comfy-custom-nodes
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill design involves processing potentially untrusted data from external sources, which is a known vulnerability surface for indirect prompt injection attacks.\n
- Ingestion points: Technical knowledge slices retrieved via the
comfycommand and general code-search results (SKILL.md).\n - Boundary markers: The skill does not define strict delimiters for external content but does include a specific instructional guardrail advising the agent to treat search results as untrusted references rather than authoritative data.\n
- Capability inventory: The agent is empowered to perform code generation, fixing, and validation tasks, which could be exploited if malicious instructions are successfully injected through the documentation or search results.\n
- Sanitization: No sanitization, filtering, or structured schema validation is specified for the external data ingested by the agent.
Audit Metadata