The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains examples of malicious prompts (e.g., 'Ignore previous instructions', 'Print your system prompt') used to demonstrate vulnerabilities in LLM-based applications. These are informational data samples and not active instructions to the AI agent.
[DATA_EXFILTRATION]: Provides references to sensitive file paths such as ~/.aws/credentials, /etc/passwd, and .env as common targets in security research. These references are part of the documentation and do not involve unauthorized data access or exfiltration by the skill.
[OBFUSCATION]: Documents various obfuscation techniques used by attackers, including homoglyph characters in URLs and Unicode tags to hide instructions. These are provided as examples for training and identification purposes.
[REMOTE_CODE_EXECUTION]: Includes descriptions and payloads for RCE vulnerabilities, such as shell pipelines using base64 and curl. These are educational snippets intended for security auditing.
[COMMAND_EXECUTION]: Lists various security tool commands (e.g., sqlmap, nuclei, ffuf) for manual execution by a security researcher.

web2-vuln-classes