knowledge-work

Structure task-oriented responses so the reader can act immediately. Load by default when the user must perform a procedure, troubleshoot, choose, recover from an error, or continue unfinished work. Do not wait for explicit ADHD wording, and never diagnose or label the user.

中文 AI 味去除与文章润色 Skill。适用于把 AI 生成稿改得更自然、更有读者感，同时保留原意、逻辑和关键事实。

Route prose editing to this plugin's first-party writing Skills. Use for human-readable Markdown or chat prose, not code or machine output.

Orchestrate hard research with host-native web discovery, MCP capture/anchors, typed claims, a fresh seal, and post-seal outbound handoff. Use for multi-source or evidence-backed research without provider API keys or standalone search CLIs.

Use the smallest useful text diagram, tree, diff, table, or Mermaid view to explain relationships, sequence, hierarchy, state changes, or change shape. Use it when a visual materially improves understanding; do not trigger for simple facts, one-step actions, or ordinary lists.

员工要生成日报、周报、阶段总结、工作复盘或改进计划时使用；统一编排本地证据、员工确认、成长承诺、TL 验证和完整性封印。

Fill evidence gaps in a work report with a bounded one-question-at-a-time interview. Use during daily, weekly, or summary authoring when data gaps remain; do not use as a standalone grilling session or to write the saved report.

TL 要审阅员工日报、周报、阶段总结、承诺兑现或考核证据时使用；只读检查事实、证据强度、改进闭环和待验证项。

检查和清理中英文文本里的 AI 套路，适用于“去 AI 味”“说人话”“自然一点”“别像模板”“先标问题”这类改写和审稿需求；按场景控制力度，同时保留事实、术语、语域和责任主体。

Remove signs of AI-generated writing from text. Use when editing or reviewing text to make it sound more natural and human-written. Based on Wikipedia's comprehensive "Signs of AI writing" guide. Detects and fixes patterns including: inflated symbolism, promotional language, superficial -ing analyses, vague attributions, em dash overuse, rule of three, AI vocabulary words, passive voice, negative parallelisms, and filler phrases.

Review and rewrite Chinese or English prose to reduce AI-generated patterns. Use for de-AI polishing, natural-language rewrites, robotic or formulaic writing, and publication cleanup. Run the bundled deterministic Markdown analyzer before editing, inspect every reported location, then read the full article for semantic patterns the rules cannot enumerate. The installed PostToolUse Hook independently rescans observed Markdown writes.

Ultra-compressed communication. Use only when the user explicitly asks for fewer tokens, caveman mode, or brief output.