spec
Pass
Audited by Gen Agent Trust Hub on Nov 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external, potentially untrusted data including user-provided feature ideas, brainstorms, and existing project source code across multiple layers (entities, handlers, controllers).
- Ingestion points: Ingests user requirement text (idea/brainstorm), project source code files (entities, validators, handlers, controllers), and existing specification documents.
- Boundary markers: The skill mandates the use of 'tech-free' prose in business sections and 'abstract anchors' (e.g., [Source: namespace/service/id]) to isolate business logic from implementation details, which serves as a natural boundary against technical injection.
- Capability inventory: The skill has the capability to write to the file system (updating specs and reports), spawn sub-agents for parallel processing, and execute local analysis scripts.
- Sanitization: Implements strict mandates (M1-M7) that require technical identifiers to be kept out of narrative prose, though it does not explicitly mention automated sanitization of user-provided ideas.
- [COMMAND_EXECUTION]: The skill frequently executes shell commands and local scripts to perform codebase analysis.
- Evidence: Instructions include the execution of
git ls-files,grep,rg, and local scripts such asnode .claude/scripts/ai-signal-scan.cjsandpython .claude/scripts/code_graph. These are used to discover entities, operations, and actors within the project structure.
Audit Metadata