The Agent Skills Directory

[DATA_EXFILTRATION]: The uxaudit-locator subagent is designed to scan project directories for sensitive configuration files, including .env, package.json, and environment-specific declarations. While intended for environment discovery, the combination of reading these files and the presence of network-capable subagents (uxaudit-scout, uxaudit-journey-compiler utilizing WebFetch and WebSearch) creates a risk surface for data exfiltration.
[COMMAND_EXECUTION]: The orchestrator and subagents employ Bash to launch background development servers, manage processes, and run a suite of Python and Node.js utility scripts, such as run_all_checks.py and capture.mjs.
[REMOTE_CODE_EXECUTION]: The skill uses a 'Journey Compiler' to transform natural language descriptions into executable JSON scripts. These scripts are subsequently interpreted and executed by capture_journey.mjs to perform automated browser actions, representing dynamic code generation and execution.
[EXTERNAL_DOWNLOADS]: Subagents are configured with WebSearch and WebFetch tools to retrieve external documentation and project context during the audit's discovery phase.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted application data (HTML, code, and screenshots) to drive automated testing logic. (File: SKILL.md)
Ingestion points: Project source files, package.json, and rendered application screenshots.
Boundary markers: None explicitly mentioned for data interpolation.
Capability inventory: Subagents possess Bash, Write, and WebFetch capabilities.
Sanitization: No explicit sanitization or validation of project-sourced data is described.

uxaudit