scaffold
Pass
Audited by Gen Agent Trust Hub on Nov 17, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes a 'Plan' containing architecture decisions (Workflow Step 1). While this represents a potential injection surface, the skill implements strong mitigations: it requires a mandatory validation phase and a manual human-in-the-loop confirmation gate ('ask user question tool') before generating any code or executing tasks based on the plan data.
- [DYNAMIC_EXECUTION]: The primary purpose of the skill is to generate project scaffolding (abstractions, infrastructure, and examples) at runtime. To mitigate risks associated with generated code, the skill enforces strict isolation: all examples must reside in a dedicated 'examples/' directory, use a specific '.example.' file naming convention, and be explicitly excluded from production builds.
- [COMMAND_EXECUTION]: The skill performs local command execution for verification and environment analysis. This includes executing project-configured build or type-check commands during the 'Verification Gate' and running a local script ('node .claude/scripts/ai-signal-scan.cjs') to detect AI-specific configuration changes. These operations are scoped to the project environment.
- [SAFE]: The skill demonstrates multiple safety-oriented design patterns, such as 'Activation Guards' that check for existing foundations before proceeding to prevent accidental overwriting, and a specific rule forbidding the inclusion of real secrets or credentials in generated examples, requiring placeholders instead.
Audit Metadata