product-demo-local-video

Pass

Audited by Gen Agent Trust Hub on Oct 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external application interfaces and chat logs, which could theoretically contain instructions designed to influence the agent's behavior during transcription or video synthesis.
  • Ingestion points: The skill captures real-app UI footage and two-account communication exchanges (found in SKILL.md and references/immersive-walkthroughs.md).
  • Boundary markers: The instructions require the agent to transcribe actual speech and compare it against expected dialogue to detect drifts or invented content, and use contact sheets for visual verification (references/bunko-native-dialogue.md).
  • Capability inventory: The agent is authorized to execute video rendering and publishing tools such as lazyedit-publish-workflow based on the ingested data (SKILL.md).
  • Sanitization: The skill provides explicit mandates to protect private accounts, reuse protected credential files with minimal access, and inspect all outputs for secrets or unrelated third-party data (references/immersive-walkthroughs.md).
  • [COMMAND_EXECUTION]: The skill relies on the execution of local command-line tools to perform its primary video processing and publishing functions.
  • Evidence: The skill instructs the agent to use lazyedit-publish-workflow with various flags (e.g., --languages, --no-portrait-blur-fill, --no-publish) and interact with the LocalVideoGen engine parser (SKILL.md and references/bunko-native-dialogue.md).
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 19, 2026, 06:25 PM