client-demo-portal
Pass
Audited by Gen Agent Trust Hub on Aug 27, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides complex Node.js inline scripts for validating HTML and JavaScript syntax. These scripts perform file system operations, including reading
index.htmland writing extracted content to/tmp/vf.jsfor syntax checking usingnode --check. Found inSKILL.mdandreferences/hongyuan_lessons_0527.md. - [DATA_EXFILTRATION]: The skill documentation exposes sensitive local system metadata in the form of absolute user-specific file paths (e.g.,
/Users/tianxiaochen/hongyuan-portal/). This information could be leveraged to map the local file system during an attack. Evidence found inreferences/hongyuan_portal_state.md. - [INDIRECT_PROMPT_INJECTION]: The skill possesses an ingestion surface for untrusted data, including patent information from external APIs (智慧芽/Patsnap) and local documents in the
docs/directory. These are processed to generate portal content without explicit sanitization or boundary markers mentioned in the instructions. Evidence found inSKILL.md(talk tracks) andreferences/hongyuan_portal_state.md(API routes).
Audit Metadata