locate-ui-from-code
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill interacts with live web pages, which are untrusted data sources. Maliciously crafted content on a page, such as prompt injection hidden in text or attributes, could potentially influence the agent's behavior. The skill operates with high-privilege capabilities and lacks explicit sanitization for data extracted from the browser.\n
- Ingestion points: Browser snapshots and DOM evaluations performed during discovery and navigation (SKILL.md, Phase 5).\n
- Boundary markers: No specific delimiters or instructions are provided for handling content returned from the browser.\n
- Capability inventory: The skill utilizes shell command execution (Bash), file modification (Write/Edit), and complex browser automation (playwright-cli, chrome-devtools).\n
- Sanitization: The skill does not define methods to filter or escape data retrieved from external web pages.\n- [DYNAMIC_EXECUTION]: The skill dynamically generates and executes JavaScript within the browser context to create a visual highlight overlay. It uses a predefined script template and performs evaluation via tools like
evalorRuntime.evaluate. This behavior is core to the skill's primary purpose.\n- [EXTERNAL_DOWNLOADS]: The skill documentation refers to installing the@playwright/clipackage from the NPM registry. This is a legitimate and widely-used browser automation tool from a well-known provider.\n- [COMMAND_EXECUTION]: The instructions involve running shell commands such asgrep,kill-port, and development scripts (pnpm dev) to trace code and prepare the testing environment.
Audit Metadata