The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/agent.py executes local system commands to enumerate WMI subscriptions.
Evidence: The script uses subprocess.run to call wmic to query the root\subscription namespace for filters, consumers, and bindings.
Evidence: The script uses subprocess.run to execute a PowerShell block that retrieves and formats WMI subscription data into JSON.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted log data which could be crafted to influence an agent's behavior during the analysis phase.
Ingestion points: scripts/agent.py reads Sysmon EVTX files; scripts/process.py parses JSON and CSV log files.
Boundary markers: Absent. The scripts extract data directly from logs without wrapping content in delimiters or applying "ignore" instructions.
Capability inventory: The skill can execute local commands (wmic, powershell) and perform local file writes (scripts/process.py generates finding reports).
Sanitization: The scripts match suspicious strings via regex but do not sanitize or escape the extracted log content before reporting it.
[EXTERNAL_DOWNLOADS]: The skill documentation and code reference external libraries required for log parsing.
Evidence: references/api-reference.md and scripts/agent.py identify the python-evtx library as a dependency for processing Sysmon Event logs.

hunting-for-persistence-via-wmi-subscriptions