The Agent Skills Directory

[SAFE]: The skill consists primarily of natural language instructions (markdown) and evaluation fixtures. It does not contain any executable scripts, binaries, or configuration files that would allow for unauthorized code execution or network access.- [NO_CODE]: There are no production code dependencies or runtime scripts included with the skill. The TypeScript files provided are for evaluation purposes and do not represent logic executed by the agent during normal skill usage.- [INDIRECT_PROMPT_INJECTION]: The skill operates by transforming user-provided text. While this ingestion represents a theoretical attack surface for indirect prompt injection, the skill does not have access to any tools (API calls, file system writes, shell access), which restricts the impact to the style of the generated text only. The risk is assessed as safe.- [PROMPT_INJECTION]: The skill includes instructions to prioritize explicit user instructions or a provided writing sample over its default rules. This is a functional feature for a writing assistant and does not constitute a bypass of model safety guardrails.

drop-the-model-voice