hrt-dark-mode-opt-in
Pass
Audited by Gen Agent Trust Hub on Oct 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill facilitates a transition to a high-autonomy state based on user-supplied triggers, bypassing standard confirmation gates.
- Ingestion point: Natural language request from the user for 'dark mode'.
- Boundary markers: The instructions mandate displaying a verbatim prototype risk warning before activation.
- Capability inventory: Enables the agent to proceed through proposal, specs, design, align, and apply phases without human-in-the-loop review.
- Sanitization: Requires explicit naming of the variant (planning-only or +implementation) and a confirmation response.
- [COMMAND_EXECUTION]: The '+implementation' variant permits the agent to automatically proceed to the code application phase without human review.
- Evidence: The skill creates a 'dark.md' file which acts as a signal for the agent to skip confirmation steps during the 'apply' phase, potentially leading to automated file system changes and command execution for code implementation.
Audit Metadata