ai-infrastructure-modal
Pass
Audited by Gen Agent Trust Hub on Nov 4, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill provides templates for creating web endpoints and model inference services that ingest untrusted user input. These endpoints lack explicit boundary markers or sanitization logic, creating a surface for indirect prompt injection.\n
- Ingestion points: Inference functions in SKILL.md and examples/core.md (e.g., predict, classify, generate, chatCompletion).\n
- Boundary markers: None identified in code examples.\n
- Capability inventory: GPU inference compute, subprocess execution (vLLM server), and persistent volume storage.\n
- Sanitization: No sanitization or input validation logic is present in the examples.\n- [REMOTE_CODE_EXECUTION]: Documentation examples in the reference guide demonstrate the use of shell pipes to execute remote content during container image construction, such as fetching from example.com and piping to bash.\n- [COMMAND_EXECUTION]: The skill includes code patterns that spawn subprocesses to manage services, such as using subprocess.Popen to launch an OpenAI-compatible API server for model inference.\n- [EXTERNAL_DOWNLOADS]: The skill defines container build environments that fetch dependencies from standard package registries (PyPI, npm) and includes documentation examples for downloading external resources during the image build phase.
Audit Metadata