measure-ui
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to interact with and read data from external web pages to perform UI measurements, which creates a potential surface for instructions to be ingested from untrusted sources.
- Ingestion points: Browser content accessed via drivers like 'control-ui' or 'orca' as described in SKILL.md and fidelity.md.
- Boundary markers: No specific delimiters or 'ignore embedded instructions' warnings are provided for the data read from the browser.
- Capability inventory: The skill utilizes JavaScript execution (via browser evaluate), mouse movement, keyboard input, and hit-testing capabilities (interaction-states.md).
- Sanitization: No explicit sanitization or validation of the retrieved DOM data is described in the measurement loops.
- [DYNAMIC_EXECUTION]: The skill uses JavaScript snippets to be executed dynamically within the browser environment to retrieve style and layout information.
- Execution method: The agent is guided to use browser drivers to execute JS code for retrieving computed values.
- Evidence: Code blocks in SKILL.md and the references folder for accessing styleSheets, getComputedStyle, and getBoundingClientRect.
- Intent: The execution is restricted to the browser context for the purpose of UI verification and matches the stated primary use case of the skill.
Audit Metadata