The Agent Skills Directory

[COMMAND_EXECUTION]: Dynamic module and playbook loading. The module_engine.py and playbook_loader.py scripts use importlib.import_module to load and execute Python files from the modules/ and playbooks/ directories based on directory listings. This design allows for the execution of arbitrary Python code if malicious files are placed in those directories.
[PROMPT_INJECTION]: Indirect Prompt Injection surface.
Ingestion points: Security alert data (such as command lines, process names, and user identifiers) received via the webhook_receiver.py endpoint is processed by AI agents in the EDRAnalyzer module.
Boundary markers: None identified. The data is interpolated directly into agent prompts (e.g., 'Extract IOCs from this alert: {state[alert_data]}') without delimiters or instructions to ignore embedded commands.
Capability inventory: The platform is designed to automate case management, enrich artifacts, and generate security recommendations via the SIRPClient.
Sanitization: No sanitization or escaping of external alert content is observed before it is sent to the LLM for analysis.
[EXTERNAL_DOWNLOADS]: The installation process involves cloning a repository from GitHub and downloading several third-party Python dependencies. These are standard operations for deploying this type of platform.

agentic-soc-platform