dont-make-me-think
Pass
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructs the agent to execute several shell commands, including environment checks, git synchronization operations (fetch and rebase), and the execution of a local Python script for image analysis.
- [EXTERNAL_DOWNLOADS]: The 'Dependency Preflight' section of the instructions suggests downloading and installing external tools, including the 'browse' skill from a third-party GitHub repository (garrytan/gstack) and the 'agent-skill-manager' package via npm.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted content from live websites and UI screenshots, creating a vulnerability surface for indirect prompt injection.
- Ingestion points: The skill reads data from live URLs (via the browse tool), screenshot analysis results, and raw HTML/CSS/JS code provided as input.
- Boundary markers: There are no specific delimiters or instructions defined to prevent the agent from following malicious commands embedded within the audited UI content.
- Capability inventory: The skill includes high-privilege capabilities such as modifying local source code (Redesign Mode), performing git operations, and accessing the network through the browse tool.
- Sanitization: No explicit sanitization or filtering of the external content is implemented before it is analyzed by the agent.
- [DYNAMIC_EXECUTION]: The skill executes 'scripts/process_screenshots.py' as part of the image processing workflow. While this is a local script included in the skill package, it represents a pattern of executing code that acts on untrusted image input.
Audit Metadata