The Agent Skills Directory

[PROMPT_INJECTION]: Several templates, specifically references/templates/promptfoo-llm-wiki-redteam.yaml and references/templates/promptfoo-llm-wiki.yaml, contain strings like 'Ignore all prior instructions' and 'reveal the system prompt'. These are part of a security evaluation suite (Promptfoo) designed to test the agent's ability to resist injection attacks. These are data artifacts for testing, not active instructions for the agent.
[INDIRECT_PROMPT_INJECTION]: The skill documents an attack surface where untrusted wiki content or web clips could contain malicious instructions. It mitigates this by providing templates for 'redaction-policy.yml' and 'security-scorecard.yaml', and by recommending automated testing for 'suspicious prompt-like text' during ingestion.
[DATA_EXPOSURE]: The skill includes synthetic 'canary' secrets (e.g., sk-test-canary-000000 and AKIA0000000000000000) in the promptfoo-llm-wiki-redteam.yaml file. These are used to verify that the redaction and security filtering tools correctly identify and block secret leakage. No real credentials or sensitive paths are exposed.
[EXTERNAL_DOWNLOADS]: The CI/CD templates (references/templates/gitlab-llm-wiki-ci.yml) reference the official GitLab CLI image (registry.gitlab.com/gitlab-org/cli:latest). This is a reference to a well-known service for legitimate development operations.

llm-wiki-gitlab-operating-model