The Agent Skills Directory

[COMMAND_EXECUTION]: Provides an extensive catalog of shell command payloads for both Linux and Windows environments, utilizing operators like ;, |, &&, and backticks to achieve arbitrary execution on target systems.
[REMOTE_CODE_EXECUTION]: Instructs the agent on methods to achieve RCE, including command substitution ($(...)), polyglot payloads for different quoting contexts, and argument injection into programs like ssh, tar, and rsync.
[DATA_EXFILTRATION]: Outlines several methods for exfiltrating sensitive data from targets, including time-based conditional sleeps, Out-of-Band (OOB) DNS lookups via tools like interactsh or dig, and HTTP-based exfiltration using curl or wget.
[PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by processing untrusted data from target applications while possessing high-impact capabilities.
Ingestion points: Processes untrusted inputs from target application parameters, headers, and filenames (SKILL.md).
Boundary markers: Absent; there are no instructions provided to separate or delimit processed target data from the agent's internal instructions.
Capability inventory: Includes full OS shell command execution, network operations (DNS, HTTP), and file system writes across multiple scripts and steps (SKILL.md).
Sanitization: Absent; the skill is designed to bypass sanitization rather than apply it to ingested data.

command-injection