wxa-skills-eval
Pass
Audited by Gen Agent Trust Hub on Jul 7, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The tool executes the WeChat Developer Tools CLI (wechatidecli) to run and analyze mini-programs. On Windows, it invokes PowerShell with the -ExecutionPolicy Bypass flag to provide a graphical directory selection dialog as described in the troubleshooting documentation.
- [PROMPT_INJECTION]: The evaluation engine exhibits an indirect prompt injection surface. It ingests untrusted metadata and documentation (app.json, mcp.json, and API descriptions) from the Mini Program projects being evaluated. This content is interpolated into LLM prompts during the intent generation and evaluation phases without explicit sanitization shown in the analyzed files.
- [EXTERNAL_DOWNLOADS]: The web-based evaluation interface and generated reports fetch typography resources from Google Fonts, which is a well-known and standard service.
- [PROMPT_INJECTION]: Mandatory Evidence Chain for Indirect Prompt Injection surface:
- Ingestion points: The start and gen_api_deps nodes read project configuration files (app.json, mcp.json) and source code fingerprints.
- Boundary markers: None explicitly mentioned in the configuration or pipeline documentation for separating untrusted project data from system instructions.
- Capability inventory: The skill has the capability to execute shell commands (via the DevTools adapter) and access the local file system for report generation.
- Sanitization: Not observed in the provided frontend assets or instruction files.
Audit Metadata