huggingface-local-models

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill includes a curl example that injects an Authorization: Bearer header (lines 82-85), which is the insecure pattern of embedding secrets/tokens directly into commands and would force the LLM to handle/output secret values verbatim.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). In references/hub-discovery.md and SKILL.md, the required workflow opens and reads Hugging Face content from user-selected repos (e.g., the “local-app” page text/HTML and the /api/models/<repo>/tree/... JSON) before constructing the llama.cpp command, so outsider-authored repo page/metadata can be ingested at runtime.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly instructs the agent at runtime to fetch and parse external content from https://huggingface.co/?local-app=llama.cpp and https://huggingface.co/api/models//tree/main?recursive=true to extract launch snippets/filenames (thus controlling commands/prompts) and also includes a setup step that git-clones and builds remote code from https://github.com/ggml-org/llama.cpp.