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
Risk Level
SAFE
Analyzed
Aug 1, 2026, 05:34 AM
Security Audit — agent-trust-hub — using-model-endpoint