The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted binary files and memory dumps which could contain malicious instructions designed to influence the agent's behavior during analysis.
Ingestion points: Untrusted data enters the agent context via malware samples processed by scripts/triage.py, scripts/auto_unpack.py, and scripts/cs_config_extract.py, as well as memory dumps analyzed by scripts/mem_triage.py and network captures in scripts/beacon_profiler.py.
Boundary markers: The skill documentation explicitly instructs the user to perform dynamic analysis and detonation within isolated virtual machines (VMs) and snapshotted environments to mitigate risks.
Capability inventory: The skill scripts can execute external shell commands via subprocess.run (to call tools like Volatility, capa, and FLOSS) and perform local file writes to save extracted payloads and reports.
Sanitization: Shell commands are constructed using list-based arguments and shlex.split, which prevents basic command injection via malicious filenames.
[COMMAND_EXECUTION]: Several scripts execute external binaries to perform technical analysis tasks.
Evidence: scripts/triage.py invokes floss, capa, and yara-x to identify binary capabilities. scripts/mem_triage.py wraps the Volatility 3 framework (vol) to triage memory images for injection artifacts.
[EXTERNAL_DOWNLOADS]: The skill documentation and scripts reference several external dependencies and specialized security tools.
Evidence: The skill relies on Python packages such as pefile, unicorn, capstone, scapy, and yara-x. It also references repositories and tools from well-known sources, including Mandiant's capa, the Canadian Centre for Cyber Security's MACO framework, and VirusTotal's YARA-X engine.

malware-analysis