentional data exfiltration (e.g., reading sensitive system files like `~/.ssh`, `~/.aws`, or `os.environ` and sending them to unauthorized endpoints or LLMs), unauthorized persistence (beyond the stated cron job automation for the skill's own scripts), or obfuscation. The LLM prompts are constructed from video transcripts, not sensitive system data. While the skill requires access to browser cookies (via `yt-dlp`) and executes external commands, these are necessary for its legitimate function and are implemented with reasonable security practices. The transparency about these capabilities and the lack of malicious intent make it benign.