launch-kit
Pass
Audited by Gen Agent Trust Hub on Oct 18, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and act upon instructions contained within local project files, creating a surface for potential indirect prompt injection.\n
- Ingestion points: The agent reads instructions and configurations from BRAND.md, LAUNCH.md, actions.json, and project-specific markdown recipes found in the workspace.\n
- Boundary markers: No specific delimiters or safety warnings for embedded content are defined in the instructions.\n
- Capability inventory: The skill utilizes npm for package installation and the vos CLI for video processing, recording, and network-based publishing.\n
- Sanitization: The skill does not describe explicit sanitization or validation mechanisms for the content read from these markdown and JSON files.\n- [EXTERNAL_DOWNLOADS]: The skill installs and uses a specific CLI tool from the vendor's ecosystem.\n
- Evidence: The instructions include 'npm i -D @vosjs/cli' to install the necessary pipeline tools. This is a vendor-owned resource consistent with the skill's authorship.\n- [DATA_EXPOSURE]: The skill includes functionality to publish user interaction data as part of its deployment flow.\n
- Evidence: The vos push command includes the --prompt flag which uploads the user's raw prompt to a watch page. The skill mitigates this by instructing the agent to inform the user of this behavior and providing guidance on how to manage the visibility of the published content.
Audit Metadata