The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructions utilize sudo mount, sudo losetup, and sudo umount to manage and inspect disk images within files like advanced-disk-and-memory.md and filesystem-and-archive-recovery.md. This level of access is necessary for forensic mounting but represents a privilege escalation vector.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted external artifacts, including PCAP files, disk images, and system logs. Evidence of this attack surface is found in SKILL.md (which defines the ingestion of disk, memory, and network captures) and in the reference scripts that perform automated parsing of these artifacts. There are no explicit boundary markers or output sanitization rules defined for the agent's interaction with this data.
[EXTERNAL_DOWNLOADS]: The reference materials suggest downloading several specialized forensic tools from external GitHub repositories, including openwall/john, gtoubassi/femtozip, kimci86/bkcrack, and others. These downloads are associated with well-known forensic projects but represent external dependencies.
[REMOTE_CODE_EXECUTION]: Instructions in audio-and-archive-stego.md, filesystem-and-archive-recovery.md, and steganography.md involve cloning and building tools from remote repositories (e.g., git clone followed by make for tools like john and femtozip).
[DYNAMIC_EXECUTION]: The linux-forensics.md file describes the use of pyrasite to attach to running Python processes via ptrace to recover in-memory source code, which involves injecting code into active processes.
[COMMAND_EXECUTION]: The skill heavily relies on executing a wide range of shell commands and custom Python scripts to perform complex forensics tasks such as data reassembly, decryption, and signal processing.

forensics-ctf