The Agent Skills Directory

[PROMPT_INJECTION]: The skill outlines methodologies that involve ingesting and displaying untrusted data from external servers, which creates an attack surface for indirect prompt injection. If an agent executes these scripts and monitors the output, a malicious server could return crafted content designed to influence the agent's behavior.
Ingestion points: The Python script in Pattern 5 captures and prints raw server response bodies using print(status, body). Similar patterns are described for biometric verification and SSO audits.
Boundary markers: No specific boundary markers or instructions to treat external data as untrusted are included in the provided scripts.
Capability inventory: The skill leverages network communication tools (curl, ffuf, aiohttp) and shell command execution.
Sanitization: The skill lacks mechanisms to sanitize or validate the content of HTTP responses before they are processed by the agent's console.
[EXTERNAL_DOWNLOADS]: The skill depends on external libraries and third-party security tools to perform its intended functions.
Python packages: The code in Pattern 5 requires the aiohttp library, which must be installed from a public registry.
[DATA_EXFILTRATION]: The skill encourages the capture and use of sensitive session identifiers and cookies (e.g., YOUR_SESSION) to test authentication flows. Additionally, it contains instructions for network operations targeting non-whitelisted domains (e.g., target.com). While these are standard practices for security auditing, they involve the handling of sensitive data and external communication.
[COMMAND_EXECUTION]: The skill provides multiple shell-based workflows utilizing tools such as ffuf, curl, jq, and grep for automated security testing and log analysis. These commands are intended to be executed in the environment where the audit is performed.

hunt-mfa-bypass