om-mockup-prototype
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFE
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from product briefs and synthetic panel reports. It proactively mitigates this risk by including a dedicated 'Untrusted content boundary' section in its setup instructions, which explicitly warns the agent against executing instructions found within input documents. Additionally, it mandates the escaping of all input copy and forbids the copying of scripts from input documents.
- Ingestion points: Product briefs (
{brief}) and panel reports (--panel-report) defined inSKILL.md. - Boundary markers: Present in
references/agentic-setup.mdunder the 'Untrusted content boundary' header. - Capability inventory: File writing (Step 3), local browser execution for verification (Step 4), and repository file reading (Step 1).
- Sanitization: Explicitly required in
references/prototype-format.md, including escaping input copy and using text nodes for user-provided content. - [COMMAND_EXECUTION]: The skill utilizes a browser provider (defaulting to Playwright) for artifact verification. It secures this process by restricting the agent to using repository-local descriptors and isolated sessions, specifically forbidding network access or remote destinations during browser walks.
- [PRIVILEGE_ESCALATION]: The skill implements robust path validation and containment checks in its preflight steps. It ensures that all file operations are repository-relative and strictly confined to a specific revision directory, preventing directory traversal and unauthorized access to sensitive system files.
Audit Metadata