The Agent Skills Directory

[PROMPT_INJECTION]: The reference files contain various strings commonly associated with prompt injection attacks, such as 'Ignore previous instructions' and 'reveal your system prompt'. Analysis confirms these are exclusively used as part of adversarial test cases and educational content regarding prompt defense strategies, not as instructions for the agent.\n- [REMOTE_CODE_EXECUTION]: Multiple documents provide code snippets in Python and TypeScript demonstrating how to implement evaluation frameworks or parse structured outputs. These are provided as illustrative implementation examples for users and are not executed by the skill itself during its operation.\n- [SAFE]: The skill provides robust guidelines for secure prompt engineering, including the use of XML delimiters for input sandboxing, instruction hierarchies for priority enforcement, and canary tokens to detect prompt leakage. No malicious code or exfiltration patterns were detected.

prompt-engineer