demo-director
Pass
Audited by Gen Agent Trust Hub on Sep 28, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The
scripts/demo_checks.pyscript executes theffmpegandffprobecommands viasubprocess.runto validate the technical properties of generated video files. These calls are limited to verification tasks and use a list-based invocation method that prevents shell injection. Additionally, the skill utilizes system tools likecaffeinateto manage power states during recording, which is done through explicit user-approved sessions. - [DYNAMIC_EXECUTION]: The skill provides an optional native macOS adapter in
scripts/macos.swift. Users are instructed to compile this source code locally using theswiftccompiler to enable platform-specific UI automation and recording capabilities. This is a documented, opt-in component that does not execute external or unverified code. - [INDIRECT_PROMPT_INJECTION]: The skill processes external data such as application windows and browser URLs. It mitigates the risk of processing malicious content by enforcing strict manual approval gates for the interview, the timed script, and a short preflight recording before any full automation or capture is performed.
- [SAFE]: The skill includes system discovery tools like
scripts/host_inventory.pyand thelistcommand in the macOS adapter. These tools are strictly gated by user consent and follow a least-privilege approach by explicitly omitting sensitive identifiers like hostnames, usernames, and network addresses, and they do not perform any network exfiltration.
Audit Metadata