Back to Details

skill-auto-update

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Skill Auto Update

技能自动更新

Use this skill to keep project skills aligned with what feedback keeps teaching.
使用此技能可确保项目技能与反馈持续传递的内容保持一致。

Activation

激活方式

This skill is most effective when the workspace 
AGENTS.md
tells the AI to use it on every turn. Add a short rule such as: 
Use $skill-auto-update on every turn to distill durable feedback into the narrowest existing skill.
Treat routine use as a silent background check. Do not announce that you are using this skill unless it produces a user-relevant skill edit or the user explicitly asks about it.
当工作区的
AGENTS.md
要求AI在每一轮交互中使用此技能时，该技能的效果最佳。添加一条简短规则，例如：
在每一轮交互中使用$skill-auto-update，将持续的反馈提炼到最贴合的现有技能中。
将常规使用视为后台静默检查。除非产生了与用户相关的技能编辑，或是用户明确询问，否则无需告知用户你正在使用此技能。

Run this loop

执行循环

  1. Scan the current turn for durable feedback signals such as user corrections, review comments, comments on issues or pull requests, repeated instructions, and explicit "do this / avoid this" guidance.
  2. Ignore one-off facts, task-local preferences, and literal examples that do not imply a reusable rule.
  3. Rewrite the signal as a principle: capture the idea, not the anecdote; state the invariant in one short sentence; prefer imperative wording such as 
    Prefer
    , 
    Keep
    , 
    Move
    , or 
    Avoid
    .
  4. Translate taste or persona goals into concrete engineering rules; prefer 
    Keep functions small and cohesive
    over 
    code like Uncle Bob
    .
  5. Update the narrowest existing skill that owns the rule. Prefer updating an existing skill over creating a new one.
  6. Search the target skill for the same idea, including paraphrases and nearby rules.
  7. If the idea already exists, do not duplicate it. Tighten the existing rule only when the new feedback adds missing scope, rationale, or priority.
  8. When a new rule subsumes older examples, remove or compress the examples instead of keeping both.
  9. If the same rule is missed again, increase its score instead of adding another bullet.
  10. Keep the edit minimal. Prefer one changed bullet over new sections, long explanations, copied examples, repeated examples, or checklists of symptoms.
  11. If no durable lesson exists, make no skill edit and do not mention the check.
  1. 扫描当前轮次中的持续反馈信号，例如用户修正、评审意见、问题或拉取请求中的评论、重复指令，以及明确的“执行/避免”指导。
  2. 忽略一次性事实、任务本地偏好，以及不涉及可复用规则的字面示例。
  3. 将信号重写为原则：捕捉核心思想而非具体轶事；用一句话陈述不变规则；优先使用祈使语气，例如
    优先
    、
    保持
    、
    迁移
    或
    避免
    。
  4. 将品味或角色目标转化为具体的工程规则；例如，优先使用
    保持函数小巧且内聚
    而非
    像Uncle Bob那样编码
    。
  5. 更新包含该规则的最贴合的现有技能。优先更新现有技能而非创建新技能。
  6. 在目标技能中搜索相同的思想，包括同义表述和相关规则。
  7. 如果该思想已存在，则无需重复。仅当新反馈补充了缺失的范围、理由或优先级时，才优化现有规则。
  8. 当新规则涵盖旧示例时，移除或压缩旧示例，而非同时保留两者。
  9. 如果同一规则再次被疏漏，则提高其分数，而非添加新的项目符号。
  10. 保持编辑最小化。优先修改单个项目符号，而非添加新章节、长篇解释、复制示例、重复示例或症状清单。
  11. 如果不存在可复用的经验教训，则不进行技能编辑，也无需提及此次检查。

Score rules

分数规则

  • Treat score as an optional weight for repeated misses.
  • Use no score marker for a fresh rule unless recurrence already matters.
  • Append 
     [score: N]
    to the rule when increasing weight.
  • Start at 
    2
    for the second clear miss and increment by one for later misses.
  • Cap scores at 
    5
    .
  • Do not raise score for a mere style preference.
  • 将分数视为重复疏漏的可选权重。
  • 对于新规则，除非重复问题已很重要，否则不添加分数标记。
  • 当需要增加权重时，在规则后追加
     [score: N]
    。
  • 第二次明确疏漏时起始分数为
    2
    ，后续每多一次疏漏加1分。
  • 分数上限为
    5
    。
  • 仅因风格偏好的情况，不提高分数。

Rule shape

规则格式

  • Good: 
    Keep shared components data-driven; move module-specific policy back to the owning feature. [score: 3]
  • Bad: 
    Do not repeat the status-action-button mistake from the custom module table.
  • 规范示例：
    保持共享组件数据驱动；将模块专属策略移回所属功能模块。[score: 3]
  • 错误示例：
    不要重复自定义模块表格中状态操作按钮的错误。

Editing standard

编辑标准

  • Keep skill text concise and imperative.
  • Avoid historical commentary, changelogs, and case-specific prose.
  • Prefer strengthening an existing rule over adding a sibling rule that says the same thing differently.
  • Prefer general rules over examples; add at most one generic example only when the rule would otherwise be ambiguous.
  • When a rule and an example say the same thing, keep the rule and delete the example.
  • Do not multiply examples for the same rule; replace them with one generic example or none.
  • 保持技能文本简洁且使用祈使语气。
  • 避免历史注释、变更日志和针对特定案例的描述。
  • 优先强化现有规则，而非添加表述不同但意思相同的同级规则。
  • 优先使用通用规则而非示例；仅当规则会存在歧义时，最多添加一个通用示例。
  • 当规则和示例表述的内容相同时，保留规则并删除示例。
  • 不为同一规则添加多个示例；替换为一个通用示例或不添加示例。