The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted binary samples (e.g., suspect.exe), which represents a potential attack surface where a malicious file could exploit vulnerabilities in the parsing libraries or analysis tools.
Ingestion points: Binary data is read and parsed in SKILL.md (via heredoc scripts) and scripts/agent.py (via pefile and struct).
Capability inventory: The skill performs file read/write operations and executes the upx utility via system calls.
Boundary markers: The instructions explicitly advise using an isolated analysis VM without network connectivity as a security boundary.
Sanitization: No programmatic sanitization of binary content is present, which is standard for analysis tools but remains a risk factor.
[COMMAND_EXECUTION]: The scripts/agent.py script executes the upx command-line utility using the subprocess.run function.
Evidence: The implementation in unpack_upx uses a list-based argument format ["upx", "-d", "-o", output_path, filepath], which safely avoids shell interpolation and protects against simple command injection vectors through filenames.

analyzing-packed-malware-with-upx-unpacker