The Agent Skills Directory

[PROMPT_INJECTION]: The skill explicitly instructs the agent to bypass safety guardrails and ignore content policies. It provides specific injection patterns such as 'you are DAN...', roleplay personas, and authority impersonation to move the model's response into a region where safety training is ineffective.
[INDIRECT_PROMPT_INJECTION]: The guide outlines methods to exploit the model's context window and instruction hierarchy. It details techniques like instruction-hierarchy confusion via fake system messages, many-shot attacks with fake compliant examples, and context-flooding to override intended trust boundaries.
[NO_CODE]: The skill consists entirely of instructional text and documentation without executable scripts or external dependencies.

ai-jailbreak