hyperframes-creative
Warn
Audited by Gen Agent Trust Hub on Nov 14, 2026
Risk Level: MEDIUMREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: The skill uses a custom utility in
scripts/package-loader.mjsto automatically install missing Node.js dependencies (such assharpand@hyperframes/producer) usingnpm installat runtime. While it includes the--ignore-scriptsflag to prevent post-install scripts from running, this pattern still introduces external code into the execution environment dynamically. - [COMMAND_EXECUTION]:
- The script
scripts/extract-audio-data.pyusessubprocess.runto execute theffmpegcommand-line tool for audio processing. - The documentation in
references/design-picker.mdinstructs the AI agent to start a background HTTP server usingpython3 -m http.server. This exposes the project directory on the local network, which may be unintended depending on the deployment environment. - [DYNAMIC_EXECUTION]: The
scripts/package-loader.mjsutility usesspawnSyncto re-spawn the current Node.js process with a modified environment (augmentedPATH) after completing its automated dependency installation. - [INDIRECT_PROMPT_INJECTION]: The skill's primary workflow, described in
references/prompt-expansion.md, involves ingesting untrusted user prompts and expanding them into detailed scene-by-scene instructions for sub-agents. - Ingestion points: Untrusted user input enters the context during the initial project prompt stage.
- Boundary markers: The instructions do not specify the use of delimiters or warnings to ignore embedded instructions within user content.
- Capability inventory: The skill can execute shell commands, manage its own Node.js environment, and start network servers.
- Sanitization: There are no explicit sanitization or validation steps before user input is expanded and processed by other automated tasks.
Audit Metadata