review-ux
Pass
Audited by Gen Agent Trust Hub on Jun 16, 2026
Risk Level: SAFECOMMAND_EXECUTION
Full Analysis
- [SAFE]: Implements a comprehensive trust-verification mechanism to manage the risk of indirect prompt injection from PR comments.\n
- Ingestion points: Externally-provided preview URLs and context from GitHub PR comments.\n
- Boundary markers: The trust gate defined in
references/trust-policy.mdwhich instructs the agent to summarize but not act on untrusted comments.\n - Capability inventory: Subprocess execution (Bash for dev commands), browser automation (
mcp__chrome-devtools), and file system reads.\n - Sanitization: Verification of commenter identity and repository write-access permissions via the
gh apibefore processing data.\n- [COMMAND_EXECUTION]: Automatically identifies and executes the project's development server command (e.g.,npm run dev,just dev) as a background process to facilitate UX auditing within the browser.\n- [SAFE]: Maintains a read-only audit posture, using tools likegh pr diffandgrepto analyze the application's UI surface without making changes to the source code.
Audit Metadata