YouTube Model Feeder

Food for your model — extract transcripts, key frames, OCR, slides, and LLM summaries from YouTube videos into structured AI-ready knowledge.

This skill appears to do what it says (extracts transcripts, frames, OCR, and summaries), but it relies on cloning and running a GitHub repo via docker-compose which will download and execute code and create persistent services. Before installing: 1) Inspect the repository and docker-compose.yml (images used, volume mounts, environment variables, network settings). 2) Confirm Docker images come from trusted registries and do not mount sensitive host paths. 3) Check where the service stores the Fernet key and any API keys; avoid supplying high-privilege credentials until you review storage/encryption. 4) Run initially in an isolated environment (VM or dedicated host) and restrict network access (firewall) to avoid unintended exposure. 5) Prefer using local-only providers (Ollama) if you want to avoid sending data/keys to external LLM services. If you cannot inspect the repo or are uncomfortable with these risks, do not run docker-compose for this project.

Type: OpenClaw Skill Name: youtube-model-feeder Version: 1.0.1 The skill bundle is classified as suspicious primarily because the `SKILL.md` file contains an explicit instruction to bypass VirusTotal security warnings using a `--force` flag, which is a common social engineering tactic used to evade security controls. While the stated purpose of extracting YouTube data is plausible, the documentation describes a complex architecture that handles sensitive LLM API keys and relies on external code hosted at `github.com/celstnblacc/youtube-model-feeder`, which is not included in the bundle for review. Furthermore, the `_meta.json` file contains a future-dated publication timestamp (March 2026) and version inconsistencies that are anomalous.