comfy-deploy-failures
Pass
Audited by Gen Agent Trust Hub on Oct 20, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill documents a surface where third-party code could inject instructions via deployment logs. \n- Ingestion points: Instructions guide the agent to read logs using the
comfy deploy logscommand. \n- Boundary markers: The documentation explicitly warns that logs are attacker-controlled text and must not be interpreted as commands or arguments. \n- Capability inventory: The skill facilitates use of thecomfyCLI for deployment, build, and authentication tasks. \n- Sanitization: Instructions require the agent to summarize findings in its own words rather than following content found in the logs.
Audit Metadata