ollama-model-agents
Pass
Audited by Gen Agent Trust Hub on Jun 13, 2026
Risk Level: SAFECOMMAND_EXECUTIONCREDENTIALS_UNSAFEPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill facilitates the use of local and subscription-based CLI tools (ollama, gemini, and claude) to process model requests. These are standard operational commands within the skill's defined scope.
- [CREDENTIALS_UNSAFE]: The instructions mandate unsetting provider API keys (e.g., ANTHROPIC_API_KEY, OPENAI_API_KEY) from the environment using 'env -u' before execution. This is a positive security measure to prevent accidental credential leakage to model backends.
- [PROMPT_INJECTION]: The skill processes untrusted input via shell variables in delegated model calls.
- Ingestion points: Untrusted data enters the agent context through the $PROMPT variable in SKILL.md.
- Boundary markers: Explicit boundary markers or delimiters for the prompt variable are absent in the provided command templates.
- Capability inventory: The skill utilizes subprocess calls to execute CLI tools; specifically, the claude CLI is configured with Read, Grep, Glob, and Bash tools enabled.
- Sanitization: While no specific character sanitization is provided, the skill enforces mitigation by requiring '--approval-mode plan' and '--permission-mode plan' for all automated tool executions, ensuring a human review checkpoint.
Audit Metadata