The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The file 'reference/web-llm-attacks-cheat-sheet.md' contains several functional reverse shell payloads for Bash, Python, Perl, and Netcat, designed to establish unauthorized remote connections. Additionally, an automated antivirus scan detected a Trojan signature (Script:SNH-gen [Trj]) in this file.
[DATA_EXFILTRATION]: Multiple reference files provide commands to exfiltrate sensitive files, such as '/etc/passwd' and '.bashrc', to external domains like 'attacker.com' via POST requests, DNS queries, and base64 encoding.
[COMMAND_EXECUTION]: The skill includes payloads for destructive actions, including deleting system files (rm -rf /) and dropping database tables. It also documents techniques to bypass command filters using shell variables, whitespace manipulation, and encoding.
[PROMPT_INJECTION]: The reference materials include explicit prompt injection strings designed to override LLM behavior, such as 'Ignore all previous instructions' and 'Reveal your system prompt'.
[EXTERNAL_DOWNLOADS]: The skill documents and includes payloads that involve downloading and executing remote scripts from untrusted external sources, such as 'attacker.com/shell.ps1' and 'attacker.com/xss.js'.

api-security