The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in Mode 3b, where it reads user-uploaded compliance reports in PDF, CSV, or Excel formats to update the local tracker.
Ingestion points: ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/CLAUDE.md (local config), ~/.claude/plugins/config/claude-for-legal-zh/corporate-legal/entities/compliance-tracker.yaml (local data), and external user-provided report files (PDF, CSV, Excel) via Mode 3b.
Boundary markers: Absent for input data processing; the skill relies on the agent's ability to interpret structured or unstructured documents without explicit delimiters or instructions to ignore embedded commands within the reports.
Capability inventory: File system read/write access to specific plugin configuration directories and text output generation for CSV/markdown tables.
Sanitization: The skill includes a robust 'Formula Injection Defense' section, instructing the agent to prepend a single quote to cells starting with =, +, -, @, and other control characters when exporting to CSV or tables to prevent DDE/CSV injection attacks on the user's spreadsheet software.
[COMMAND_EXECUTION]: The skill instructions specify access to sensitive local file paths within the user's home directory (~/.claude/plugins/config/...). While these paths are scoped to the skill's specific legal plugin, the ability to read and modify configuration files in these locations represents an elevated privilege within the agent's environment.

entity-compliance