cool-website
Pass
Audited by Gen Agent Trust Hub on Sep 26, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
execFileSyncin several Node.js scripts to perform system operations. scripts/doctor.mjsandscripts/encode.shsearch for and executeffmpegfrom various system paths or environment variables likeCOOL_WEBSITE_FFMPEGto validate the environment and encode video assets.scripts/doctor.mjsandscripts/kie.mjsexecutenodesub-processes to run helper scripts and probes.- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface as it integrates untrusted data from multiple sources into its operational context.
- Ingestion points: The skill ingests user-provided interview answers (defined in Step 0) and external keyword/market data fetched via the DataForSEO MCP (defined in Step 0a and Step 3 of the SEO workflow).
- Boundary markers: There are no explicit security boundary markers or specific 'ignore embedded instructions' delimiters used when interpolating this external data into the website structure or generation prompts.
- Capability inventory: The skill possesses significant capabilities including shell command execution (
ffmpeg), file system writes (fs.writeFileSyncfor HTML, CSS, JS, and asset creation), and network access (fetchcalls to video and SEO APIs). - Sanitization: The documentation does not specify rigorous sanitization or filtering logic for data retrieved from external SEO queries before it is used to influence the final page content and metadata.
- [EXTERNAL_DOWNLOADS]: The skill fetches content and interacts with external AI services to generate video assets.
scripts/kie.mjssends base64-encoded local image data tokieai.redpandaai.coandapi.kie.aifor video generation tasks.- The skill also supports the official Higgsfield REST API at
api.higgsfield.aifor similar purposes. - These operations are documented as part of the core functionality for producing cinematic scroll-sites.
Audit Metadata