The Agent Skills Directory

Indirect Prompt Injection (MEDIUM): The skill provides a significant surface for indirect prompt injection by processing external untrusted data while having access to side-effect-producing tools.
Ingestion points: log_analyzer.py and stack_trace_parser.py process external log files and stack trace text.
Boundary markers: Absent; the scripts do not use delimiters to isolate untrusted content from agent instructions.
Capability inventory: performance_check.py can perform arbitrary outbound HTTP GET requests and enumerate all running system processes.
Sanitization: Absent; external content is parsed via regex and outputted without sanitization or escaping.
Data Exposure & Exfiltration (LOW): The performance_check.py script performs system reconnaissance by listing all running processes, PIDs, and resource usage. It also allows making HTTP GET requests to non-whitelisted domains, which could be used for basic SSRF or signaling.
Unverifiable Dependencies & Remote Code Execution (LOW): The skill requires the installation of external Python packages psutil and requests to function.
COMMAND_EXECUTION (LOW): The skill relies on the execution of local Python scripts to perform its analysis and system checks.

intelligent-debugger