agentsclimarketplace

Ccdawn evaluation

Skill CCDawn/codex-skills/skills/engineering/ccdawn-evaluation

Chinese-first Codex skill pack for intent alignment, dynamic routing, lightweight development, code review, UI design, and AI research workflows.

Install
npx -y skills add CCDawn/codex-skills --skill ccdawn-evaluation

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

  • 4 stars4 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.

What its author says it does

Copied from the file, not written here

Use when CCDawn workflow needs a Chinese-first evaluation adapter and no more specific existing skill already owns the review, debugging, PR, planning, feedback, verification, or goal task.

The file declares its own license as MIT. That is the author’s claim about this one file, and it is not the same thing as the license GitHub reports for the repository, which is listed with the other numbers below.

SKILL.md

2.6 KB, as published. Nobody here has run it

CCDawn Evaluation

目标

评价方案、流程、skill、输出质量、合理性、灵活性或下一步价值。只在没有更具体 owner 时使用；不重复 PR、项目、bug、复用研究或 planning 流程。

BRT interface

  • Context Boundary: 被评价对象、用户标准、证据来源、允许动作和已排除的专项 owner。
  • Output Contract: 简洁判断、1-3 个会改变行动的依据或建议和 Route Out。
  • Allowed Action: 只读评价；用户要求优化时路由最具体执行 owner，不在评价过程中顺手修改。
  • Success Evidence: 判断绑定具体对象、证据、影响和可验证下一步。
  • Stop Condition: 对象/目标不明、存在更具体 owner、关键证据缺失、实施越界或高风险取舍未确认。
  • Route Out: 更具体 owner、ccdawn-brt、ccdawn-planning、FAST_PATH、ccdawn-completion-summary 或 BLOCKED。

统一调用契约

  • 只处理 BRT interface 范围；不匹配时回 ccdawn-brt 或更具体 owner，复合任务不吞其他 owner。
  • 用户可见内容默认中文；保留技术字面量；只报结论、证据、风险和产出；Route Out 仅以 BRT interface 为准，末行写 下一步建议: <一个具体动作>。

评价方法

  1. 先内部确认没有更具体 owner；无需向用户输出“复用检查”。
  2. 只选会改变结论或下一动作的维度；证据已足够时停止，不固定遍历完整质量清单。
  3. 每个判断给最短证据、影响和最小有效改进，不用“更完善/更灵活”等空话。
  4. 最多给 3 个建议，按依赖、ROI 和误改风险排序；其余只在用户要求完整审计时展开。
  5. 用户已要求“继续/优化/修复”时，立即路由最具体执行 owner，并沿已授权范围连续推进；评价 skill 本身不再复制执行流程。

内部可以多视角自审，但没有 finding 不展示矩阵，也不输出固定角色、严重度体系或 SAFE_DIRECT/PLAN_THEN_EXECUTE 分类。

输出

评估结论: GOOD / ACCEPTABLE_WITH_RISK / NEEDS_CHANGE / BLOCKED
关键依据: <只保留会改变结论的证据>
建议（0-3 项）:
1. <最小动作>；原因；完成条件
剩余风险: ...
下一步建议: <一个具体动作>

评价对象不清时回 BRT 对齐；评价变成项目/PR/bug/复用专项时立即 Route Out，不在本 skill 扩写对应流程。

Keep looking

Skills are one crate of 328,083. 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.