chengfeng-videocut-workbench
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill facilitates interaction with a local command-line interface tool named
workbenchto perform video editing tasks. It provides structured command templates (e.g.,["workbench", "workflow-get", "--json"]) and guides the agent on how to locate and verify the runtime executable through version checks rather than arbitrary disk scanning.\n- [INDIRECT_PROMPT_INJECTION]: The skill ingests user input and project data to construct editing commands, creating a potential attack surface. To mitigate this, the instructions mandate the verification of unique identifiers (such aswordIdandclipId) and require the agent to perform read-back operations to confirm project state after every modification.\n- [SAFE]: The instructions incorporate multiple security constraints, including restricting API origins to local loopback addresses (127.0.0.1or[::1]), explicitly forbidding the use ofnpxfor dynamic tool acquisition, and instructing the agent to cease operations if required capabilities or commands are missing.
Audit Metadata