The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains a direct remote script execution pattern: bash -c \"$(curl -fsSL https://gef.blah.cat/sh)\" in references/gdb-debugging.md. This pattern downloads and executes code from an unverified external source.
[EXTERNAL_DOWNLOADS]: The skill instructs the agent to clone repositories from GitHub that are not part of the trusted vendor list, including firmadyne/firmadyne, pr0v3rbs/FirmAE, and pwndbg/pwndbg. These are then installed or configured using scripts such as sudo ./install.sh and sudo ./run.sh which run with elevated privileges.
[PRIVILEGE_ESCALATION]: Extensive use of sudo is required for many steps, including installing system packages, mounting filesystems (sudo mount), and creating network interfaces (sudo tunctl). While necessary for firmware emulation, it exposes the host to significant risk.
[COMMAND_EXECUTION]: The skill executes various system-level commands and tools such as qemu, iptables, and tcpdump. The use of sudo with these commands increases the potential impact of exploitation.
[INDIRECT_PROMPT_INJECTION]: The skill acts as a vulnerability surface because it processes untrusted firmware data. 1. Ingestion points: Extracted firmware filesystems (squashfs-root/, rootfs.ext4) and guest network traffic. 2. Boundary markers: The skill includes warnings that emulation environments are not security boundaries. 3. Capability inventory: The agent can execute shell commands via sudo, modify network configurations, and perform network requests. 4. Sanitization: There is no explicit sanitization or filtering of content from the firmware or the output of the emulation tools.

firmware-emulation