The Agent Skills Directory

[SAFE]: The skill is composed entirely of instructional markdown and YAML metadata. It does not include any executable scripts, shell commands, or tool definitions that could pose a direct security risk.
[EXTERNAL_DOWNLOADS]: The skill contains numerous hyperlinks to external research papers, technical documentation, and official guidelines. These references target reputable academic repositories (such as arXiv, ACL Anthology, and OpenReview), trusted technology organizations (including OpenAI, Anthropic, and LangChain), and official government or international bodies (NASA and the World Health Organization). These are informational citations intended to ground the skill's methodology in established AI reliability research.
[DATA_EXFILTRATION]: The skill explicitly includes a 'Disclosure Boundary' section that instructs the agent on how to handle sensitive information like secrets, credentials, and PII. It mandates counting these items for completeness while redacting the actual values to prevent accidental exposure or exfiltration.
[PROMPT_INJECTION]: While the skill discusses prompt injection, it does so from a defensive perspective. It provides rules to identify and ignore 'instruction-formatted distraction' (embedded instructions in processed data) and specifically warns the agent not to render prompt-injection payloads found in input data.

methodical