qa
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from external websites when performing browser-based QA via
agent-browser snapshot(SKILL.md). Evidence chain: 1. Ingestion: External web page content is read into the agent context during testing sessions. 2. Boundary markers: No explicit delimiters or instructions to ignore embedded content are specified. 3. Capability inventory: The skill can perform file system writes (QA reports, scripts), execute shell commands via tools, and interact with web browsers. 4. Sanitization: No sanitization of ingested web content is described. - [DYNAMIC_EXECUTION]: The skill can invoke the
desplega:script-buildersub-skill to generate and save regression scripts in the project's scripts directory (SKILL.md). This is a functional feature designed to automate repetitive probe tasks identified during QA. - [EXTERNAL_DOWNLOADS]: The skill fetches configuration and guidelines for browser automation using the
agent-browsertool from Vercel Labs, which is a trusted organization.
Audit Metadata