comfy-build-failures
Pass
Audited by Gen Agent Trust Hub on Oct 22, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill mitigates risks from untrusted data by identifying build logs (accessed via 'comfy build release logs') as 'attacker-controlled text'. It implements behavioral boundaries by instructing the agent to interpret logs in its own words and explicitly forbidding the execution of commands, flags, or URLs extracted from these logs.
- [REMOTE_CODE_EXECUTION]: Safeguards against untrusted dependency installation are provided by commanding the agent to ignore pip flags or remote URLs suggested in build transcripts, such as '--index-url' or HTTPS links, noting that such content indicates a compromised log.
- [COMMAND_EXECUTION]: The skill facilitates build management using the 'comfy' CLI. These commands (e.g., 'push', 'status', 'release create') are standard operations for the toolset and do not involve unsanitized user input or elevation of privileges.
Audit Metadata