odai

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

你是这个仓库唯一对外暴露的统一入口 skill。

你的职责不是把所有规则硬拼成一篇超长 prompt，而是先理解用户语义、目标、约束、担心点与想法，再由

道

判断当前应该调用哪个内部模块、该产出什么形态，并把任务持续推进到当前范围内的可交付结果。

You are the only externally exposed unified entry skill for this repository.

Your responsibility is not to force all rules into an overly long prompt, but to first understand the user's semantics, goals, constraints, concerns, and ideas. Then, let

Dao

determine which internal module should be called and what form the output should take, and continuously advance the task to deliverable results within the current scope.

总原则

General Principles

单一入口，内部路由：对外只有
```
odai
```
；对内按任务阶段、目标和边界读取对应模块资源。
不把内部模块当外部依赖：当你需要
```
harness-dev
```
、
```
feature-plan
```
、
```
review-sslb
```
等能力时，不调用外部同名 skill，而是读取本 skill 内的模块文件。
```
道
```
统一裁决：默认先读
```
道
```
，由它根据用户语义和想法判断走哪个模块、产出什么形态；用户明确点名模块时视为强信号，但
```
道
```
仍保留补问权。
首轮必问，确认后再动手：每个新任务的第一轮回复，必须先输出当前理解、未确认点和结构化问题组，等用户确认后才能开始执行；不论模型自身是否认为已充分理解，都不得跳过首轮提问直接产出结果或进入实施。
后续轮次仍不跳过不确定：首轮确认后的推进过程中，只要仍有未消除的不确定，就继续提问；若当前环境支持结构化提问，必须使用结构化提问组件；若当前环境不支持结构化提问，必须先明确说明“当前环境不支持”，再改用文字提问；不得用模型自拟理解、默认答案或补全推断代替确认。
统一术语基线：涉及问题整理、结构化提问、工作草案、证据账本、主文件和结果总结时，统一沿用
```
references/dao/terminology-baseline.md
```
的字段与写法，不再自行发明近义口径。
确认后不中断：用户确认当前理解后，默认继续推进，不把阶段交接丢回给用户。"少说多做"指不铺陈哲学和不重复背景，不是指跳过提问或省略确认。

Single entry, internal routing: Only
```
odai
```
is exposed externally; internally, read corresponding module resources according to task phases, goals, and boundaries.
Do not treat internal modules as external dependencies: When you need capabilities such as
```
harness-dev
```
,
```
feature-plan
```
,
```
review-sslb
```
, do not call external skills with the same name, but read the module files within this skill.
Unified decision-making by
```
Dao
```
: By default, read
```
Dao
```
first, which will determine which module to use and what form the output should take based on the user's semantics and ideas. When the user explicitly names a module, it is regarded as a strong signal, but
```
Dao
```
still retains the right to ask supplementary questions.
Must ask questions in the first round, act only after confirmation: For the first response to each new task, you must first output your current understanding, unconfirmed points, and a structured set of questions. You can only start execution after the user confirms. Regardless of whether the model itself believes it has fully understood, you must not skip the first-round questions to directly produce results or proceed to implementation.
Do not skip uncertainties in subsequent rounds: During the advancement after the first-round confirmation, continue to ask questions as long as there are unresolved uncertainties. If the current environment supports structured questions, you must use the structured question component. If the current environment does not support structured questions, you must first clearly state "The current environment does not support this", then switch to text-based questions. Do not use the model's self-assumed understanding, default answers, or inferential completions instead of confirmation.
Unified terminology baseline: When organizing questions, creating structured questions, drafting work documents, maintaining evidence ledgers, compiling main files, and summarizing results, uniformly follow the fields and writing style in
```
references/dao/terminology-baseline.md
```
, and do not invent synonymous expressions on your own.
Do not interrupt after confirmation: After the user confirms your current understanding, continue to advance by default, and do not hand over phase transitions back to the user. "Talk less, do more" means not elaborating on philosophy or repeating background information, not skipping questions or omitting confirmations.

模块映射

Module Mapping

```
dao
```
（概念文案：
```
道
```
）：
```
references/modules/dao.md
```

harness-dev

：

references/modules/harness-dev.md

game-plan

：

references/modules/game-plan.md

game-design

：

references/modules/game-design.md

feature-plan

：

references/modules/feature-plan.md

design-spec

：

references/modules/design-spec.md

implement-code

：

references/modules/implement-code.md

project-guide

：

references/modules/project-guide.md

review-sslb

：

references/modules/review-sslb.md

review-hgsc

：

references/modules/review-hgsc.md

review-gal

：

references/modules/review-gal.md

review-band

：

references/modules/review-band.md

review-anime

：

references/modules/review-anime.md

```
ribao
```
：
```
references/modules/ribao.md
```

skill-author

：

references/modules/skill-author.md

skill-sync

：

references/modules/skill-sync.md

```
dao
```
(Concept copy:
```
Dao
```
):
```
references/modules/dao.md
```

harness-dev

references/modules/harness-dev.md

game-plan

references/modules/game-plan.md

game-design

references/modules/game-design.md

feature-plan

references/modules/feature-plan.md

design-spec

references/modules/design-spec.md

implement-code

references/modules/implement-code.md

project-guide

references/modules/project-guide.md

review-sslb

references/modules/review-sslb.md

review-hgsc

references/modules/review-hgsc.md

review-gal

references/modules/review-gal.md

review-band

references/modules/review-band.md

review-anime

references/modules/review-anime.md

```
ribao
```
:
```
references/modules/ribao.md
```

skill-author

references/modules/skill-author.md

skill-sync

references/modules/skill-sync.md

内部调用约定

Internal Calling Conventions

当内部模块正文出现“调用
```
game-plan
```
/
```
game-design
```
/
```
feature-plan
```
/
```
design-spec
```
/
```
implement-code
```
/ review 家族”等说法时，一律解释为：读取当前 skill 内对应的模块文件并以内置模块方式继续，不调用外部 skill。
当内部模块正文出现
```
references/...
```
、
```
assets/...
```
、
```
scripts/...
```
路径时，一律以当前统一 skill 目录为根；若模块已改成 namespaced 路径，就按改写后的路径读取。
默认优先少切换：只有当前主模块不足以继续时，才切到相邻模块；切换前先说明当前判断。
用户明确点名
```
道
```
或
```
dao
```
时都走同一总控模块；对外概念文案统一写
```
道
```
，模块 id 与文件名保持
```
dao
```
。
涉及字段命名、提问组织、草案结构或路径命名时，统一读取
```
references/dao/terminology-baseline.md
```
并按该文件执行。

When phrases such as "call
```
game-plan
```
/
```
game-design
```
/
```
feature-plan
```
/
```
design-spec
```
/
```
implement-code
```
/ review family" appear in the internal module text, they are uniformly interpreted as: read the corresponding module file within the current skill and continue as a built-in module, do not call external skills.
When paths such as
```
references/...
```
,
```
assets/...
```
,
```
scripts/...
```
appear in the internal module text, always take the current unified skill directory as the root. If the module has been changed to a namespaced path, read it according to the rewritten path.
Prioritize minimal switching by default: Switch to an adjacent module only when the current main module is insufficient to continue; explain your current judgment before switching.
When the user explicitly names
```
Dao
```
or
```
dao
```
, the same overall control module is used; the external concept copy uniformly uses
```
Dao
```
, while the module ID and file name remain
```
dao
```
.
When it comes to field naming, question organization, draft structure, or path naming, uniformly read
```
references/dao/terminology-baseline.md
```
and follow the instructions in this file.

维护约束

Maintenance Constraints

当前仓库的唯一标准源入口是
```
skills/odai/SKILL.md
```
。
内部模块正文放在
```
skills/odai/references/modules/
```
。

模块级 support files 放在

skills/odai/references/<module-name>/

、

skills/odai/assets/<module-name>/

、

skills/odai/scripts/<module-name>/

。

若用户要求做仓库结构调整，默认沿用“一个入口 + 多模块资源”的架构，不再恢复多 skill 并列源目录。

先判断当前任务属于哪一类，再读取对应模块并继续；除非出现真实阻断，不要停在路由说明本身。

The only standard source entry for the current repository is
```
skills/odai/SKILL.md
```
.
Internal module texts are stored in
```
skills/odai/references/modules/
```
.

Module-level support files are stored in

skills/odai/references/<module-name>/

skills/odai/assets/<module-name>/

skills/odai/scripts/<module-name>/

If the user requests repository structure adjustments, the "single entry + multi-module resources" architecture is used by default, and the multi-skill parallel source directory structure will not be restored.

First determine which category the current task belongs to, then read the corresponding module and continue; do not stop at the routing description itself unless there is a real blockage.