Back to Details

explore

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Project Explorer

项目探索工具

Deploy parallel discovery agents to catalog a project's capabilities and identify strategic improvements and expansion opportunities. Produces a structured report of existing features grouped by domain, plus prioritized recommendations.
Write
and 
Edit
are intentionally excluded from allowed-tools. This skill MUST NOT modify any files.
部署并行发现Agent来盘点项目能力，识别战略改进和扩展机会。生成按领域分组的现有功能结构化报告，以及优先级排序的改进建议。
Write
和
Edit
工具被有意排除在允许使用的工具之外。本Skill绝对不能修改任何文件。

Quick Start

快速开始

Follow all 4 steps sequentially.
按顺序完成以下4个步骤。

Workflow

工作流程

Step 1: Orient

步骤1：定位

Gather project context to feed into agent prompts. Do NOT present this to the user yet.
  1. Read these files (skip any that don't exist):
    • README.md
      or 
      README.*
    • package.json
      , 
      Cargo.toml
      , 
      go.mod
      , 
      pyproject.toml
      , 
      Gemfile
      , 
      pom.xml
      , 
      build.gradle
    • docker-compose.yml
      , 
      Dockerfile
    • CLAUDE.md
      , 
      .claude/settings.json
    • Top-level directory listing
  2. Extract orientation context:
    • Project name — from package manifest or README title
    • Purpose — one-sentence description from README or manifest
    • Stack — languages, frameworks, key dependencies
    • Scale — approximate file count and directory depth
    • Architecture type — monolith, monorepo, microservices, library, CLI tool, etc.
  3. Count source files (exclude 
    node_modules
    , 
    vendor
    , 
    target
    , 
    dist
    , 
    .git
    , 
    __pycache__
    , 
    build
    ):
    bash
    find . -type f \( -name '*.ts' -o -name '*.tsx' -o -name '*.js' -o -name '*.jsx' -o -name '*.py' -o -name '*.rb' -o -name '*.go' -o -name '*.rs' -o -name '*.java' -o -name '*.kt' -o -name '*.swift' -o -name '*.c' -o -name '*.cpp' -o -name '*.h' -o -name '*.cs' -o -name '*.ex' -o -name '*.exs' -o -name '*.php' -o -name '*.vue' -o -name '*.svelte' \) -not -path '*/node_modules/*' -not -path '*/vendor/*' -not -path '*/target/*' -not -path '*/dist/*' -not -path '*/.git/*' -not -path '*/__pycache__/*' -not -path '*/build/*' | wc -l
  4. Small-project escape hatch: If there are fewer than 10 source files, skip Step 2 (parallel agents) entirely. Instead, read all source files directly, then jump to Step 3 with the information gathered from your own reading. This avoids the overhead of parallel agents for trivial projects.
  5. Parse focus area (if provided):
    • /explore focus: API layer
      -> set 
      FOCUS = "API layer"
      and include it in all agent prompts
    • /explore focus: auth
      -> set 
      FOCUS = "auth"
      and include it in all agent prompts
    • /explore
      (bare) -> 
      FOCUS = null
      , full-project exploration
收集项目上下文以输入到Agent的提示词中。目前不要将这些内容展示给用户。
  1. 读取以下文件（跳过不存在的文件）：
    • README.md
      或 
      README.*
    • package.json
      、
      Cargo.toml
      、
      go.mod
      、
      pyproject.toml
      、
      Gemfile
      、
      pom.xml
      、
      build.gradle
    • docker-compose.yml
      、
      Dockerfile
    • CLAUDE.md
      、
      .claude/settings.json
    • 顶层目录列表
  2. 提取定位上下文：
    • 项目名称 — 来自包清单或README标题
    • 项目目标 — 来自README或包清单的一句话描述
    • 技术栈 — 编程语言、框架、关键依赖
    • 规模 — 大致的文件数量和目录深度
    • 架构类型 — 单体应用、单体仓库、微服务、类库、CLI工具等
  3. 统计源代码文件数量（排除
    node_modules
    、
    vendor
    、
    target
    、
    dist
    、
    .git
    、
    __pycache__
    、
    build
    目录）：
    bash
    find . -type f \( -name '*.ts' -o -name '*.tsx' -o -name '*.js' -o -name '*.jsx' -o -name '*.py' -o -name '*.rb' -o -name '*.go' -o -name '*.rs' -o -name '*.java' -o -name '*.kt' -o -name '*.swift' -o -name '*.c' -o -name '*.cpp' -o -name '*.h' -o -name '*.cs' -o -name '*.ex' -o -name '*.exs' -o -name '*.php' -o -name '*.vue' -o -name '*.svelte' \) -not -path '*/node_modules/*' -not -path '*/vendor/*' -not -path '*/target/*' -not -path '*/dist/*' -not -path '*/.git/*' -not -path '*/__pycache__/*' -not -path '*/build/*' | wc -l
  4. 小型项目快捷通道：如果源代码文件少于10个，完全跳过步骤2（并行Agent）。改为直接读取所有源代码文件，然后使用自行收集到的信息跳至步骤3。这样可以避免在小型项目中使用并行Agent带来的额外开销。
  5. 解析聚焦领域（如果提供）：
    • /explore focus: API layer
      -> 设置
      FOCUS = "API layer"
      并将其包含在所有Agent的提示词中
    • /explore focus: auth
      -> 设置
      FOCUS = "auth"
      并将其包含在所有Agent的提示词中
    • /explore
      （无参数）-> 
      FOCUS = null
      ，探索整个项目

Step 2: Deploy 3 Discovery Agents in Parallel

步骤2：并行部署3个发现Agent

Launch all 3 agents simultaneously using the 
Task
tool. Each agent receives the orientation context from Step 1.
使用
Task
工具同时启动所有3个Agent。每个Agent都会接收步骤1中收集的定位上下文。

Agent Roster

Agent清单

Architecture Cartographer (
Plan
)
  • Mission: Map the project's structural organization — modules, components, data models, integrations, and boundaries.
  • Exploration targets: Directory structure and module hierarchy, data models and schemas, external integrations (APIs, databases, message queues, third-party services), internal boundaries (what talks to what), configuration and environment management.
  • Strategy: Start with the directory tree and entry points. Follow imports to map the dependency graph. Identify clear module boundaries and where they blur. Note shared types and cross-cutting concerns.
  • Output format: Group findings by architectural domain. For each domain, list components with file paths and a one-line description of responsibility.
Capability Analyst (
Explore
)
  • Mission: Inventory all user-facing and developer-facing capabilities — routes, commands, APIs, background jobs, UI screens, event handlers.
  • Exploration targets: HTTP routes and API endpoints, CLI commands and subcommands, background jobs and scheduled tasks, event handlers and webhooks, UI pages and components (if applicable), exported functions and public APIs (for libraries).
  • Strategy: Search for route definitions, command registrations, job schedulers, event listeners, and exported interfaces. Read handler implementations to understand what each capability does. For web apps, trace from routes to controllers to services.
  • Output format: List each capability with: name, type (route/command/job/etc.), file path, and a one-sentence description of what it does.
Infrastructure Scout (
general-purpose
)
  • Mission: Assess the project's operational maturity — CI/CD, testing, deployment, observability, and developer experience tooling.
  • Exploration targets: CI/CD configuration (GitHub Actions, GitLab CI, Jenkinsfile, etc.), test setup and coverage patterns, deployment configuration (Docker, K8s, serverless, PaaS), observability (logging, metrics, tracing, error tracking), developer experience (linting, formatting, pre-commit hooks, scripts, Makefile), documentation (README quality, API docs, architecture docs).
  • Strategy: Check for CI config files, test directories, Dockerfiles, deployment manifests, monitoring config. Read package scripts and Makefile targets. Assess documentation completeness.
  • Output format: For each infrastructure area, report: what exists (with file paths), what's configured, and notable gaps.
架构制图师 (
Plan
)
  • 任务：绘制项目的结构组织图——模块、组件、数据模型、集成和边界。
  • 探索目标：目录结构和模块层级、数据模型和 schema、外部集成（API、数据库、消息队列、第三方服务）、内部边界（组件间的调用关系）、配置和环境管理。
  • 策略：从目录树和入口文件开始。通过导入关系映射依赖图。识别清晰的模块边界以及边界模糊的地方。记录共享类型和横切关注点。
  • 输出格式：按架构领域分组展示发现结果。对于每个领域，列出组件及其文件路径和一句职责描述。
能力分析师 (
Explore
)
  • 任务：盘点所有面向用户和面向开发者的能力——路由、命令、API、后台任务、UI界面、事件处理器。
  • 探索目标：HTTP路由和API端点、CLI命令和子命令、后台任务和定时任务、事件处理器和Webhook、UI页面和组件（如果适用）、导出函数和公共API（针对类库）。
  • 策略：搜索路由定义、命令注册、任务调度器、事件监听器和导出接口。读取处理器实现以了解每个能力的功能。对于Web应用，从路由追踪到控制器再到服务层。
  • 输出格式：列出每个能力，包含：名称、类型（路由/命令/任务等）、文件路径、以及一句功能描述。
基础设施侦察员 (
general-purpose
)
  • 任务：评估项目的运维成熟度——CI/CD、测试、部署、可观测性和开发者体验工具。
  • 探索目标：CI/CD配置（GitHub Actions、GitLab CI、Jenkinsfile等）、测试设置和覆盖模式、部署配置（Docker、K8s、无服务器、PaaS）、可观测性（日志、指标、追踪、错误追踪）、开发者体验（代码检查、格式化、提交前钩子、脚本、Makefile）、文档（README质量、API文档、架构文档）。
  • 策略：检查CI配置文件、测试目录、Dockerfile、部署清单、监控配置。读取包脚本和Makefile目标。评估文档的完整性。
  • 输出格式：对于每个基础设施领域，报告：已存在的内容（带文件路径）、配置情况、以及显著的缺口。

Agent Prompt Template

Agent提示词模板

For each agent, construct the 
Task
prompt as:
You are the {Agent Name} exploring a project at {project_root}.
为每个Agent构建
Task
提示词如下：
You are the {Agent Name} exploring a project at {project_root}.

Orientation

Orientation

  • Project: {name} — {purpose}
  • Stack: {stack}
  • Scale: ~{file_count} source files
  • Architecture: {architecture_type} {if FOCUS: - Focus area: Narrow your exploration to aspects related to "{FOCUS}", but still cover your full mission within that lens.}
  • Project: {name} — {purpose}
  • Stack: {stack}
  • Scale: ~{file_count} source files
  • Architecture: {architecture_type} {if FOCUS: - Focus area: Narrow your exploration to aspects related to "{FOCUS}", but still cover your full mission within that lens.}

Your Mission

Your Mission

{mission from the agent roster above}
{mission from the agent roster above}

Exploration Targets

Exploration Targets

{exploration targets from the agent roster above}
{exploration targets from the agent roster above}

Strategy

Strategy

{strategy from the agent roster above}
{strategy from the agent roster above}

Rules

Rules

  • This is a READ-ONLY exploration. Do not modify any files.
  • Include specific file paths for every finding.
  • Group findings by domain/area.
  • Be comprehensive but concise — list capabilities, don't explain obvious ones in detail.
  • If a target area has nothing to report, say so briefly rather than omitting it.
  • This is a READ-ONLY exploration. Do not modify any files.
  • Include specific file paths for every finding.
  • Group findings by domain/area.
  • Be comprehensive but concise — list capabilities, don't explain obvious ones in detail.
  • If a target area has nothing to report, say so briefly rather than omitting it.

Output Format

Output Format

{output format from the agent roster above}
undefined
{output format from the agent roster above}
undefined

Step 3: Synthesize

步骤3：整合分析结果

Once all agents return (or after your own reading for small projects):
  1. Merge and deduplicate agent reports into a unified domain-grouped view. Domains should emerge naturally from the project (e.g., "Authentication", "API", "Data Layer", "Background Processing") rather than mapping 1:1 to agents.
  2. Generate recommendations in three categories:
    Improving existing capabilities — things that already exist but could be better:
    • Incomplete features (partial implementations, TODO markers, stub endpoints)
    • Missing infrastructure (no tests for critical paths, no CI, no error tracking)
    • Robustness gaps (missing error handling at boundaries, no retry logic, no rate limiting)
    • Tag each: High / Medium / Exploratory
    Novel capability opportunities — things that don't exist but naturally extend what does:
    • Natural feature extensions based on existing data models and integrations
    • New integrations that complement the current stack
    • User-facing features enabled by existing infrastructure
    • Tag each: High / Medium / Exploratory
    Other suggestions — DX, ecosystem, operational maturity:
    • Developer experience improvements (better scripts, documentation, onboarding)
    • Operational maturity (monitoring, alerting, deployment automation)
    • Ecosystem plays (plugins, extensions, API consumers)
    • Tag each: High / Medium / Exploratory
  3. Impact tagging criteria:
    • High — Addresses a clear gap that affects users or reliability today
    • Medium — Meaningful improvement but not urgent; good next-quarter work
    • Exploratory — Interesting direction worth investigating; may or may not pan out
当所有Agent返回结果后（或对于小型项目，在自行读取文件后）：
  1. 合并并去重Agent的报告，形成统一的按领域分组的视图。领域应从项目中自然提炼（例如“认证”、“API”、“数据层”、“后台处理”），而非与Agent一一对应。
  2. 生成三类改进建议：
    现有能力改进 — 已存在但可以优化的内容：
    • 未完成的功能（部分实现、TODO标记、占位端点）
    • 缺失的基础设施（关键路径无测试、无CI、无错误追踪）
    • 健壮性缺口（边界处缺少错误处理、无重试逻辑、无速率限制）
    • 每个建议标记：高 / 中 / 探索性
    新能力机会 — 目前不存在但可自然扩展现有功能的内容：
    • 基于现有数据模型和集成的自然功能扩展
    • 与当前技术栈互补的新集成
    • 现有基础设施支持的面向用户的功能
    • 每个建议标记：高 / 中 / 探索性
    其他建议 — 开发者体验、生态系统、运维成熟度：
    • 开发者体验改进（更完善的脚本、文档、入职流程）
    • 运维成熟度提升（监控、告警、部署自动化）
    • 生态系统拓展（插件、扩展、API消费者）
    • 每个建议标记：高 / 中 / 探索性
  3. 影响标记标准：
    • 高 — 解决当前影响用户或可靠性的明确缺口
    • 中 — 有意义的改进但不紧急；适合作为下一季度的工作
    • 探索性 — 值得研究的有趣方向；可能成功也可能失败

Step 4: Present Report

步骤4：呈现报告

Output the final report in this format:
markdown
undefined
按以下格式输出最终报告：
markdown
undefined

Explore: {project name}

Explore: {project name}

Purpose: {one-sentence description} Stack: {languages, frameworks, key deps} Architecture: {architecture type} Scale: ~{file_count} source files across {directory_count} directories {if FOCUS: Focus: {FOCUS}}
Purpose: {one-sentence description} Stack: {languages, frameworks, key deps} Architecture: {architecture type} Scale: ~{file_count} source files across {directory_count} directories {if FOCUS: Focus: {FOCUS}}

Capabilities

Capabilities

{For each domain, a table or grouped list:}
{For each domain, a table or grouped list:}

{Domain Name}

{Domain Name}

CapabilityTypeLocationDescription
{name}{route/command/job/etc.}
{file:line}
{one-sentence description}
{Repeat for each domain}
CapabilityTypeLocationDescription
{name}{route/command/job/etc.}
{file:line}
{one-sentence description}
{Repeat for each domain}

Strategic Improvements

Strategic Improvements

Improvements to existing capabilities:
#RecommendationImpactDomainDetails
1{title}High{domain}{one-sentence explanation}
2{title}Medium{domain}{one-sentence explanation}
Improvements to existing capabilities:
#RecommendationImpactDomainDetails
1{title}High{domain}{one-sentence explanation}
2{title}Medium{domain}{one-sentence explanation}

Novel Opportunities

Novel Opportunities

New capabilities this project could grow into:
#OpportunityImpactBuilds OnDetails
1{title}High{existing capability/domain}{one-sentence explanation}
2{title}Exploratory{existing capability/domain}{one-sentence explanation}
New capabilities this project could grow into:
#OpportunityImpactBuilds OnDetails
1{title}High{existing capability/domain}{one-sentence explanation}
2{title}Exploratory{existing capability/domain}{one-sentence explanation}

Other Suggestions

Other Suggestions

DX, operational, and ecosystem improvements:
#SuggestionImpactAreaDetails
1{title}Medium{area}{one-sentence explanation}
DX, operational, and ecosystem improvements:
#SuggestionImpactAreaDetails
1{title}Medium{area}{one-sentence explanation}

What's Next?

What's Next?

Pick an option or tell me what you'd like to do:
  1. Deep-dive into a specific domain (give me the name)
  2. Elaborate on a recommendation (give me the number and table)
  3. Compare two directions — I'll analyze trade-offs
  4. Export this report to a markdown file
undefined
Pick an option or tell me what you'd like to do:
  1. Deep-dive into a specific domain (give me the name)
  2. Elaborate on a recommendation (give me the number and table)
  3. Compare two directions — I'll analyze trade-offs
  4. Export this report to a markdown file
undefined

Guidelines

指南

  • This skill is strictly read-only. Never modify any files.
  • Deploy all 3 agents in parallel, never sequentially (unless small-project path).
  • Recommendations should be strategic, not code-level. "Add rate limiting to the API gateway" not "fix the off-by-one on line 42."
  • Include file paths in the capabilities table so the user can navigate directly.
  • Keep recommendations to 3-5 per category. Quality over quantity.
  • The follow-up menu is important — always include it.
  • If the user asks to deep-dive into a domain, spawn a targeted agent to explore that domain in depth.
  • If the user asks to elaborate on a recommendation, provide a concrete implementation sketch (still read-only — describe the approach, don't write the code).
  • If the user asks to compare directions, present a structured trade-off analysis with pros, cons, effort, and risk for each option.
  • 本Skill严格为只读模式。绝不修改任何文件。
  • 并行部署所有3个Agent，绝不串行部署（除非使用小型项目快捷通道）。
  • 建议应具有战略性，而非代码层面的细节。例如“为API网关添加速率限制”而非“修复第42行的off-by-one错误”。
  • 在能力表中包含文件路径，以便用户可以直接导航到对应位置。
  • 每个类别的建议保持3-5个。质量优先于数量。
  • 后续菜单很重要 — 务必包含。
  • 如果用户要求深入探索某个领域，启动一个针对性的Agent来深度探索该领域。
  • 如果用户要求详细说明某个建议，提供具体的实现方案草图（仍为只读模式 — 描述方法，不要编写代码）。
  • 如果用户要求比较不同方向，提供结构化的权衡分析，包含每个选项的优缺点、工作量和风险。