agentsclimarketplace

Course quality editor

Skill Ivor-NCUT/course-producer/skills/course-quality-editor

对整门课程和单章执行认知交付、模块顺序、逻辑连续、信息密度、口播节奏、事实证据与 AI 写作坏味道审校,生成 review.jsonl 和保留讲师立场的修订稿。Use when the user says 课程审校、检查课程逻辑、去 AI 味、检查逐字稿、精修课程、检查事实证据,或 Course Producer 进入 quality_review 阶段。From its SKILL.md

Install
npx -y skills add Ivor-NCUT/course-producer --skill course-quality-editor

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

3.7 KB, ~1.2k tokens by cl100k_base, as published. Nobody here has run it

课程质量审校

围绕已有稿件审校和修订,不另起炉灶重写整门课。开始前读取课程定位、蓝图、知识卡、全部章节,以及 知识库/方法论/课程审校.md

审校顺序

  1. 运行 node tools/course-lint.mjs <lessons-dir> --output <review.jsonl>,得到可复现的禁用句式、模板转场、字幕式断行和证据提示。
  2. 做课程级检查:模块是否按学习成果排序,第一、第二模块能否独立交付结果,章节是否螺旋复用,承诺是否越过定位边界。
  3. 做章节级检查:教学任务、认知起终点、挑战情境、段落到段落的因果/动作连接、信息密度、术语解释和前后桥接。
  4. 核查事实证据:数字、案例、背书、高风险承诺是否对应 knowledge_card_id 与来源;找不到时标为补材料,不能润色成确定事实。
  5. 对每个问题选择动作:
    • auto_fix:不改变观点的措辞、重复、缺桥接、段落节奏与明确格式问题;
    • needs_evidence:缺数据、案例、授权或原始来源;
    • needs_decision:课程承诺、品牌立场、目标人群、价格、公开风险或证据冲突需要用户判断。
  6. 先复制原稿到 .course-producer/artifacts/revised-lessons/ 再应用 auto_fix。不覆盖已确认原稿,除非用户明确要求原位修改。
  7. 把所有发现逐行写入 .course-producer/artifacts/review.jsonl;修复后更新 statusrevision_locator,保留问题原始证据。
  8. 重新运行确定性扫描并做课程级复核。高风险问题未解决时不能把 quality_review 标记完成。

review.jsonl 契约

{"review_id":"rev-001","scope":"lesson","artifact":"lessons/01.md","locator":"## 方法","category":"logic_gap","severity":"high","evidence":"上一段产物没有进入本段动作","action":"auto_fix","status":"fixed","suggestion":"补充产物到动作的桥接","revision_locator":"revised-lessons/01.md#方法"}

category 可使用 course_orderlogic_gaprepetitiondensityspoken_flowai_smellunexplained_termunsupported_claimposition_conflict。不要输出主观“AI 概率”。

保真边界

  • 可以改结构、顺序、句群、转场、重复和术语解释。
  • 不改变讲师观点、战略立场、案例因果、数字口径和课程承诺。
  • 对观点有疑问时写 needs_decision,不要用审校者立场替换。
  • 逐字原话只在确认是引用错误时改;普通口语不因“不够精致”被抹平。

完成检查

  • 课程级与章节级审校都已执行。
  • review 每条有位置、具体证据、严重度、动作和状态。
  • 自动修复后的稿件可回放,原稿仍可恢复。
  • 数字、案例、背书和高风险承诺有来源或明确待补。
  • 没有用概率分数代替具体文本证据。
  • 修订没有改变讲师观点和战略立场。

完成后可进入 course-lark-delivery;用户对修订结果反馈时按审美对齐入口处理。

Feedback Learning

用户对修订稿给出反馈、亲自改稿、选择版本或明确满意时,先判断是否只是当前作品修改。用户确认要沉淀后,调用 course-aesthetic-alignment 记录 edit trace、正反例或 preference pair;单次满意不能直接更新本 Skill 的审校规则。

What ships with it: 1 file

273 B alongside SKILL.md

agents/

Keep looking

Skills are one crate of 325,949. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.