The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The documentation in SKILL.md recommends the use of commands requiring elevated privileges. Specifically, it suggests using sudo mount to mount extracted firmware images and sudo chroot to emulate the firmware environment using QEMU.
[COMMAND_EXECUTION]: The automation script scripts/agent.py utilizes the subprocess module to execute external binaries. It calls binwalk for scanning and extraction, and the file utility to determine binary architectures within the extracted filesystem.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted firmware images, which serves as an attack surface for indirect prompt injection.
Ingestion points: The scripts/agent.py script reads text and binary data directly from the extracted firmware filesystem (e.g., lines 88, 110, 155) to search for strings and patterns.
Boundary markers: No explicit delimiters or instructions are used to notify the agent that the extracted data is untrusted and should not be interpreted as commands.
Capability inventory: The skill allows for command execution via subprocess and recommends high-privilege operations like sudo in its workflow.
Sanitization: The script extracts strings and IOCs without sanitization or escaping before including them in the final JSON report that may be reviewed by the agent.
[EXTERNAL_DOWNLOADS]: The skill references and downloads several external security tools and configuration resources from public GitHub repositories, including ReFirmLabs/binwalk, firmadyne/firmadyne, LongSoft/UEFITool, chipsec/chipsec, and e-m-b-a/emba.

performing-firmware-malware-analysis