The Agent Skills Directory

[DYNAMIC_EXECUTION]: The code in module_engine.py and playbook_loader.py dynamically loads Python modules from the filesystem using importlib.import_module. It iterates through the modules/ and playbooks/ directories and imports any .py files found. If a malicious file is placed in these directories, the platform will automatically execute its contents when the engine or loader starts.
[INDIRECT_PROMPT_INJECTION]: The skill has a vulnerability surface for indirect prompt injection due to how it handles external alert data.
Ingestion points: Security alerts are received from external systems (like SIEMs) via the /webhook/alert endpoint in webhook_receiver.py.
Boundary markers: The prompt templates in LanggraphAgent do not use clear delimiters or instructions to ignore embedded commands when interpolating alert_data into the AI context.
Capability inventory: The platform has the capability to perform network requests to external APIs (OpenAI, VirusTotal, AbuseIPDB, SIRP) and manipulate Redis streams, which an injected instruction could attempt to misuse.
Sanitization: The skill lacks explicit sanitization or filtering of the alert payload before passing it to the AI agent for analysis.
[EXTERNAL_DOWNLOADS]: The installation instructions guide the user to download source code from an external repository at https://github.com/FunnyWolf/agentic-soc-platform.git.
[COMMAND_EXECUTION]: The deployment documentation involves executing multiple shell commands, including git clone, docker-compose up, pip install, and running several Python management and engine scripts, which triggers the execution of downloaded code.

agentic-soc-platform