The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/agent.py script and accompanying workflow instructions use subprocess.run to execute external binary analysis tools such as de4dot, diec, and monodis. While the script uses argument lists to prevent shell injection, these commands are executed on untrusted file paths provided by the user.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and parse untrusted .NET assemblies, creating a vulnerability surface where malicious content within the analyzed binary could attempt to manipulate the agent.\n
Ingestion points: File reading and binary header parsing in scripts/agent.py and SKILL.md.\n
Boundary markers: The skill does not implement boundary markers to isolate extracted strings or metadata from the analysis logic.\n
Capability inventory: The skill possesses file read/write access and the ability to execute subprocesses.\n
Sanitization: Indicators are extracted using regular expressions, but general metadata is processed without extensive sanitization.\n- [DYNAMIC_EXECUTION]: The script dynamically invokes external decompilation and deobfuscation tools at runtime to process malware samples.\n
Evidence: Multiple subprocess.run calls in scripts/agent.py are used to wrap functionality from external utilities.\n- [EXTERNAL_DOWNLOADS]: The skill references well-known external repositories for analysis tools such as dnSpyEx and de4dot on GitHub. These references are standard for the malware analysis use case and target established community tools.

reverse-engineering-dotnet-malware-with-dnspy