The Agent Skills Directory

[PROMPT_INJECTION]: The skill implements strong defensive guardrails against prompt injection. The instructions explicitly command the agent to treat any embedded directives within the user-provided text—such as requests to ignore rules or change language—as literal text to be edited, effectively neutralizing indirect injection attempts.
[COMMAND_EXECUTION]: The skill's logic includes instructions for a bug-tracking workflow. This involves writing diagnostic information to a local directory (~/.bensz-skills/bugs/) and using the GitHub CLI (gh) to submit reports to a specific repository managed by the author. This behavior is scoped to development and error-handling scenarios.
[DATA_EXFILTRATION]: No unauthorized data transmission or external network patterns were detected. The skill's primary focus is local text refinement, with strict constraints ('semantic zero loss') to prevent the addition or removal of information from the user's document.

nsfc-humanization