penpot-foundations

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

penpot-foundations — token & library foundation

penpot-foundations — token 与库基础

1. Title + How it works

1. 标题与工作原理

penpot-foundations

is the load-bearing skill of the kit: nothing downstream is correct without a governed token layer. Every mutation goes through

execute_code

; validate visually with

export_shape

; read structure with

penpotUtils.shapeStructure

(full tool surface:

shared/penpot-mcp-tool-reference.md

). This skill reads tokens with

penpotUtils.tokenOverview()

, creates them through

penpot.library.local.tokens

(

addSet

set.addToken({type,name,value})

), and binds them to shapes with

shape.applyToken(token, properties)

. It works in three tiers — primitive → semantic → component (see

shared/tokens-schema.json

) — and can also infer a token system from an existing hardcoded design.

penpot-foundations

是本工具包的核心支撑技能：没有规范化的token层，下游所有内容都无法保证正确性。所有变更都通过

execute_code

执行；使用

export_shape

进行视觉验证；通过

penpotUtils.shapeStructure

读取结构（完整工具说明见

shared/penpot-mcp-tool-reference.md

）。本技能通过

penpotUtils.tokenOverview()

读取token，通过

penpot.library.local.tokens

（

addSet

、

set.addToken({type,name,value})

）创建token，并通过

shape.applyToken(token, properties)

将token绑定到形状。它采用三层架构——基础层 → 语义层 → 组件层（见

shared/tokens-schema.json

），还可以从现有硬编码设计中推断出token系统。

2. The One Rule That Matters Most

2. 最重要的核心规则

Tokens before everything, and never one-shot. Build the foundation in small, idempotent steps: primitives first, then semantic aliases that reference them, then (only if needed) component tokens. Validate after each tier with

tokenOverview()

. Never invent values silently — propose the scale and let a human approve it at the checkpoint.

Token优先于一切，且绝不能一次性完成。 以小型、幂等的步骤构建基础：先做基础token，再做引用它们的语义别名，最后（仅在需要时）做组件token。每完成一层都用

tokenOverview()

验证。绝不能凭空创造值——提出刻度方案，由人工在检查点确认。

3. Penpot MCP Tool Reference

3. Penpot MCP 工具参考

Full surface:

shared/penpot-mcp-tool-reference.md

. Domain calls this skill leans on:

penpotUtils.tokenOverview()

shapeStructure()

(discovery + inference);

penpot.library.local.tokens.addSet

set.addToken

shape.applyToken

(writes);

penpot_api_info('TokenSet'|'Token'|'TokenCatalog')

before relying on member signatures.

完整接口说明见

shared/penpot-mcp-tool-reference.md

。本技能主要依赖的领域调用：

penpotUtils.tokenOverview()

shapeStructure()

（发现与推断）；

penpot.library.local.tokens.addSet

set.addToken

shape.applyToken

（写入操作）；在使用成员签名前，可调用

penpot_api_info('TokenSet'|'Token'|'TokenCatalog')

查看详情。

4. Plugin API Essentials

4. 插件 API 核心要点

Gotcha numbers refer to

shared/plugin-api-gotchas.md

#8 token creation —

penpot.library.local.tokens

;

addSet({name})

addTheme({group,name})

take OBJECT args; new sets are INACTIVE (

set.toggleActive()

); token

value

is always a string (literal or

"{reference}"

); read

token.resolvedValue

Use the exact token

type

strings from

shared/tokens-schema.json

color

dimension

spacing

typography

shadow

opacity

borderRadius

borderWidth

fontWeights

fontSizes

fontFamilies

letterSpacing

textDecoration

textCase

. (Note camelCase/plural — NOT

border-radius

font-size

#2 async token application — this skill's critical failure mode.
```
shape.applyToken(token, properties)
```
is async (~100 ms) and binds by NAME (bindings re-resolve when the active set/theme changes — that is what makes dark mode work). Apply in one
```
execute_code
```
call, verify
```
shape.tokens
```
in a LATER call. Never mass-apply in a single call — hundreds of
```
applyToken
```
s in one loop race and scramble bindings; apply in chunks of ≈25–40 shapes per call, verify each chunk, retry misses. Text/Rectangle fill bindings are the flakiest (Boards are reliable); bindings are sticky — overwrite a wrong one by applying the correct token, don't try to clear it via
```
fills
```
.

Verify any unfamiliar member with

penpot_api_info('TokenSet')

penpot_api_info('Token')

first.

注意事项编号对应

shared/plugin-api-gotchas.md

文件。

#8 token 创建 — 通过
```
penpot.library.local.tokens
```
操作；
```
addSet({name})
```
/
```
addTheme({group,name})
```
接收对象参数；新创建的集合默认处于未激活状态（需调用
```
set.toggleActive()
```
激活）；token 的
```
value
```
始终为字符串（字面量或
```
"{reference}"
```
引用格式）；可读取
```
token.resolvedValue
```
获取解析后的值。

请使用

shared/tokens-schema.json

中定义的精确token

type

字符串：

color

、

dimension

、

spacing

、

typography

、

shadow

、

opacity

、

borderRadius

、

borderWidth

、

fontWeights

、

fontSizes

、

fontFamilies

、

letterSpacing

、

textDecoration

、

textCase

。（注意是驼峰命名/复数形式——不是

border-radius

或

font-size

。）

#2 异步token应用——本技能的关键故障点。
```
shape.applyToken(token, properties)
```
是异步操作（约100毫秒），且按名称绑定（当激活的集合/主题变化时，绑定会重新解析——这是深色模式生效的原理）。请在一次
```
execute_code
```
调用中完成应用，在后续调用中验证
```
shape.tokens
```
。绝不要在单次调用中批量应用——循环中执行数百个
```
applyToken
```
会引发竞态，导致绑定混乱；请按每次调用约25-40个形状的批次应用，每批验证后重试失败项。文本/矩形的填充绑定最不稳定（画板绑定是可靠的）；绑定是持久的——如果绑定错误，直接应用正确的token覆盖即可，不要尝试通过
```
fills
```
清除。

遇到不熟悉的成员，请先调用

penpot_api_info('TokenSet')

penpot_api_info('Token')

确认。

5. Token-Aware Brief Contract

5. 具备Token意识的需求约定

Before mutating, restate the request as:

Context — product, brand, existing system (if any), light/dark needs.
Objective — single and specific (e.g., "create primitive + semantic color/spacing tokens and a dark theme").
Inputs — existing tokens, brand palette, type scale ratio (default Minor Third 1.2), spacing grid (default 4px).
Constraints — no hardcoded values; semantic tokens must reference primitives; spacing on the 4px grid.
Acceptance Criteria — every token resolves; semantic tier references primitives; all spacing is a 4px multiple; (if requested) a working dark theme.

Act as a senior design-systems engineer who never hardcodes values.

在进行变更前，请将需求重述为以下部分：

上下文 — 产品、品牌、现有系统（如有）、浅色/深色模式需求。
目标 — 单一且明确（例如："创建基础+语义颜色/间距token和深色主题"）。
输入 — 现有token、品牌调色板、字号比例（默认小三度 1.2）、间距网格（默认4px）。
约束 — 不允许硬编码值；语义token必须引用基础token；间距遵循4px网格。
验收标准 — 所有token都能解析；语义层引用基础层；所有间距都是4px的倍数；（如要求）深色主题可正常工作。

请扮演从不硬编码值的资深设计系统工程师。

6. Mandatory Workflow

6. 强制工作流

Phase 0 — Discovery (read-only).

high_level_overview

, then

penpotUtils.tokenOverview()

and a

shapeStructure(penpot.currentPage.root, 3)

read. Record what already exists in the

RUN_ID

ledger. ✋ Checkpoint: confirm scope (new system vs. extend vs. infer-from-design).

Phase 1 — Primitives. Create the

primitives

set and activate it (

set.toggleActive()

— new sets are created inactive), then add raw color ramps, the spacing scale (4px grid), radius scale, and type scale (font sizes via Minor Third). Use

scripts/createTokenSet.js

. ✋ Checkpoint: review the primitive scales (

tokenOverview()

+ naming).

Phase 2 — Semantic. Add intent aliases that reference primitives, routed into two sets (the referenced

primitives

set must be active or references fail validation):

colour semantics (

color.action.primary.bg

→

{color.blue.500}

) → modes/light
;

mode-invariant semantics (
```
spacing.*
```
,
```
radius.*
```
,
```
font.*
```
) → semantic
.

scripts/createSemanticTokens.js

does this split by token type. This is what makes dark mode a clean parallel colour set. ✋ Checkpoint: review semantic mapping.

Phase 3 — Dark mode (optional). Create the parallel modes/dark
colour set (same colour names as

modes/light

, dark values) with

scripts/createThemes.js

; leave

modes/light

active as default and hand the toggle to the user. Cover every

modes/light

colour name (the script returns a

missing

list). Dark mode works — bindings re-resolve by token name when the user activates

modes/dark

— but the agent cannot render the switch from the plugin (

theme.addSet()

doesn't persist; the plugin's read lags). The user toggles

modes/dark

on /

modes/light

off in Penpot's Tokens panel (see

references/04-themes-modes.md

). ✋ Checkpoint: verify dark

resolvedValue

s, restore light, and tell the user how to switch.

Phase 4 — Apply / Infer / Bind. Either bind tokens to a selection (

scripts/applyTokensToShapes.js

), infer tokens from a hardcoded design (

scripts/inferTokensFromShapes.js

), or — fundamental — auto-bind tokens to every unbound element by exact value match so the user doesn't have to do it by hand (

scripts/bindTokensToElements.js

). This is also the precondition for theming: a design only follows a light/dark switch if its shapes are bound to token names (bindings re-resolve by name when sets/themes change). Apply in chunks and verify —

applyToken

is async and mass-application in one call races/scrambles bindings; Text/Rectangle fill bindings are flaky and need verify+retry (see

shared/plugin-api-gotchas.md

). Value-only matches are a heuristic: flag inverse-colored chrome for a semantic-role review. ✋ Checkpoint: review bindings + coverage.

Phase 5 — Validate.

scripts/validateTokens.js

— orphans, unresolved references, off-grid spacing. Produce the structured report.

阶段0 — 发现（只读）。 先执行

high_level_overview

，然后运行

penpotUtils.tokenOverview()

和

shapeStructure(penpot.currentPage.root, 3)

读取结构。在

RUN_ID

台账中记录已有的内容。 ✋ 检查点：确认范围（新建系统 / 扩展现有系统 / 从设计推断）。

阶段1 — 基础token。 创建

primitives

集合并激活它（调用

set.toggleActive()

—— 新集合默认未激活），然后添加原始色阶、间距刻度（4px网格）、圆角刻度和字号刻度（按小三度比例的字号）。使用

scripts/createTokenSet.js

脚本。 ✋ 检查点：审核基础刻度（

tokenOverview()

+ 命名）。

阶段2 — 语义token。 添加引用基础token的意图别名，分为两个集合（被引用的

primitives

集合必须处于激活状态，否则引用会验证失败）：

颜色语义（

color.action.primary.bg

→

{color.blue.500}

）→ 存入 modes/light
；

模式无关语义（
```
spacing.*
```
、
```
radius.*
```
、
```
font.*
```
）→ 存入 semantic
。

scripts/createSemanticTokens.js

脚本会按token类型自动拆分。这是深色模式能作为平行颜色集干净实现的原因。 ✋ 检查点：审核语义映射关系。

阶段3 — 深色模式（可选）。 使用

scripts/createThemes.js

创建平行的 modes/dark
颜色集合（颜色名称与

modes/light

完全一致，值为深色版本）；保持

modes/light

为默认激活状态，将切换权交给用户。需覆盖所有

modes/light

中的颜色名称（脚本会返回

missing

缺失列表）。深色模式可以正常工作——当用户激活

modes/dark

时，绑定会按token名称重新解析——但Agent无法在插件中渲染切换效果（

theme.addSet()

不会持久化；插件的读取会有延迟）。用户需要在Penpot的Token面板中手动开启

modes/dark

/ 关闭

modes/light

（见

references/04-themes-modes.md

）。 ✋ 检查点：验证深色模式的

resolvedValue

，恢复浅色模式，并告知用户切换方法。

阶段4 — 应用/推断/绑定。 可以将token绑定到选中内容（

scripts/applyTokensToShapes.js

），从硬编码设计推断token（

scripts/inferTokensFromShapes.js

），或者——核心功能——通过精确值匹配自动将token绑定到所有未绑定元素，免去用户手动操作（

scripts/bindTokensToElements.js

）。这也是主题化的前提：只有当设计中的形状都绑定到token名称时，设计才能跟随浅色/深色切换变化（集合/主题变化时，绑定会按名称重新解析）。请分批应用并验证——

applyToken

是异步操作，单次调用中批量应用会引发竞态/绑定混乱；文本/矩形的填充绑定不稳定，需要验证+重试（见

shared/plugin-api-gotchas.md

）。仅按值匹配是启发式方法：需标记反色的界面控件，供语义角色审核。 ✋ 检查点：审核绑定情况与覆盖率。

阶段5 — 验证。 运行

scripts/validateTokens.js

—— 检查孤立token、未解析引用、偏离网格的间距。生成结构化报告。

7. Critical Rules

7. 关键规则

Primitives → semantic → component. Shapes bind to semantic tokens, never primitives.
All spacing/padding/gap/radius values are multiples of 4px.
Never create a token you can't justify; propose new tokens at a checkpoint, don't auto-create them.
Idempotent: check for an existing set/token by name before adding (no duplicates).
Type sizes follow a Minor Third (1.2) scale unless the brand dictates otherwise.
Do not apply and read back a token in the same
```
execute_code
```
call (async).
Keep the
```
RUN_ID
```
ledger updated so a truncated session can resume.

基础层 → 语义层 → 组件层。形状绑定到语义token，绝不直接绑定基础token。
所有间距/内边距/间隙/圆角值都是4px的倍数。
绝不创建无法说明用途的token；在检查点提出新token建议，不要自动创建。
幂等性：添加前先检查是否已存在同名集合/token（避免重复）。
除非品牌有特殊要求，字号遵循小三度（1.2）比例。
不要在同一次
```
execute_code
```
调用中既应用token又回读结果（异步特性）。
保持
```
RUN_ID
```
台账更新，以便会话中断后可以恢复。

8. Domain Architecture

8. 领域架构

Three token tiers (see

shared/tokens-schema.json

primitives — raw, context-free scale (
```
color.blue.500
```
,
```
spacing.4
```
,
```
radius.md
```
,
```
font.size.300
```
). Never bound directly to shapes when a semantic token exists.
semantic — intent aliases referencing primitives (
```
color.text.default
```
→
```
{color.gray.900}
```
,
```
spacing.inset.md
```
→
```
{spacing.4}
```
). This is what shapes bind to.
components — optional per-component overrides referencing semantic tokens. Create only when a component needs its own surface.

Set layout (see

shared/modes-and-policies.md

): mode-invariant semantics live in

semantic

; colour semantics live in

modes/light

modes/dark

(same names in both). Themes are modeled as Penpot token themes (

addTheme({ group, name })

) that toggle the matching

modes/*

set; primitives and the mode-invariant

semantic

set stay constant.

三层token架构（见

shared/tokens-schema.json

）：

基础层（primitives） — 原始的、无上下文的刻度（
```
color.blue.500
```
、
```
spacing.4
```
、
```
radius.md
```
、
```
font.size.300
```
）。当存在语义token时，绝不要直接绑定到形状。
语义层（semantic） — 引用基础token的意图别名（
```
color.text.default
```
→
```
{color.gray.900}
```
、
```
spacing.inset.md
```
→
```
{spacing.4}
```
）。形状绑定的就是这一层。
组件层（components） — 可选的、引用语义token的单组件覆盖值。仅当组件需要专属表层时创建。

集合布局（见

shared/modes-and-policies.md

）：模式无关的语义token存放在

semantic

集合中；颜色语义token存放在

modes/light

modes/dark

集合中（两者名称完全一致）。主题以Penpot token主题的形式实现（

addTheme({ group, name })

），用于切换对应的

modes/*

集合；基础集合和模式无关的

semantic

集合保持不变。

9. Modes & Policies

9. 模式与策略

Default review. Creating a new token, building a theme, or applying tokens to shapes all require a checkpoint. The only auto-applicable action (safe set,

shared/modes-and-policies.md

) is replacing a raw value that is exactly equal to an existing token's resolved value with that token.

默认采用审核模式。创建新token、构建主题或将token应用到形状都需要经过检查点。唯一可自动执行的操作（安全操作，见

shared/modes-and-policies.md

）是：将与现有token解析后值完全相等的原始值替换为该token。

10. State Management

10. 状态管理

Ledger keys under

RUN_ID

(see

shared/state-management.md

phase

created:[{kind:'set'|'token'|'theme', name, id}]

proposed:[...]

applied:[...]

. Re-derive reality with

tokenOverview()

after truncation before continuing.

RUN_ID

下的台账键（见

shared/state-management.md

）：

phase

、

created:[{kind:'set'|'token'|'theme', name, id}]

、

proposed:[...]

、

applied:[...]

。会话中断后继续操作前，请用

tokenOverview()

重新获取实际状态。

11. User Checkpoints

11. 用户检查点

After phase	Artifacts shown	What we ask
0 Discovery	existing tokens summary	New / extend / infer?
1 Primitives	`tokenOverview()` of primitives	Approve scales & naming?
2 Semantic	semantic→primitive mapping	Approve intent mapping?
3 Themes	resolved values per theme	Approve theme switch?
4 Apply/Infer	before/after on a sample	Approve bindings?

阶段完成后	展示产物	我们确认的内容
0 发现阶段	现有token汇总	新建 / 扩展 / 推断？
1 基础token阶段	基础token的 `tokenOverview()` 结果	是否认可刻度与命名？
2 语义token阶段	语义→基础映射关系	是否认可意图映射？
3 主题阶段	各主题的解析后值	是否认可主题切换？
4 应用/推断阶段	样本的前后对比	是否认可绑定结果？

12. Naming Conventions

12. 命名规范

Dot-notation per

shared/naming-conventions.md

: sets are

primitives

semantic

(mode-invariant) /

modes/light

modes/dark

(colour) / optional

components

; tokens are lowercase dot paths; themes grouped

mode

with names

Light

Dark

High Contrast

遵循

shared/naming-conventions.md

的点分命名法：集合名为

primitives

semantic

（模式无关） /

modes/light

modes/dark

（颜色） / 可选的

components

；token为小写点路径；主题分组为

mode

，名称为

Light

Dark

High Contrast

。

13. Anti-Rationalization Table

13. 反合理化对照表

Excuse the LLM makes	Why it's wrong	Deterministic countermeasure
"I'll hardcode this hex now and tokenize later."	Hardcoded values break theming and governance; "later" never comes.	Stop. Read `tokenOverview()` , bind to a semantic token, or propose one for review.
"I'll bind shapes straight to a primitive to save a step."	Primitives can't switch per theme; dark mode breaks.	Bind to a semantic token that references the primitive.
"This spacing is 18px, close enough."	Off-grid values erode rhythm and fail governance.	Round to the nearest 4px-grid token (16 or 20). Document any exception only with human approval.
"I'll create a dozen tokens up front to be safe."	Unused/orphan tokens are debt.	Create only what's needed now; propose the rest at a checkpoint.

LLM常找的借口	错误原因	确定性应对措施
"我先硬编码这个色值，之后再token化。"	硬编码值会破坏主题化和治理；"之后"永远不会来。	停止操作。读取 `tokenOverview()` ，绑定到语义token，或提出新token供审核。
"我直接把形状绑定到基础token，省一步。"	基础token不能按主题切换；深色模式会失效。	绑定到引用该基础token的语义token。
"这个间距是18px，差不多就行。"	偏离网格的值会破坏节奏，且不符合治理要求。	四舍五入到最近的4px网格token（16或20）。所有例外都需经人工批准并记录。
"我先创建十几个token备用，更稳妥。"	未使用的/孤立的token是技术债务。	只创建当前需要的内容；其余的在检查点提出建议。

14. Helper Code Snippets

14. 辅助代码片段

// Idempotent set + token creation (Phase 1)
const tok = penpot.library.local.tokens;
let prim = tok.sets.find(s => s.name === "primitives") || tok.addSet({ name: "primitives" });
if (!prim.active) prim.toggleActive();          // new sets are INACTIVE by default
if (!prim.tokens.find(t => t.name === "color.blue.500")) {
  prim.addToken({ type: "color", name: "color.blue.500", value: "#0066FF" });
}
return { set: prim.name, count: prim.tokens.length };

// Semantic alias referencing a primitive (Phase 2)
const tok = penpot.library.local.tokens;
let sem = tok.sets.find(s => s.name === "semantic") || tok.addSet({ name: "semantic" });
if (!sem.active) sem.toggleActive();            // and the referenced 'primitives' set must be active too
if (!sem.tokens.find(t => t.name === "color.action.primary.bg")) {
  sem.addToken({ type: "color", name: "color.action.primary.bg", value: "{color.blue.500}" });
}
return { ok: true };

// Idempotent set + token creation (Phase 1)
const tok = penpot.library.local.tokens;
let prim = tok.sets.find(s => s.name === "primitives") || tok.addSet({ name: "primitives" });
if (!prim.active) prim.toggleActive();          // new sets are INACTIVE by default
if (!prim.tokens.find(t => t.name === "color.blue.500")) {
  prim.addToken({ type: "color", name: "color.blue.500", value: "#0066FF" });
}
return { set: prim.name, count: prim.tokens.length };

// Semantic alias referencing a primitive (Phase 2)
const tok = penpot.library.local.tokens;
let sem = tok.sets.find(s => s.name === "semantic") || tok.addSet({ name: "semantic" });
if (!sem.active) sem.toggleActive();            // and the referenced 'primitives' set must be active too
if (!sem.tokens.find(t => t.name === "color.action.primary.bg")) {
  sem.addToken({ type: "color", name: "color.action.primary.bg", value: "{color.blue.500}" });
}
return { ok: true };

15. Reference Resources

15. 参考资源

penpot_api_info('TokenCatalog')

penpot_api_info('TokenSet')

penpot_api_info('Token')

shared/tokens-schema.json

(taxonomy + real type strings),

shared/plugin-api-gotchas.md

penpot_api_info('TokenCatalog')

、

penpot_api_info('TokenSet')

、

penpot_api_info('Token')

。

shared/tokens-schema.json

（分类体系 + 实际类型字符串）、

shared/plugin-api-gotchas.md

。

16. Supporting Files

16. 配套文件

references/

File	Loaded for
`01-discovery.md`	Phase 0 read-only inventory
`02-token-architecture.md`	tiers, scales, type ratio
`03-infer-from-design.md`	building tokens from a hardcoded design
`04-themes-modes.md`	light/dark theme modeling
`05-naming.md`	token naming specifics
`06-anti-rationalization.md`	governance excuses + countermeasures
`07-error-recovery.md`	common failures & fixes

scripts/

File	Purpose
`inspectFileStructure.js`	read pages/library/tokens
`createTokenSet.js`	idempotent primitives set + tokens
`createSemanticTokens.js`	semantic aliases referencing primitives
`inferTokensFromShapes.js`	propose tokens from raw values
`applyTokensToShapes.js`	bind tokens to a selection
`bindTokensToElements.js`	auto-bind tokens to unbound elements by value (chunked + verify) — enables theming
`createThemes.js`	light/dark themes
`validateTokens.js`	orphans, unresolved refs, off-grid

Doctrine paths.

shared/…

and

policies/…

resolve inside this bundle in native installs (vendored by the installer); in a Claude Code plugin install they live at the plugin root —

${CLAUDE_PLUGIN_ROOT}/shared/…

, two directories up from this file.

references/

文件	加载用途
`01-discovery.md`	阶段0 只读清单
`02-token-architecture.md`	层级、刻度、字号比例
`03-infer-from-design.md`	从硬编码设计构建token
`04-themes-modes.md`	浅色/深色主题建模
`05-naming.md`	token命名细节
`06-anti-rationalization.md`	治理借口与应对措施
`07-error-recovery.md`	常见故障与修复方法

scripts/

文件	用途
`inspectFileStructure.js`	读取页面/库/token
`createTokenSet.js`	幂等创建基础token集合与token
`createSemanticTokens.js`	创建引用基础token的语义别名
`inferTokensFromShapes.js`	从原始值提出token建议
`applyTokensToShapes.js`	将token绑定到选中内容
`bindTokensToElements.js`	按值自动绑定token到未绑定元素（分批+验证）—— 实现主题化
`createThemes.js`	浅色/深色主题
`validateTokens.js`	检查孤立token、未解析引用、偏离网格

规则文件路径：原生安装时，

shared/…

和

policies/…

位于本包内部（由安装程序内置）；在 Claude Code 插件安装场景下，它们位于插件根目录——

${CLAUDE_PLUGIN_ROOT}/shared/…

，即本文件的上两级目录。