The Agent Skills Directory

[Privilege Escalation] (HIGH): The skill instructs the agent to use powershell -ExecutionPolicy Bypass in SKILL.md. This command explicitly disables host-level security protections designed to prevent the execution of untrusted or unsigned scripts on Windows systems.
[Indirect Prompt Injection] (HIGH): The skill is designed to ingest and process external files from the data directory (CSV/XLSX). Since the processing logic includes shell and Python execution, it constitutes a high-risk surface where malicious data in those files could be used to execute arbitrary commands or influence agent behavior. (Evidence: 1. Ingestion point: data folder; 2. Boundary markers: Absent; 3. Capability inventory: PowerShell and Python execution; 4. Sanitization: Absent).
[Unverifiable Dependencies] (MEDIUM): The PowerShell script executes tools/hvdc_ops.py, which is not included in the provided file list. This prevents a complete audit of the code being executed and its handling of external data.
[Obfuscation] (LOW): The SKILL.md file contains the string '以묐떒' (Korean for 'stop' or 'interruption'). While likely a UI status indicator, the use of non-standard characters in instruction sets can be a technique used to evade simple string-matching security filters.

etl-duckdb