The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides numerous pre-written payloads for executing code on remote servers across various languages and frameworks.
Evidence includes Jinja2 RCE payloads using popen and read in server-side.md.
Evidence includes EJS template injection for system command execution via child_process.
Evidence includes Perl 2-argument open() command injection and Ruby instance_eval breakout techniques.
[DATA_EXFILTRATION]: Instructions are provided for leaking sensitive system files and credentials to external domains.
client-side.md contains payloads for stealing document.cookie via fetch to exfil.com.
server-side.md demonstrates exfiltrating flags or files using curl to webhook.site.
cves.md describes methods for leaking credentials via curl's .netrc handling and browser Link headers.
[DYNAMIC_EXECUTION]: The documentation includes sophisticated techniques for escaping Node.js VM sandboxes.
node-and-prototype.md details VM escapes using this.constructor.constructor to access the host's process and child_process modules.
It describes a "Full Chain" attack combining prototype pollution with document.write to bypass script evaluation restrictions in the happy-dom library.
[COMMAND_EXECUTION]: The skill lists multiple ways to execute arbitrary shell commands and bypass security filters.
SKILL.md and server-side.md provide command injection patterns using backticks, semicolons, and subshells.
Techniques for bypassing keyword blocklists for tools like cat or head are provided using sed, awk, and tac.
[PRIVILEGE_ESCALATION]: Several techniques for escalating user privileges and bypassing authentication are documented.
auth-and-access.md describes JWT algorithm confusion (RS256 to HS256) and balance replay attacks to gain unauthorized value.
server-side.md explains Python shared object (.so) hijacking by writing malicious files to the filesystem and forcing a re-import to gain process control.

ctf-web