The Agent Skills Directory

Safe Text Processing: The skill is designed strictly for prose manipulation. It lacks any tool-calling capabilities (no allowed-tools in frontmatter) or shell execution, which significantly minimizes its security footprint.
Indirect Prompt Injection Surface: The skill processes user-supplied drafts, which constitutes a potential ingestion point for indirect instructions. However, because the skill has no access to sensitive capabilities like file-system writes, network operations, or tool execution, this surface does not present a security risk.
Strict Fidelity Guidelines: The instructions explicitly forbid the AI from inventing claims, statistics, or sources, which serves as a safety guardrail against hallucinations and the creation of misinformation.
Internal Evaluation Logic: The inclusion of an internal evaluation file (eval.md) ensures the AI audits its own output for quality and adherence to instructions before responding, which is a best practice for complex prompting tasks.

human-writing