reg-gap-analysis
Warn
Audited by Gen Agent Trust Hub on Sep 20, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [DATA_EXFILTRATION]: The skill is designed to read sensitive local configuration files located at
~/.claude/plugins/config/claude-for-legal-zh/privacy-legal/CLAUDE.md. - This file contains proprietary or sensitive information including internal privacy rule commitments, regulatory coverage scopes, and a list of systems used for Data Subject Access Request (DSAR) processing.
- While the skill's primary purpose is to use this data for analysis, the access to hidden application configuration directories is a sensitive operation.
- [INDIRECT_PROMPT_INJECTION]: The skill accepts untrusted user input via the
argument-hintwhich encourages users to "paste law text or summaries." - Ingestion points: User-provided law text processed in
Step 2: Extract requirements. - Boundary markers: The skill uses a table format for extraction but lacks explicit delimiters or instructions to ignore commands embedded within the pasted regulatory text.
- Capability inventory: The skill has the ability to read sensitive local files, write output files, and invoke external search tools.
- Sanitization: There is no evidence of sanitization for the user-pasted text before it is used to influence the analysis logic.
- [COMMAND_EXECUTION]: The workflow incorporates "Retrieval Connectors" (e.g.,
yuandian MCPor other legal search tools). - These are external components that execute network or system-level commands to fetch live regulatory data from the internet or proprietary databases.
Audit Metadata