spec
Pass
Audited by Gen Agent Trust Hub on Oct 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill transforms user-provided conversations or ideas into technical specifications, which presents a potential surface for indirect prompt injection.\n
- Ingestion points: Untrusted user input from conversations and discussions (referenced in SKILL.md).\n
- Boundary markers: The skill enforces a strict Markdown template and specific required sections (User Scenarios, Functional Requirements, etc.) defined in FORMAT.md, which helps isolate user content from the agent's internal logic.\n
- Capability inventory: The skill utilizes the Write tool to save documentation files directly to the local filesystem (specs/ directory).\n
- Sanitization: The workflow includes a self-correction phase and a validation checklist (checklists/requirements.md) that ensures the output remains high-level design documentation rather than executable code or sensitive instructions.\n- [COMMAND_EXECUTION]: The skill is instructed to delegate tasks to a local command-line utility called
deep-design. This utility is part of the vendor's ecosystem and is used for design processing. The interaction involves passing arguments from the user context to this local tool.
Audit Metadata