The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions grant the agent the ability to execute a wide variety of high-risk security tools. These include automated exploitation frameworks like Metasploit (msfconsole), database injection tools like sqlmap, and credential auditing tools like crackmapexec. The agent is encouraged to use these tools autonomously across various attack phases.
[PRIVILEGE_ESCALATION]: The skill guides the agent in workflows designed to gain unauthorized elevated access. This includes instructions for credential dumping using mimikatz (specifically targeting logon passwords) and impacket-secretsdump. It also suggests configuring Docker environments with NET_ADMIN and NET_RAW capabilities, which allow the container to bypass standard network isolation and perform low-level packet manipulation.
[DATA_EXFILTRATION]: The skill includes patterns for accessing highly sensitive system and network data. It specifically mentions accessing cloud metadata services via the 169.254.169.254 endpoint to harvest IAM credentials and dumping NTLM hashes for lateral movement. These capabilities are inherently designed to extract sensitive information from the environment being tested.
[INDIRECT_PROMPT_INJECTION]: The skill designs a workflow where the agent ingests and acts upon data retrieved from external network targets, which could be used to manipulate the agent's behavior.
Ingestion points: Data is ingested from external network sources via tools such as nmap, nikto, subfinder, whatweb, and API fuzzers (referenced in SKILL.md).
Boundary markers: None. The skill does not provide delimiters or instructions for the agent to treat output from these tools as untrusted data.
Capability inventory: The skill provides full shell access via local CLI, SSH, and Docker, including network exploitation and file-writing capabilities (referenced in SKILL.md).
Sanitization: None. The instructions prompt the agent to "analyze and iterate" based on the raw results of the exploitation and reconnaissance tools.
[EXTERNAL_DOWNLOADS]: The skill documentation suggests downloading and installing software from external sources at runtime, including pulling the kalilinux/kali-rolling Docker image and installing the kali-linux-default package bundle. These operations introduce external code into the execution environment.

kali-pentest-ai-agent