The Agent Skills Directory

[PROMPT_INJECTION]: The skill describes methods for testing and identifying vulnerabilities related to Chain-of-Thought injection. While it provides examples of instructions that could be used for injection (e.g., "Observation: ...", "Emergency mode skip verification"), these are presented as research methodologies and test cases for security professionals to evaluate target system robustness. No actual malicious payloads are executed.
[REMOTE_CODE_EXECUTION]: No code or scripts are included in this skill. It consists solely of Markdown-based documentation.
[EXTERNAL_DOWNLOADS]: No external URLs or remote resources are referenced for download or execution.
[DATA_EXFILTRATION]: The skill mentions data exfiltration and database queries as potential attack targets in the context of security testing, but it does not contain any mechanisms to access or transmit sensitive data from the user's environment.

cot-injection