culture

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

/culture

Two modes:

Internal mode (default). Other workflow skills — and
```
/cheese
```
itself — invoke
```
/culture
```
silently as a thinking pass: restate the question, list assumptions, name candidate options, run a quick shape check, pick the next action. The dialogue does not surface to the user; only the resulting decision (and any code edits the calling skill makes) does. The most common callers are:
- ```
/cheese
```
  step 1 — silent classification reasoning before announce.
- ```
/cheese
```
  tier-2 escalation (see
```
skills/cheese/SKILL.md
```
  § Escalation) — fills missing context when the cook-fast-path clarity check fails on the raw input; the synthesis lands in the mini-spec's
```
## Provenance
```
  section.
- Other workflow skills' own pre-dispatch reasoning passes (mold, cook taste-test, etc.).
User-facing mode. The user has explicitly opted out of writes for this session. Conversation is the deliverable; no code, no spec, no PR. Reach this mode only when the user said "no writes" / "rubber-duck this" / "just talk" or equivalent.

Do not use the user-facing mode when the user wants a written spec (

/mold

), implementation (

/cook

), review (

/age

), or external evidence gathering (

/briesearch

) — those targets get the internal-mode call instead, and the calling skill does the work.

两种模式：

内部模式（默认）。其他工作流技能——包括
```
/cheese
```
本身——会静默调用
```
/culture
```
作为思考环节：重述问题、列出假设、列举候选方案、快速进行形态检查、选择下一步行动。这段对话不会展示给用户；只有最终的决策（以及调用技能所做的任何代码修改）会呈现给用户。最常见的调用者包括：
- ```
/cheese
```
  第一步——在告知用户之前进行静默分类推理。
- ```
/cheese
```
  二级升级（详见
```
skills/cheese/SKILL.md
```
  § 升级）——当快速处理路径的清晰度检查在原始输入上失败时，补充缺失的上下文；合成内容会放入迷你规范的
```
## Provenance
```
  部分。
- 其他工作流技能自身的预调度推理环节（mold、cook尝试验证等）。
面向用户模式。用户明确选择在本次会话中不进行写入操作。交付物为对话内容；不生成代码、规范文档或PR。只有当用户说出“no writes”/“rubber-duck this”/“just talk”或类似表述时，才进入此模式。

当用户需要书面规范（

/mold

）、代码实现（

/cook

）、代码评审（

/age

）或外部证据收集（

/briesearch

）时，请勿使用面向用户模式——这些场景应调用内部模式，由调用技能完成相应工作。

Invariant

不变规则

/culture

does not write production code, commit changes, open PRs, or mutate project state. The only sanctioned artifact is the opt-in notes handoff at

.cheese/notes/<slug>.md

(see

## Handoff slug

below), written only at session end of a user-facing session and only when the user asks for notes — never during dialogue, never in internal mode. If a user-facing conversation reveals that something should be built, route to

/mold

/cook

and stop. In internal mode, just return the decision and let the calling skill act.

/culture

不会写入生产代码、提交变更、创建PR或修改项目状态。唯一被允许生成的产物是可选的会话结束时的交接文件

.cheese/notes/<slug>.md

（详见下文

## 交接文件

），仅在面向用户的会话结束且用户要求记录时才会生成——对话过程中或内部模式下绝不会生成。如果面向用户的对话中发现需要构建的内容，引导至

/mold

或

/cook

并停止当前操作。在内部模式下，仅返回决策结果，由调用技能执行后续操作。

Flow

流程

Both modes share the same reasoning loop. The difference is what the agent does with the output: internal mode returns a single decision back to the calling skill; user-facing mode renders the dialogue and waits for the user.

Restate the question or tension in one sentence. If the question rests on a false premise or a loaded assumption, name it.
Identify assumptions, constraints, and decision criteria.
Explore trade-offs and likely blast radius. When the trade-off hinges on "what does this touch", run a read-only shape check on the candidate seam — a
```
cheez-search
```
callers query (
```
tilth_search kind: "callers"
```
) plus
```
tilth_deps
```
— and label each option
```
[low | medium | high blast radius]
```
. Procedure mirrors
```
../mold/references/shape-check.md
```
; culture stops at the verdict and never drafts signatures. Steelman the rejected option before settling on a recommendation.
Use evidence only when it helps the reasoning; avoid deep research unless explicitly asked.
Converge on a single recommended next action. In internal mode, return that recommendation and stop. In user-facing mode, render a compact summary, open questions tagged with confidence (
```
certain | speculating | don't know
```
), and a
```
## Handoff
```
prompt (see below).

Default the model's own contribution to maximum useful depth — full pseudocode signatures over hand-waving, named edge cases over "consider edge cases", concrete file:line evidence over vague pointers. Smallest-useful-question discipline applies only to what you ask the user, never to what you offer them.

两种模式共享相同的推理循环。区别在于Agent对输出结果的处理方式：内部模式将单个决策结果返回给调用技能；面向用户模式则展示对话内容并等待用户反馈。

用一句话重述问题或核心矛盾。如果问题基于错误前提或隐含假设，需明确指出。
识别假设条件、约束因素和决策标准。
分析权衡因素及可能的影响范围。当权衡取决于“该内容会影响哪些部分”时，对候选衔接点进行只读形态检查——调用
```
cheez-search
```
查询（
```
tilth_search kind: "callers"
```
）并结合
```
tilth_deps
```
——为每个选项标记
```
[低 | 中 | 高影响范围]
```
。流程与
```
../mold/references/shape-check.md
```
一致；culture环节仅得出结论，绝不会草拟签名。在确定推荐方案前，需充分论证被否决的选项。
仅在有助于推理时使用证据；除非明确要求，避免深入研究。
收敛至单个推荐的下一步行动。在内部模式下，返回该推荐并停止操作。在面向用户模式下，生成简洁的总结、带有置信度标签（
```
certain | speculating | don't know
```
）的待解决问题，以及
```
## 交接
```
提示（详见下文）。

默认模型输出应达到最大实用深度——提供完整的伪代码签名而非模糊描述，明确列出边缘案例而非仅提示“考虑边缘案例”，给出具体的文件:行证据而非模糊指向。最小有效问题原则仅适用于向用户提出的问题，不适用于向用户提供的内容。

Preferred tools and fallbacks

首选工具及备选方案

Code search and reading go through the

cheez-*

skills (

/cheez-search

/cheez-read

) — see those skills for tool selection rules. Blast-radius reads specifically use

cheez-search

callers (

kind: "callers"

) plus

tilth_deps

(read-only shape check); culture stops at the verdict and never drafts signatures.

Beyond

cheez-*

there are culture-specific tools:

Need	Prefer	Fallback
Visualizing diffs or examples	`delta`	plain `git diff`
External sanity check	`/briesearch`	clearly mark as an assumption

Missing optional tools should not interrupt the conversation. Keep tool use light; this is a thinking session.

代码搜索和阅读通过

cheez-*

技能（

/cheez-search

、

/cheez-read

）完成——详见这些技能的工具选择规则。影响范围读取专门使用

cheez-search

调用者查询（

kind: "callers"

）并结合

tilth_deps

（只读形态检查）；culture环节仅得出结论，绝不会草拟签名。

除

cheez-*

外，还有culture专属工具：

需求	首选工具	备选方案
可视化差异或示例	`delta`	普通 `git diff`
外部合理性校验	`/briesearch`	明确标记为假设

缺失可选工具不应中断对话。工具使用应尽量精简；这是一个思考会话。

Output

输出

Return a short conversational summary:

Current understanding
Trade-offs or options
Open questions

返回简短的对话式总结：

当前理解
权衡因素或可选方案
待解决问题

Handoff slug

交接文件

When (and only when) the user asks for notes at session end, write an optional handoff slug to

.cheese/notes/<slug>.md

. The slug uses the same minimum schema as the other phase handoffs so

/cheese --continue <slug>

can read it:

markdown

status: ok | halt: <one-line reason>
next: mold | cook | ultracook | stop
artifact: <path-if-any>
<one-line orientation: what the culture session converged on>

next:

is culture-specific — values are

mold

(fuzzy idea, route to spec),

cook

(clear ask, route to implementation),

ultracook

(clear ask with high blast radius, route to autonomous fresh-context chain), or

stop

(no further action). The orientation line captures the punchline of the dialogue in one factual sentence; deeper notes go in the body of the same file.

The notes slug is the only thing culture is allowed to write. No commits, no PRs, no production-code edits — those route to

/mold

/cook

当（且仅当）用户在会话结束时要求记录时，生成可选的交接文件至

.cheese/notes/<slug>.md

。该文件使用与其他阶段交接文件相同的最小化 schema，以便

/cheese --continue <slug>

可以读取：

markdown

status: ok | halt: <单行原因>
next: mold | cook | ultracook | stop
artifact: <如有路径则填写>
<单行定位：culture会话达成的核心结论>

next:

是culture专属字段——可选值为

mold

（模糊想法，引导至规范文档）、

cook

（明确需求，引导至代码实现）、

ultracook

（明确需求且高影响范围，引导至自主新上下文链）或

stop

（无进一步操作）。定位行用一句事实性语句概括对话的核心要点；更详细的记录放在同一文件的正文中。

该记录文件是culture唯一被允许生成的内容。禁止提交变更、创建PR或修改生产代码——这些操作需引导至

/mold

或

/cook

。

Handoff

交接

Pipeline: [culture] → mold → cook → press → age → cure → ship

When the conversation reveals real work, ask via the shared handoff gate in

../../shared/handoff-gate.md

. Lead each option with the verb (what the user wants to do next); the skill command is the backing detail. Before asking, render a compact context packet so the downstream skill can dispatch without losing the discussion:

yaml

handoff_context:
  source_skill: /culture
  summary: <one factual sentence>
  open_questions: [<only blockers, if any>]
  artifact: <.cheese/notes/<slug>.md if written, otherwise none>

Default options (pick at most three of these plus a stop):

Shape this into a written spec (recommended when the idea is still fuzzy) —
```
/mold
```
with the context packet, or
```
/mold .cheese/notes/<slug>.md
```
when a notes slug exists.
Implement it directly (recommended when the ask is clear and unambiguous) —
```
/cook
```
with the context packet as the accepted contract.
Implement and auto-review —
```
/cook --auto
```
with the context packet, chains through
```
/press → /age → /cure
```
autonomously, fixing every medium-or-above finding plus cheap (contained-fix) lows across up to two cure passes. Stops at the final cure pass; opening or updating the PR stays a manual step. Pre-select this when the conversation reached an unambiguous contract; offer the non-auto
```
/cook
```
as an alternative when the user wants per-step approval.
Research more first (when the conversation hit a factual gap external docs could close) —
```
/briesearch
```
.
Pause — dispatch none; keep the dialogue in head.

After a non-stop selection, run the selected downstream skill immediately with the context packet.

/age

is never the next step from culture — review needs a diff to look at.

流程管线： [culture] → mold → cook → press → age → cure → ship

当对话中涉及实际工作任务时，通过

../../shared/handoff-gate.md

中的共享交接入口发起请求。每个选项以动词开头（用户接下来想要执行的操作）；技能命令作为支持细节。发起请求前，生成简洁的上下文数据包，以便下游技能可以调度而不丢失之前的讨论内容：

yaml

handoff_context:
  source_skill: /culture
  summary: <一句事实性语句>
  open_questions: [<仅包含阻塞性问题（如有）>]
  artifact: <若已生成记录文件则填写`.cheese/notes/<slug>.md`，否则填写none>

默认选项（最多选择其中三个加上“暂停”选项）：

整理为书面规范文档（当想法仍模糊时推荐）——携带上下文数据包调用
```
/mold
```
，若已生成记录文件则调用
```
/mold .cheese/notes/<slug>.md
```
。
直接实现（当需求明确且无歧义时推荐）——携带上下文数据包调用
```
/cook
```
，将其作为已确认的约定。
实现并自动评审——携带上下文数据包调用
```
/cook --auto
```
，自动串联
```
/press → /age → /cure
```
流程，修复所有中高优先级问题以及最多两次cure流程中可低成本修复（范围可控）的低优先级问题。在最终cure流程结束时停止；创建或更新PR仍为手动步骤。当对话达成明确约定时默认选择此选项；若用户希望逐步骤确认，则提供非自动的
```
/cook
```
作为替代选项。
先进行更多研究（当对话遇到可通过外部文档填补的事实空白时）——调用
```
/briesearch
```
。
暂停——不调度任何技能；保持对话状态。

选择非“暂停”选项后，立即携带上下文数据包运行选定的下游技能。

/age

永远不会是culture的下一步——评审需要查看差异内容。

Rules

规则

No production-code writes, no commits, no PRs. The only sanctioned write is the opt-in
```
.cheese/notes/<slug>.md
```
handoff at session end, and only when the user asks for it.
Ask one useful question at a time when the user is exploring.
Prefer clarity over completeness.
Agree when agreement is warranted; do not manufacture counterpoints to seem balanced.
When external evidence raises an alternative ("X uses Y or Z"), name it as a trade-off in the dialogue and a candidate option — never silently recommend "add both" or "expose a knob". Design choices need explicit user adjudication, not agent inference from a citation.
Apply the shared voice kernel (lives at
```
skills/age/references/voice.md
```
in this repo): lead with the answer, flag confidence as
```
certain | speculating | don't know
```
, steelman, track contradictions across turns.

禁止写入生产代码、提交变更或创建PR。唯一被允许的写入操作是会话结束时的可选
```
.cheese/notes/<slug>.md
```
交接文件，且仅在用户要求记录时生成。
当用户在探索需求时，一次只提一个有用的问题。
优先保证清晰性而非完整性。
当需要达成共识时明确同意；不要为了显得客观而刻意制造反对意见。
当外部证据提出替代方案（“X使用Y或Z”）时，需在对话中作为权衡因素和候选选项明确指出——绝不要静默推荐“同时添加两者”或“提供配置开关”。设计选择需要用户明确裁决，而非Agent根据引用内容自行推断。
应用统一语音准则（位于本仓库的
```
skills/age/references/voice.md
```
）：先给出答案，标记置信度为
```
certain | speculating | don't know
```
，充分论证对立观点，跟踪多轮对话中的矛盾点。