Youtube Summary

Summarize any YouTube video by dropping the link in chat. Supports custom prompts — paste the URL followed by your instructions (e.g. 'focus on the technical...

This skill appears coherent and is likely safe to install for its stated purpose, but consider the following before using it: - TranscriptAPI.com will receive the video identifier and return transcripts; ensure you trust that service and understand its privacy/cost model (the README mentions a $5/mo plan). Transcripts (which may contain copyrighted or private content) are printed as JSON and then sent to your agent/LLM for summarization — avoid feeding sensitive videos unless you accept that exposure. - Store TRANSCRIPT_API_KEY securely (using the suggested 'pass' workflow or environment variable). The skill warns environment variables can be visible; using --api-key-file with a temp file reduces exposure in process listings. - Install Python deps in a virtual environment (pip install -r requirements.txt). The code calls the 'yt-dlp' binary as a fallback; ensure yt-dlp is available in your environment if you want metadata fallback to work. - The code defaults to English for transcript language detection; summaries will follow that value. If you need stronger guarantees about language detection or transcript integrity, review the extract.py implementation. - The SKILL.md prohibits fallback web_search and forbids fabricating transcripts; those are good safety constraints — verify the agent follows them. If you need anything changed (e.g., explicit declaration of yt-dlp as a required binary, different transcript provider, or not transmitting full transcripts to the model), ask the skill author or audit the runtime to implement that behavior.

Type: OpenClaw Skill Name: yt-summary Version: 1.3.2 The skill is classified as suspicious primarily due to its explicit support for 'custom prompts' which allows direct prompt injection against the underlying Large Language Model (LLM). While this is an advertised feature for customizing summaries, it represents a significant risk where a user could instruct the LLM to generate harmful content or attempt to bypass its safety mechanisms. Additionally, the `SKILL.md` file instructs the agent to execute shell commands with user-controlled input (`YOUTUBE_URL_OR_ID`). Although the `scripts/extract.py` and `scripts/utils.py` files implement robust sanitization (via `extract_video_id`) to mitigate shell injection risks, the pattern of passing unsanitized user input directly into a shell command is a design that could be vulnerable if the sanitization were ever weakened or bypassed. There is no evidence of intentional malicious behavior such as data exfiltration or backdoors.