The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill documents numerous payloads for achieving code execution through prototype pollution gadgets in popular server-side template engines such as EJS, Pug, Handlebars, and Nunjucks. Evidence includes payloads targeting properties like outputFunctionName, block, and AST nodes.
[COMMAND_EXECUTION]: The playbook outlines techniques for injecting shell commands into Node.js applications by polluting global prototype properties like shell, argv0, and NODE_OPTIONS, which are subsequently inherited by child_process functions.
[OBFUSCATION]: The instructions describe common filter bypass techniques used by attackers, including Unicode null byte escapes (\u0000) in JSON keys, the use of constructor.prototype as an alternative path to __proto__, and bracket notation variants.
[DATA_EXFILTRATION]: The skill mentions detection methodologies using response differentials, where an attacker modifies HTTP headers (like Content-Type) or status codes (e.g., setting status to 555) via prototype pollution to confirm successful injection.

prototype-pollution-advanced