The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the native Windows utility wevtutil to query Event ID 4104 (Script Block Logging) and 4103 (Module Logging) for analysis. The command execution in scripts/agent.py uses a list-based argument structure with subprocess.run, and the inputs (Event ID and max event count) are constrained as integers by the command-line parser, preventing command injection.
[INDIRECT_PROMPT_INJECTION]: The tool has an attack surface for indirect prompt injection as it ingests untrusted data from Windows Event Logs.
Ingestion points: scripts/agent.py reads log data via wevtutil and extracts PowerShell script blocks.
Boundary markers: The findings are returned to the agent in a structured JSON format, which helps differentiate metadata from processed content.
Capability inventory: The script performs read-only log queries and regex-based analysis; it does not contain file-write or network-send capabilities.
Sanitization: The script decodes and previews external content but treats it as data for inspection rather than instructions.
[SAFE]: The regex patterns used in scripts/agent.py correctly identify well-known PowerShell Empire artifacts (e.g., Invoke-Mimikatz, -noP -sta -w 1 -enc). These patterns are the targets of the detection tool and do not represent malicious behavior by the skill itself.

analyzing-powershell-empire-artifacts