ui-automation
Pass
Audited by Gen Agent Trust Hub on Oct 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with untrusted external websites and applications, creating a potential surface for indirect prompt injection.
- Ingestion points: External URLs and native applications targeted for UI automation as defined in
SKILL.mdandui-acceptance.md. - Boundary markers: Implements a strict 'Artifact Contract' in
contract.mdthat restricts agent operations to a specific run directory and requires structured JSON responses. - Capability inventory: Includes browser automation, desktop GUI interaction, and file writing capabilities through the
codexCLI tool. - Sanitization: Mandates manual review of screenshots and artifacts in
SKILL.mdStep 4 to ensure findings are verified visually before being accepted by the primary agent. - [COMMAND_EXECUTION]: The skill utilizes the
codexCLI tool for its primary purpose of UI automation. The instructions guide the agent to construct commands that output to local workspace directories, using standard shell variables for run IDs and directory management. The process does not involve downloading or executing arbitrary external scripts or acquiring elevated system privileges.
Audit Metadata