The Agent Skills Directory

[DATA_EXFILTRATION]: The skill provides template Aggressor Scripts in 'references/aggressor-scripting.md' that use Java networking APIs (java.net.URL) to send system metadata, including internal/external IPs and hostnames, to external webhooks.
[COMMAND_EXECUTION]: The methodology includes extensive instructions for executing arbitrary code on remote targets using shell, powershell, and specialized C2 commands like 'psexec', 'winrm', and 'wmi' as documented in 'SKILL.md' and 'references/beacon-operations.md'.
[CREDENTIALS_UNSAFE]: Documents procedures for harvesting sensitive authentication data, including NTLM hashes, cleartext passwords via memory dumping (Mimikatz), and Active Directory synchronization (DCSync).
[DYNAMIC_EXECUTION]: Details the use of Aggressor Scripts (.cna) for runtime customization and 'execute-assembly' for in-memory .NET program execution, allowing for dynamic capability extension without disk-based indicators.
[INDIRECT_PROMPT_INJECTION]: The skill establishes an operational environment where the agent processes data from untrusted external targets while having access to powerful execution tools, creating a surface for indirect prompt injection. Ingestion points: Data retrieved via directory listings (ls), process lists (ps), and file downloads as described in 'references/beacon-operations.md'. Boundary markers: No explicit delimiters or security prompts are recommended for handling data from compromised targets. Capability inventory: Administrative control including shell access, file manipulation, and credential harvesting. Sanitization: No sanitization logic is provided for target system outputs before they are processed by the agent.

cobalt-strike