The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains example prompt injection strings (e.g., 'Ignore all prior instructions') within the references/templates/promptfoo-llm-wiki-redteam.yaml file. These are explicitly defined as test fixtures for a red-teaming tool (promptfoo) to verify that an LLM can successfully ignore malicious instructions in a controlled testing environment.
[CREDENTIALS_UNSAFE]: The references/templates/promptfoo-llm-wiki-redteam.yaml template includes synthetic credential strings (e.g., 'sk-test-canary-000000', 'AKIA0000000000000000'). These are clearly documented as 'canary secrets' used for testing the efficacy of secret scanners and data protection controls.
[EXTERNAL_DOWNLOADS]: The documentation in references/docs/19-security-threat-model.md references official security resources and repositories from well-known organizations including OpenAI, Anthropic, OWASP, GitHub, NVIDIA, and NIST. These references are provided for educational purposes and do not involve automated execution of remote code.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest implementation details and data flow descriptions to generate a threat model. While this creates a surface for processing untrusted descriptions, the skill itself contains no executable code or scripts, and its primary function is text generation for security documentation. It explicitly instructs users to implement boundary markers and sanitization in their own LLM-Wiki systems.

llm-wiki-threat-model