The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides numerous templates for reverse shells and instructions to execute them. Evidence includes one-liners for Bash, Python, and PowerShell, as well as instructions for using netcat, Metasploit, and impacket to execute commands on remote targets. These are located in SKILL.md and references/manual-exploit-dev.md.
[REMOTE_CODE_EXECUTION]: The documentation contains explicit instructions for in-memory execution of remote scripts. Examples include piping curl output to bash, using Python's exec function with urlopen, and PowerShell's Invoke-Expression (iex) with Invoke-WebRequest (iwr). These patterns are detailed in the payload delivery techniques section of references/manual-exploit-dev.md.
[PERSISTENCE]: The skill instructs the agent on how to maintain access across sessions using persistence mechanisms. Specifically, it provides the command syntax for Metasploit's persistence module (run persistence -S -U -X -i 30 -p 4444 -r ) in references/framework-exploitation.md.
[DYNAMIC_EXECUTION]: The skill involves creating and executing code at runtime. It includes a C exploit skeleton and instructions for compiling it using gcc (gcc -o exploit exploit.c) on the fly. It also utilizes Python's exec() for executing dynamically fetched payloads. These are found in references/manual-exploit-dev.md.
[DATA_EXFILTRATION]: Instructions are provided for accessing and downloading sensitive files from targets, such as /etc/passwd and database contents via sqlmap. Meterpreter commands like 'download /etc/passwd /tmp/' are explicitly documented in references/framework-exploitation.md.
[INDIRECT_PROMPT_INJECTION]: The skill is highly vulnerable to indirect injection as its core workflow involves ingesting and executing untrusted third-party code.
Ingestion points: SKILL.md and references/exploit-research.md (GitHub PoC searches and searchsploit results).
Boundary markers: Absent; there are no instructions to delimit or ignore embedded instructions within downloaded PoCs.
Capability inventory: The skill has access to bash, python3, gcc, curl, and msfconsole across multiple files.
Sanitization: Absent; while the skill suggests checking stars and forks, it does not mandate code sanitization or sandboxing before execution.

vuln-exploit-technique