The Agent Skills Directory

Command Execution: The skill uses subprocess.run to coordinate between its internal Python scripts and to invoke the Microsoft Threat Modeling Tool. These executions are part of the core generation and validation workflow and target known local executables.
Dynamic Assembly Loading: The PowerShell fidelity check script (Deserialize-Tm7.ps1) dynamically loads .NET assemblies from the local environment. To mitigate the risk of loading untrusted code, it includes a signature verification step that requires a valid Authenticode signature from 'Microsoft Corporation' before any assembly is loaded into the process.
XML Security Hardening: The skill includes specialized logic in tm7_threat_contract.py to prevent XML External Entity (XXE) attacks. It uses the defusedxml library and performs custom scans for DTD and entity declarations across multiple character encodings (UTF-8, UTF-16, etc.) to ensure that malicious XML payloads are rejected before parsing.
Data Ingestion and Sanitization: The skill processes YAML/JSON specifications and XML threat models. While these represent a surface for indirect prompt injection, the skill's instructions and implementation emphasize treating this content as inert data, and it employs strict schema validation to prevent malicious instructions from influencing agent behavior.
Integrity Protections: Files are written using atomic patterns (temporary files followed by a replace operation) and semantic hashes are used to verify that the core structure of threat models remains unchanged during automated population tasks.

security-planning