overthink

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Follow this structured process to think through complex tasks.

遵循以下结构化流程来处理复杂任务。

Orient yourself

明确任务定位

Identify the key elements of the task:

Goal: what the user is trying to accomplish.
Failure: what would likely go wrong if answering without thinking it through.
Constraints: hard requirements, soft preferences, unknowns.
Evidence: whether facts, files, tools, examples, tests, or measurements can change the answer.
Output: what the final answer needs to be: artifact, diagnosis, recommendation, explanation, critique, plan, or direct answer.

If a missing detail would materially change the goal, hard constraints, domain, acceptable risk, or output, answer with reasonable assumptions and flag what you assumed.

Do not confuse satisfying the user's expected framing with serving the user's goal.

识别任务的关键要素：

Goal：用户想要达成的目标。
Failure：如果不经过思考就直接作答，可能会出现哪些问题。
Constraints：硬性要求、软性偏好、未知因素。
Evidence：事实、文件、工具、示例、测试或测量结果是否会改变答案。
Output：最终答案需要呈现的形式：成品、诊断结果、建议、解释、评论、计划或直接答复。

如果缺失的细节会实质性地改变目标、硬性约束、领域范围、可接受风险或输出形式，请基于合理假设作答，并标注出你的假设内容。

不要将满足用户的预期表述与服务用户的真实目标混淆。

Build a thinking process

构建思考流程

Pick and chain the thinking modes below for the task.

Do not read the files upfront, just decide the modes you will require and in what order:

```
references/model.md
```
: expose hidden assumptions and map out system boundaries before trying to solve the problem.
```
references/create.md
```
: expand the option space by generating structurally different branches and lateral approaches, not just surface variants.
```
references/diagnose.md
```
: turn symptoms into competing hypotheses to find the root cause of a failure using distinguishing evidence.
```
references/verify.md
```
: check load-bearing claims against external evidence to prevent hallucination or trusting unsupported answers.
```
references/evaluate.md
```
: systematically weigh criteria to rigorously assess an idea, artifact, or set of options.
```
references/decide.md
```
: cut through ambiguity to commit to a single recommendation, choice, or next step under constraints.
```
references/taste.md
```
: transform vague aesthetic opinions into disciplined taste judgments about form, feeling, medium, and constraint-fit.
```
references/synthesize.md
```
: combine perspectives from multiple modes to reconcile tensions, find new insights, and form a coherent final answer.

根据任务需求，选择并串联以下思考模式。

不要预先阅读文件，只需确定你需要使用的模式及其顺序：

```
references/model.md
```
：在尝试解决问题之前，揭示隐藏假设并梳理系统边界。
```
references/create.md
```
：通过生成结构不同的分支和横向方法来拓展选项范围，而不仅仅是表面变体。
```
references/diagnose.md
```
：将症状转化为相互竞争的假设，利用区分性证据找出故障的根本原因。
```
references/verify.md
```
：对照外部证据检查核心主张，防止幻觉或采信无依据的答案。
```
references/evaluate.md
```
：系统地权衡标准，严谨评估某个想法、成品或一组选项。
```
references/decide.md
```
：在约束条件下消除歧义，确定单一建议、选择或下一步行动。
```
references/taste.md
```
：将模糊的审美观点转化为关于形式、感受、媒介和约束适配性的严谨审美判断。
```
references/synthesize.md
```
：结合多种模式的视角，调和矛盾、发现新见解并形成连贯的最终答案。

Some examples

部分示例

Fixing a tricky bug:
```
diagnose
```
->
```
verify
```
Designing a new feature:
```
model
```
->
```
create
```
->
```
evaluate
```
->
```
taste
```
->
```
synthesize
```
Reviewing complex code:
```
model
```
->
```
evaluate
```
->
```
verify
```
Solving an open-ended problem:
```
model
```
->
```
create
```
->
```
evaluate
```
->
```
decide
```
Critiquing an idea or artifact:
```
model
```
->
```
evaluate
```
Making a tough call under uncertainty:
```
model
```
->
```
evaluate
```
->
```
decide
```

修复棘手bug：
```
diagnose
```
->
```
verify
```
设计新功能：
```
model
```
->
```
create
```
->
```
evaluate
```
->
```
taste
```
->
```
synthesize
```
评审复杂代码：
```
model
```
->
```
evaluate
```
->
```
verify
```
解决开放式问题：
```
model
```
->
```
create
```
->
```
evaluate
```
->
```
decide
```
评论某个想法或成品：
```
model
```
->
```
evaluate
```
在不确定性下做出艰难抉择：
```
model
```
->
```
evaluate
```
->
```
decide
```

Use a todo ledger

使用待办事项清单

Create a private todo list containing just the names of the selected modes.

Example:

[ ] model
[ ] decide
[ ] verify

Read the first mode's reference file. Based on the reference, update the todo list to add sub-items for each question you need to answer, each step you need to perform, or information to gather.

Example:

[ ] model
[ ] model: What is being treated as fixed, impossible, or irrelevant?
[ ] model: Replace a proxy metric with the underlying goal.
[ ] decide
[ ] verify

Mark a mode as done only when all its sub-items are complete or you could not complete them reliably. Then move to the next mode.

When moving to the next mode, use the former's insights, open questions, and remaining uncertainties to decide your approach.

If a mode's reference directs you to another mode you have already completed, use the insights from that prior run rather than re-running it, unless the situation has materially changed (e.g., a creative operator redefined the problem).

创建一份私人待办事项清单，仅包含所选思考模式的名称。

示例：

[ ] model
[ ] decide
[ ] verify

阅读第一个模式的参考文件。根据参考内容，更新待办事项清单，添加你需要回答的每个问题、需要执行的每个步骤或需要收集的信息作为子项。

示例：

[ ] model
[ ] model: What is being treated as fixed, impossible, or irrelevant?
[ ] model: Replace a proxy metric with the underlying goal.
[ ] decide
[ ] verify

只有当某个模式的所有子项都完成，或者你无法可靠完成它们时，才标记该模式为已完成。然后进入下一个模式。

进入下一个模式时，利用前一个模式的见解、未解决的问题和剩余的不确定性来确定你的方法。

如果某个模式的参考文件指向你已完成的另一个模式，请使用之前运行该模式得到的见解，而非重新运行，除非情况发生了实质性变化（例如，创意操作重新定义了问题）。

Subagents

Subagents（子代理）

If subagents are available, use them when:

The task requires external observation (current events, human sentiment, specific documentation, or extensive reading). This delegates information-gathering and protects your context for reasoning.
There are materially distinct angles that can be explored independently.

If explicit user permission is required to spawn subagents, pause and ask: "Do you want me to use subagents to explore separate directions in parallel before I decide?"

Whenever appropriate, continue the local critical-path work while subagents run.

如果可以使用Subagents，请在以下场景中使用：

任务需要外部观察（当前事件、人类情绪、特定文档或大量阅读）。这可以将信息收集工作委托出去，为你的推理保留上下文空间。
存在可以独立探索的实质性不同角度。

如果生成Subagents需要用户明确许可，请暂停并询问：“是否需要我使用Subagents并行探索不同方向后再做决定？”

只要合适，在Subagents运行的同时继续处理本地关键路径的工作。

Output

输出要求

Surface evidence, criteria, assumptions, and uncertainty that affect the user's action or understanding. Omit everything else.
Put caveats near the claims they limit.
Preserve nuance where flattening would mislead.
Do not make the final answer look more certain than the evidence permits. A polished explanation can make weak support feel stronger than it is.
When constraints conflict, surface the conflict or choose according to hierarchy and task goal. Do not silently pretend all constraints can be satisfied.
Ignore generic praise, motivational filler, or inflated significance.

披露影响用户行动或理解的证据、标准、假设和不确定性。省略其他所有内容。
将警告信息放在其对应的主张附近。
在简化会导致误导的地方保留细节差异。
不要让最终答案看起来比证据所能支持的更确定。经过润色的解释可能会让薄弱的支撑显得比实际更强。
当约束条件发生冲突时，披露冲突或根据优先级和任务目标进行选择。不要默默假装所有约束都能被满足。
忽略通用赞美、激励性套话或夸大的重要性。

IMPORTANT

重要提示

Do:

Follow the todo process faithfully, even if it seems slow or verbose.
Self-evaluation is a filter, not proof. Prefer external signals when available: tests, logs, calculations, code execution, sources, quotes, examples, measurements, observed behavior, constraints from the prompt, or user-provided evidence.

Don't:

Do NOT read multiple mode files at once.
Do NOT optimize for delivering a final answer faster.
Do NOT ignore the explicit instructions in this skill file unless user or system instructions say otherwise.

请务必：

严格遵循待办事项流程，即使它看起来缓慢或繁琐。
自我评估只是一个筛选器，而非证明。如有可用的外部信号，请优先使用：测试、日志、计算、代码执行、来源、引用、示例、测量结果、观察到的行为、提示中的约束或用户提供的证据。

请勿：

不要同时阅读多个模式文件。
不要为了更快交付最终答案而进行优化。
不要忽略本技能文件中的明确指令，除非用户或系统另有说明。