navigate

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Navigate

Meta — Utility. Artifact awareness and multi-phase orchestration.

Core Question: "What exists, and how do I track a complex workflow across sessions?"

Note: Skill routing (matching a user's goal to the right skills) is the agent's job — it reads skill descriptions and proposes matches proactively on every response. Navigate does NOT route. It provides artifact state awareness and persistent workflow tracking that the agent can't do implicitly.

Meta — 实用工具。工件感知与多阶段编排。

核心问题： "现有哪些工件？如何跨会话跟踪复杂工作流？"

注意： 技能路由（将用户目标匹配到对应技能）是Agent的职责——它会读取技能描述并在每次响应中主动推荐匹配的技能。Navigate不负责路由，它提供Agent无法隐式实现的工件状态感知和持久化工作流跟踪能力。

Two Modes

两种模式

Status

状态模式

When: Argument is

status

or user asks "what exists", "what's stale", "what do I have"

Scan

.agents/

and report. Single-pass, no sub-agents.

Scan
```
.agents/
```
for every
```
.md
```
file, including
```
.agents/meta/
```
. For each, read frontmatter to extract
```
skill:
```
,
```
date:
```
,
```
status:
```
,
```
version:
```
. Missing fields =
```
—
```
. Most meta-skill artifacts (
```
.agents/meta/
```
) are ephemeral analysis outputs — report them but weigh them lightly in recommendations. Exception:
```
meta/review-chain-report.md
```
is consumed by
```
/ship
```
as a review gate, so treat it as a real dependency when ship is in the workflow. Also scan
```
.agents/meta/out-of-scope/
```
— report count of out-of-scope decisions and flag any whose "Revisit if" conditions may now be met.
Report as a table sorted by date (newest first). Mark STALE if
```
date:
```
> 30 days old.

| Artifact | Skill | Date | Age | Status |
|----------|-------|------|-----|--------|
| product-context.md | icp-research | 2026-03-15 | 13d | ok |
| solution-design.md | solution-design | 2026-02-10 | 46d | STALE |

.agents/

doesn't exist or is empty, say so.

Recommend the one or two skills that unblock the most, using the dependency graph below. Don't dump a flat list of everything missing — trace the graph, find the root blocker.

触发场景： 参数为

status

或用户询问“现有哪些内容”“哪些已过期”“我有什么工件”

扫描

.agents/

目录并生成报告。单次扫描，不调用子Agent。

扫描
```
.agents/
```
目录下所有
```
.md
```
文件，包括
```
.agents/meta/
```
。针对每个文件，读取前置元数据以提取
```
skill:
```
、
```
date:
```
、
```
status:
```
、
```
version:
```
字段。缺失字段显示为
```
—
```
。大多数元技能工件（
```
.agents/meta/
```
下的文件）是临时分析输出——报告中会列出，但在推荐时权重较低。例外情况：
```
meta/review-chain-report.md
```
是
```
/ship
```
技能的审核 gate，因此当工作流中包含
```
ship
```
时，需将其视为实际依赖项。同时扫描
```
.agents/meta/out-of-scope/
```
目录——报告已标记为超出范围的决策数量，并标记任何“需重新审视”条件已满足的项。
生成报告：以表格形式按日期排序（最新的在前）。若
```
date:
```
字段距今超过30天，标记为STALE（过期）。

| Artifact | Skill | Date | Age | Status |
|----------|-------|------|-----|--------|
| product-context.md | icp-research | 2026-03-15 | 13d | ok |
| solution-design.md | solution-design | 2026-02-10 | 46d | STALE |

若

.agents/

目录不存在或为空，需告知用户。

推荐：根据下方的依赖关系图，推荐1-2个最能解除阻塞的技能。不要直接列出所有缺失的内容——需追溯依赖图，找到根阻塞点。

Orchestrate

编排模式

When: Argument starts with

orchestrate

or user asks for a full workflow plan

Scan
```
.agents/
```
for existing artifacts (same as Status step 1)
Check
```
.agents/meta/out-of-scope/
```
— if the goal overlaps with a prior out-of-scope decision, surface it: "This was previously scoped out because [reason]. Revisit condition: [condition]. Proceed anyway?"
Classify the goal, match to skills using the skill registry, check dependencies, identify parallel tracks
Produce a
```
workflow-plan.md
```
artifact with phases, checkpoints, and progress tracking
Return the plan + instructions for Phase 1

Continuation: Run

/navigate orchestrate

again (no new goal) to validate current phase, update progress, and recommend next phase.

触发场景： 参数以

orchestrate

开头或用户请求完整工作流计划

扫描
```
.agents/
```
目录下的现有工件（与状态模式的步骤1相同）
检查
```
.agents/meta/out-of-scope/
```
目录——若用户目标与之前标记为超出范围的决策重叠，需提示：“此目标之前因[原因]被标记为超出范围。重新审视条件：[条件]。是否继续？”
对用户目标进行分类，使用技能注册表匹配对应技能，检查依赖关系，识别并行任务流
生成
```
workflow-plan.md
```
工件，包含阶段、检查点和进度跟踪
返回计划以及第一阶段的执行说明

续期操作： 再次调用

/navigate orchestrate

（无需新目标）即可验证当前阶段、更新进度并推荐下一阶段。

Orchestrate Output Format

编排模式输出格式

markdown

---
skill: navigate
version: 1
date: [today]
status: in-progress
goal: "[user's goal]"
---

markdown

---
skill: navigate
version: 1
date: [today]
status: in-progress
goal: "[user's goal]"
---

Workflow: [Goal Title]

Phases

Phase 1: [Name]

/[skill] -> [artifact] Decisions:
Mechanical: [auto-decided, not surfaced]
Taste: [auto-decided, surfaced at checkpoint]
User Challenge: [must ask before proceeding] Checkpoint: [validation criteria]

/[skill] -> [artifact] Decisions:
Mechanical: [auto-decided, not surfaced]
Taste: [auto-decided, surfaced at checkpoint]
User Challenge: [must ask before proceeding] Checkpoint: [validation criteria]

Phase 2: [Name]

/[skill] -> [artifact] Decisions:
[same format] Checkpoint: [validation criteria]

/[skill] -> [artifact] Decisions:
[same format] Checkpoint: [validation criteria]

Status

Current phase: 1 Next action: Run

/[first-skill] [context]

---

Current phase: 1 Next action: Run

/[first-skill] [context]

---

Decision Classification

决策分类

In Orchestrate mode, classify decisions within each phase into three types:

Type	Behavior	Examples
Mechanical	Auto-decide silently — obvious from context, low risk	"Use the ORM already in the codebase", "Follow existing naming conventions"
Taste	Auto-decide but surface at phase checkpoint for user review	"Chose polling over WebSockets because latency requirements are relaxed", "Picked 3 personas instead of 2"
User Challenge	Always ask — never assume	"Should we support mobile?", "Which pricing tier to target?", "Build vs buy for auth?"

How to classify:

If the codebase, spec, or prior artifacts answer it → Mechanical
If you have a defensible recommendation but the user might disagree → Taste
If it's expensive to reverse, subjective, or has business implications → User Challenge

In the workflow plan, annotate each phase:

undefined

在编排模式下，需将每个阶段内的决策分为三类：

类型	行为	示例
机械型	自动静默决策——根据上下文可明确判断，风险低	"使用代码库中已有的ORM"、"遵循现有命名规范"
偏好型	自动决策，但会在阶段检查点向用户展示	"因延迟要求宽松，选择轮询而非WebSocket"、"选择3个人物角色而非2个"
用户决策型	必须询问用户——绝不自行假设	"是否支持移动端？"、"针对哪个定价层级？"、"认证功能自行开发还是采购？"

分类方法：

若代码库、规格说明或现有工件已给出答案 → 机械型
若你有合理的推荐方案，但用户可能有不同意见 → 偏好型
若决策成本高、主观性强或涉及业务影响 → 用户决策型

在工作流计划中的标注示例：

undefined

Phase 2: Architecture

/system-architecture -> system-architecture.md Decisions:
Mechanical: tech stack (matches existing codebase)
Taste: DB schema normalization level — will default to 3NF, surface at checkpoint
User Challenge: self-hosted vs managed DB — must ask before proceeding


**At phase checkpoints**, present Taste decisions as a batch: "I made these calls — confirm or override before I continue." User Challenges block the phase until answered.

/system-architecture -> system-architecture.md Decisions:
Mechanical: tech stack (matches existing codebase)
Taste: DB schema normalization level — will default to 3NF, surface at checkpoint
User Challenge: self-hosted vs managed DB — must ask before proceeding


**在阶段检查点：** 将偏好型决策批量展示给用户：“我已做出以下决策——请确认或修改后再继续。” 用户决策型问题会阻塞阶段推进，直到得到答复。

Checkpoint Validation

检查点验证

Between phases in orchestrate mode:

Existence — required artifacts exist
Freshness —
```
date
```
field < 30 days
Completeness — valid frontmatter present
Status —
```
final
```
preferred;
```
draft
```
triggers warning

Phase [N] Checkpoint: PASS / WARNING / BLOCKED

在编排模式的阶段之间：

存在性 —— 所需工件已存在
新鲜度 ——
```
date
```
字段距今少于30天
完整性 —— 包含有效的前置元数据
状态 —— 优先
```
final
```
状态；
```
draft
```
状态会触发警告

Phase [N] Checkpoint: PASS / WARNING / BLOCKED

Cross-Project Mode (--cross-project)

跨项目模式（--cross-project）

When invoked with

--cross-project

Scan
```
.agents/
```
in current directory AND sibling skill repo directories
Report unified table with Project column
Trace cross-project dependencies
Recommend highest-impact action across ALL projects

当使用

--cross-project

参数调用时：

扫描当前目录及同级技能仓库目录下的
```
.agents/
```
目录
生成包含**Project（项目）**列的统一表格报告
追溯跨项目依赖关系
推荐所有项目中影响最大的操作

Dependency Graph (Canonical)

依赖关系图（标准版）

Artifacts are optional save-points, not mandatory pipeline gates. The graph shows what CAN flow between skills, not what MUST exist on disk. Conversation context from the current session is equally valid.

product-context.md <- /icp-research
|-> market-research.md <- /market-research --+
|-> problem-analysis.md <- /problem-analysis -+
|                                             +-> solution-design.md <- /solution-design
|                                             |   |-> targets.md <- /funnel-planner -> experiment-*.md
|                                             |   |-> mkt/imc-plan.md <- /imc-plan -> mkt/content/ -> mkt/*.humanized.md
|                                             |   +-> system-architecture.md <- /system-architecture
|                                             |       +-> tasks.md <- /task-breakdown
|                                             |           +-> (execute) -> meta/review-chain-report.md <- /review-chain
|                                             |               +-> ship-report.md <- /ship
|                                             |                   +-> deploy-verify-report.md <- /deploy-verify
|-> /discover (conversation context or spec.md) --+
|-> design/brand-system.md <- /brand-system
+-> design/user-flow.md <- /user-flow --> system-architecture.md, tasks.md

工件是可选的保存点，而非强制性的流水线 gate。该图展示了技能之间可以传递的内容，而非磁盘上必须存在的内容。当前会话的对话上下文同样有效。

product-context.md <- /icp-research
|-> market-research.md <- /market-research --+
|-> problem-analysis.md <- /problem-analysis -+
|                                             +-> solution-design.md <- /solution-design
|                                             |   |-> targets.md <- /funnel-planner -> experiment-*.md
|                                             |   |-> mkt/imc-plan.md <- /imc-plan -> mkt/content/ -> mkt/*.humanized.md
|                                             |   +-> system-architecture.md <- /system-architecture
|                                             |       +-> tasks.md <- /task-breakdown
|                                             |           +-> (execute) -> meta/review-chain-report.md <- /review-chain
|                                             |               +-> ship-report.md <- /ship
|                                             |                   +-> deploy-verify-report.md <- /deploy-verify
|-> /discover (conversation context or spec.md) --+
|-> design/brand-system.md <- /brand-system
+-> design/user-flow.md <- /user-flow --> system-architecture.md, tasks.md

Priority Table

优先级表

When recommending next action. Note: if discover ran in the current session, its decisions are in conversation context — no spec.md file is required.

If missing/stale	Run	Because
`product-context.md`	`/icp-research`	12+ downstream skills depend on it
`market-research.md` or `problem-analysis.md`	`/market-research` or `/problem-analysis`	Feed into `/solution-design`
`solution-design.md`	`/solution-design`	Architecture, funnel, comms need it
No clarity on what to build (no conversation or spec)	`/discover`	Need alignment before architecture
`system-architecture.md`	`/system-architecture`	Can't decompose without design
`tasks.md`	`/task-breakdown`	Needs architecture first

推荐下一步操作时参考。注意：若当前会话中已调用过

/discover

，其决策会保存在对话上下文中——无需

spec.md

文件。

若缺失/过期	调用技能	原因
`product-context.md`	`/icp-research`	有12+个下游技能依赖于此工件
`market-research.md` 或 `problem-analysis.md`	`/market-research` 或 `/problem-analysis`	为 `/solution-design` 提供输入
`solution-design.md`	`/solution-design`	架构、漏斗模型、传播策略都依赖于此
对需构建的内容不明确（无对话上下文或 `spec.md` ）	`/discover`	需先达成共识再进行架构设计
`system-architecture.md`	`/system-architecture`	无设计则无法分解任务
`tasks.md`	`/task-breakdown`	需先完成架构设计

Pre-Built Workflow Templates

预构建工作流模板

Technical Build

技术构建

Triggers: "build the app", "implement", "code it"

Phase 1: /discover (interactive conversation — optionally saves spec.md)
Phase 2: /system-architecture -> system-architecture.md
Phase 3: /task-breakdown -> tasks.md
Phase 4: (execution, /review-chain after critical tasks)
Phase 5: /code-cleanup + /technical-writer (parallel)
Phase 6: /ship -> ship-report.md
Phase 7: /deploy-verify -> deploy-verify-report.md

触发词： "构建应用"、"实现"、"编码"

Phase 1: /discover (interactive conversation — optionally saves spec.md)
Phase 2: /system-architecture -> system-architecture.md
Phase 3: /task-breakdown -> tasks.md
Phase 4: (execution, /review-chain after critical tasks)
Phase 5: /code-cleanup + /technical-writer (parallel)
Phase 6: /ship -> ship-report.md
Phase 7: /deploy-verify -> deploy-verify-report.md

Full Product Launch

完整产品发布

Triggers: "launch", "new product", "go to market"

Phase 1: /icp-research -> product-context.md
Phase 2: /market-research + /problem-analysis (parallel)
Phase 3: /solution-design -> solution-design.md
Phase 4: /brand-system + /imc-plan + /funnel-planner (parallel)
Phase 5: /user-flow -> design/user-flow.md
Phase 6: /discover (interactive conversation)
Phase 7: /system-architecture -> system-architecture.md
Phase 8: /task-breakdown -> tasks.md + execution (/review-chain after critical tasks)
Phase 9: /ship -> ship-report.md
Phase 10: /deploy-verify -> deploy-verify-report.md
Phase 11: /content-create + /copywriting -> mkt/content/
Phase 12: /lp-optimization + /seo (parallel)

触发词： "发布"、"新产品"、"推向市场"

Phase 1: /icp-research -> product-context.md
Phase 2: /market-research + /problem-analysis (parallel)
Phase 3: /solution-design -> solution-design.md
Phase 4: /brand-system + /imc-plan + /funnel-planner (parallel)
Phase 5: /user-flow -> design/user-flow.md
Phase 6: /discover (interactive conversation)
Phase 7: /system-architecture -> system-architecture.md
Phase 8: /task-breakdown -> tasks.md + execution (/review-chain after critical tasks)
Phase 9: /ship -> ship-report.md
Phase 10: /deploy-verify -> deploy-verify-report.md
Phase 11: /content-create + /copywriting -> mkt/content/
Phase 12: /lp-optimization + /seo (parallel)

Strategy Sprint

战略冲刺

Triggers: "strategy", "what should we build", "prioritize"

Phase 1: /icp-research -> product-context.md
Phase 2: /market-research + /problem-analysis (parallel)
Phase 3: /solution-design -> solution-design.md
Phase 4: /funnel-planner -> targets.md
Phase 5: /experiment -> experiment-*.md

触发词： "战略"、"我们应该构建什么"、"优先级排序"

Phase 1: /icp-research -> product-context.md
Phase 2: /market-research + /problem-analysis (parallel)
Phase 3: /solution-design -> solution-design.md
Phase 4: /funnel-planner -> targets.md
Phase 5: /experiment -> experiment-*.md

Content Campaign

内容营销活动

Triggers: "content campaign", "marketing campaign"

Phase 1: /icp-research -> product-context.md
Phase 2: /imc-plan -> mkt/imc-plan.md
Phase 3: /content-create -> mkt/content/
Phase 4: /copywriting -> mkt/content/*.copy.md
Phase 5: /humanize -> mkt/content/*.humanized.md

触发词： "内容营销活动"、"营销战役"

Phase 1: /icp-research -> product-context.md
Phase 2: /imc-plan -> mkt/imc-plan.md
Phase 3: /content-create -> mkt/content/
Phase 4: /copywriting -> mkt/content/*.copy.md
Phase 5: /humanize -> mkt/content/*.humanized.md

Landing Page

着陆页

Triggers: "landing page", "conversion page"

Phase 1: /icp-research -> product-context.md
Phase 2: /brand-system -> design/brand-system.md
Phase 3: /copywriting -> mkt/content/*.copy.md
Phase 4: /lp-optimization -> mkt/lp-optimization.md
Phase 5: /humanize -> mkt/content/*.humanized.md

触发词： "着陆页"、"转化页"

Phase 1: /icp-research -> product-context.md
Phase 2: /brand-system -> design/brand-system.md
Phase 3: /copywriting -> mkt/content/*.copy.md
Phase 4: /lp-optimization -> mkt/lp-optimization.md
Phase 5: /humanize -> mkt/content/*.humanized.md

Architecture Decision

架构决策

Triggers: "debate the tech stack", "which approach", "compare options"

Phase 1: /discover -> scope the decision (interactive)
Phase 2: /agent-room debate -> meta/agent-room-report.md
Phase 3: /system-architecture -> system-architecture.md

触发词： "讨论技术栈"、"哪种方案更好"、"对比选项"

Phase 1: /discover -> scope the decision (interactive)
Phase 2: /agent-room debate -> meta/agent-room-report.md
Phase 3: /system-architecture -> system-architecture.md

Anti-Patterns

反模式

Dumping a flat list — Trace the graph, find the root blocker. Don't list every missing artifact equally.
Ignoring artifact state — Don't recommend
```
/icp-research
```
when
```
product-context.md
```
is 3 days old.
Skipping dependency tracing — Don't recommend
```
/task-breakdown
```
when
```
system-architecture.md
```
doesn't exist.
Treating templates as rigid — Skip phases where fresh artifacts exist.
Using navigate for skill routing — The agent proposes skills proactively. Navigate is for artifact status and orchestration only.

直接列出所有缺失项 —— 需追溯依赖图，找到根阻塞点。不要将所有缺失工件同等对待。
忽略工件状态 —— 当
```
product-context.md
```
仅创建3天时，不要推荐
```
/icp-research
```
。
跳过依赖追溯 —— 当
```
system-architecture.md
```
不存在时，不要推荐
```
/task-breakdown
```
。
将模板视为刚性规则 —— 若已有新鲜工件，可跳过对应阶段。
使用Navigate进行技能路由 —— Agent会主动推荐技能。Navigate仅用于工件状态管理和编排。

Chain Position

在技能链中的定位

Meta skill — scans artifact state and recommends the highest-impact next skill to run. Start here when unsure what to do next.

元技能——扫描工件状态并推荐影响最大的下一个技能。当不确定下一步该做什么时，从这里开始。