The Agent Skills Directory

Indirect Prompt Injection (LOW): The skill processes untrusted CSV data from users which could potentially contain malicious instructions intended to influence the AI's behavior.
Ingestion points: Read your_data_file.csv in the '工具使用指南' section.
Boundary markers: Absent. The instructions do not specify delimiters or warnings to treat the CSV content strictly as data.
Capability inventory: Bash (can execute system commands), Read (can access local files), and Grep (can search file contents).
Sanitization: Absent. There is no logic provided to escape or filter content before processing.
Command Execution (SAFE): The skill uses Bash to run legitimate local validation commands like pnpm run validate-data and file -I. These are restricted to the local environment and the primary purpose of the skill.
Data Exposure (SAFE): The skill reads internal project documentation (data_architecture.md) and source code (insurance.ts) to establish validation rules. This is necessary for its intended function and does not involve external exfiltration.

data-validator