best-practices
Pass
Audited by Gen Agent Trust Hub on Oct 26, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill features an 'Adaptive Mode' (detailed in references/adaptive-mode.md) which requires the agent to read and analyze 5-15 files from a user's existing codebase to learn local naming and structural conventions. ● Ingestion points: Representative project scripts including Scripts, LocalScripts, and ModuleScripts. ● Boundary markers: Absent; the agent is instructed to match the detected style directly without explicit delimiters or safety warnings for the sampled content. ● Capability inventory: The agent has extensive file manipulation and arbitrary code execution capabilities via the Studio MCP tools (multi_edit, execute_luau). ● Sanitization: No sanitization or filtering of the analyzed code is specified.
- [REMOTE_CODE_EXECUTION]: Installation instructions for the Rokit toolchain (in references/external-editors.md) include commands for piping remote scripts from GitHub into bash and Invoke-Expression.
- [EXTERNAL_DOWNLOADS]: The skill references and provides instructions for installing various third-party development tools and libraries from external sources, including GitHub repositories and the npm registry.
Audit Metadata