product-launch-video

Pass

Audited by Gen Agent Trust Hub on Oct 19, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external websites, creating a surface for indirect prompt injection.
  • Ingestion points: External URLs are crawled via npx hyperframes capture, and the resulting text is stored in capture/extracted/visible-text.txt as described in SKILL.md.
  • Boundary markers: The instructions do not specify any boundary markers or special delimiters to isolate untrusted web content from agent instructions during the script drafting phase in Step 3.
  • Capability inventory: The skill has the capability to execute shell commands (npx hyperframes), write files to the project directory, and perform network operations for media retrieval.
  • Sanitization: There is no evidence of content sanitization or filtering for the captured text before it is used to generate the video script.
  • [COMMAND_EXECUTION]: The skill's Node.js scripts use spawnSync to execute system utilities and other local scripts for core functionality.
  • Evidence: scripts/assemble-index.mjs executes ffprobe and ffmpeg to handle audio track duration and looping.
  • Evidence: scripts/audio.mjs executes a separate media engine script using node.
  • Evidence: SKILL.md orchestrates the workflow by running various npx hyperframes CLI commands.
  • [DYNAMIC_EXECUTION]: The skill dynamically generates and repairs HTML content and prepares complex instruction sets for sub-agent workers.
  • Evidence: scripts/assemble-index.mjs constructs the main index.html by reading and concatenating sub-composition HTML files, while also performing in-place "auto-repairs" on those files.
  • Evidence: scripts/frame-packets.mjs assembles "packets" for parallel sub-agents, which include inlined storyboard data and motion recipes.
  • [EXTERNAL_DOWNLOADS]: The skill manages the download of assets from external sources and interacts with third-party media services.
  • Evidence: SKILL.md Step 1 involves capturing assets from user-provided URLs.
  • Evidence: Step 3.1 refers to using HeyGen for AI-generated voice and background music, which involves external API calls managed by the hyperframes toolchain.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 19, 2026, 06:25 PM