nemo-fabric-build-adapter

Pass

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill references and recommends fetching documentation, integration examples, and JSON schemas from NVIDIA's official GitHub repositories (e.g., github.com/NVIDIA/NeMo-Fabric).
  • [INDIRECT_PROMPT_INJECTION]: The skill defines an architecture for adapters that process potentially untrusted external data, such as agent configurations and run requests.
  • Ingestion points: Data enters the system via the AgentConfig and AgentRunRequest models within the start and invoke lifecycle methods in SKILL.md.
  • Boundary markers: The instructions mandate strict validation of untrusted values against official JSON schemas and note that the common host decodes transport envelopes.
  • Capability inventory: Adapters built using this skill execute target-native code (e.g., self.target.run) and utilize the NeMo Fabric SDK for lifecycle management.
  • Sanitization: Developers are explicitly instructed to reject unsupported values with safe error codes and are forbidden from logging credentials, environment variables, or arbitrary user input.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 10, 2026, 08:22 PM