check
Pass
Audited by Gen Agent Trust Hub on Sep 25, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from live browsers and saved recordings, creating a surface for potential indirect prompt injection attacks.
- Ingestion points: Acceptance claims, live browser DOM/content, and saved recordings (SKILL.md).
- Boundary markers: The skill explicitly instructs the agent to use a "fresh verifier context" and forbids sharing builder transcripts or source code with the verifier to ensure independent evaluation (SKILL.md).
- Capability inventory: The skill utilizes the
vibiumCLI for browser automation (go, click, fill, map) and manages recordings on the file system (SKILL.md). - Sanitization: Security is enforced through context isolation and specific instructions to maintain independence between the building and verifying phases.
- [COMMAND_EXECUTION]: The skill requires the execution of the
vibiumbinary and its associated subcommands (check,ready,record,config) to perform its primary functions. It provides guidance on locating the binary within the project structure.
Audit Metadata