minicpm5-deploy-ollama
Pass
Audited by Gen Agent Trust Hub on Aug 13, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill downloads the official installation script from ollama.com and pipes it to the shell. Ollama is a well-known and reputable service for local LLM execution, making this a standard and safe deployment step.
- [UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: It uses huggingface-cli to download model assets from Hugging Face. This is an industry-standard platform for hosting AI models, and the repository used belongs to the skill's author, openbmb.
- [INDIRECT_PROMPT_INJECTION]: The skill interpolates user-controlled variables (MODEL_NAME, GGUF_REPO) into shell commands, creating a potential attack surface. However, this is common in deployment templates and is assessed as safe in this context.
- Ingestion points: GGUF_REPO, QUANT, and MODEL_NAME variables in SKILL.md.
- Boundary markers: Absent.
- Capability inventory: Subprocess calls and file system operations via curl, huggingface-cli, and ollama in SKILL.md.
- Sanitization: Absent.
Audit Metadata