The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains reference material documenting common adversarial patterns such as "Ignore previous instructions", "DAN", and jailbreak role-play templates.
Evidence: Files references/prompting-risks.md, references/mistakes-security.md, and references/failure-taxonomy.md include these as examples for educational and audit purposes.
Context: As the skill is a prompt engineering manual, including risk patterns is consistent with its educational purpose and does not represent an attempt to hijack the agent.
[SAFE]: The skill is a "no-code" package composed exclusively of Markdown documentation.
Evidence: Analysis confirmed the absence of executable code, shell scripts, or configurations for automated tools within the skill directory.
[SAFE]: No data exfiltration or credential risks were found.
Evidence: No hardcoded secrets, API keys, or unauthorized network operations were detected across the 5,000+ lines of reference content.

prompt-engineering