using-model-endpoint
Pass
Audited by Gen Agent Trust Hub on Aug 1, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: No malicious patterns or security risks were identified. The code demonstrates a high degree of security awareness, including defensive measures for environment variable handling and credential scrubbing.
- [CREDENTIALS_UNSAFE]: While the skill manages API keys, it does so securely. The apply_auth function in provider.py enforces strict naming rules and a denylist for environment variables, preventing attackers from overwriting critical system variables like PATH or LD_PRELOAD to achieve code injection or privilege escalation.
- [DATA_EXFILTRATION]: The skill includes a dedicated token_scrub_regex to identify and redact sensitive tokens (such as nvapi-) from the output, mitigating the risk of accidental data exposure during model interactions.
- [PROMPT_INJECTION]: The skill handles responses from remote model endpoints, which presents a surface for indirect prompt injection. 1. Ingestion points: HTTP response data received from BASE_URL using httpx or requests. 2. Boundary markers: None identified. 3. Capability inventory: Network egress is limited to the configured BASE_URL; standard Python execution is available within the sandboxed kernel. 4. Sanitization: Implements token scrubbing for specific API key formats in the provider shim.
Audit Metadata