om-ux-shape
Pass
Audited by Gen Agent Trust Hub on Jul 27, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill is a documentation-driven workflow for product shaping. It does not perform unauthorized network operations, execute hidden code, or attempt to exfiltrate data.- [PROMPT_INJECTION]: The skill includes sophisticated defensive measures against prompt injection attacks. In references/agentic-setup.md, it explicitly defines an 'Untrusted content boundary', instructing the agent to treat repository and document content as data only, and to refuse and report common injection patterns like 'ignore previous instructions'.- [DATA_EXFILTRATION]: The skill enforces 'Secrets hygiene' in references/rules.md, prohibiting the inclusion of tokens, .env content, or credentials in any output or handoff reports.- [PROMPT_INJECTION]: The skill identifies a surface for indirect prompt injection by ingesting untrusted data from the repository. Ingestion points: Processes repository-local files, including .ai/agentic.config.json and potential skill extensions in .ai/skills/om-ux-shape/SKILL.md as described in references/agentic-setup.md. Boundary markers: Strong instructions in references/agentic-setup.md define the boundary between instructions and data, warning against following commands found in untrusted sources. Capability inventory: The skill does not use any tools for code execution, network requests, or file-system modifications across its referenced scripts. Sanitization: Instructions require the agent to validate externally-sourced values and use quoting during interpolation to prevent unintended command or path execution.
Audit Metadata