mirror of https://github.com/automazeio/ccpm.git synced 2025-10-09 13:41:06 +03:00

Files

Ran Aroussi 22b311fe7e Merged main back into rc

commit abf5bed203a9ac13579eac7c9299deee3b9c50c6
Author: Ran Aroussi <ranaroussi@users.noreply.github.com>
Date:   Wed Sep 24 18:46:43 2025 +0100

    Refactor Bash command patterns in settings

commit 1cc603ee2df72eb47b61df04b156dfadb70b5b58
Author: Ran Aroussi <ran@aroussi.com>
Date:   Wed Sep 24 15:23:43 2025 +0100

    Remove outdated Chinese documentation files for agents, commands, and README to streamline project resources.

commit 34d7e724e8159d5ec4aa6365940f355d8ed02872
Author: Ran Aroussi <ranaroussi@users.noreply.github.com>
Date:   Wed Sep 24 15:23:17 2025 +0100

    Revise README with new badge links and documentation

    Updated links and added Chinese documentation reference.

commit 0026086f7a118fd3c46161cde5becf52939ee6be
Author: Ran Aroussi <ranaroussi@users.noreply.github.com>
Date:   Wed Sep 24 15:17:22 2025 +0100

    RC => Main (stability and usability) (#954)

    * feat: Add automatic worktree directory fix via pre-tool-use hook

    Problem: Bash tool resets to main project directory after every command,
      breaking worktree operations for parallel agent development.

      Solution: Pre-tool-use hook that automatically detects worktree context
      and injects 'cd /path/to/worktree && ' prefix to all Bash commands.

      Features:
      - Zero-config: Agents don't need special instructions or awareness
      - Smart detection: Distinguishes worktrees (.git file) from repos (.git dir)
      - Edge case handling: Background processes, pipes, env vars, etc.
      - Debug mode: CLAUDE_HOOK_DEBUG=true for troubleshooting
      - Skip patterns: Avoids double-prefixing and unnecessary injections

      This enables true parallel development with multiple agents in different
      worktrees without any manual command prefixing or documentation burden.

      Tested via shell-shocked test suite: All 7 stress test scenarios passed.

      Or if you prefer a shorter version:

      feat: Auto-fix Bash tool worktree directory reset issue

      Implements pre-tool-use hook that transparently injects 'cd /worktree && '
      prefix when operating in git worktrees. Handles background processes, pipes,
      and other edge cases. Agents need zero awareness of the fix.

      Enables parallel multi-agent development in worktrees without manual prefixing.

    * fix: script name in hooks README file

    * fix: make bash-worktree-fix POSIX-compliant

    * Update .claude/hooks/config.json

    Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>

    * fix: improve bash worktree hook robustness and documentation

    - Fix critical bugs in worktree detection and command parsing:
      * Distinguish between git worktrees and submodules (check for /worktrees/ in gitdir path)
      * Fix TAB pattern bug using proper tokenization instead of brittle globbing
      * Fix overly broad dot-prefix skip that broke ./script.sh execution
      * Add proper path quoting to handle spaces and special characters
      * Capture all arguments with $* instead of just $1

    - Improve documentation:
      * Remove personal paths from examples
      * Restructure troubleshooting steps for clarity
      * Switch from jq to python -m json.tool for wider compatibility
      * Clarify that hook config goes in main .claude/settings.json

    - Simplify configuration:
      * Remove redundant config.json
      * Add settings.json.example showing proper integration
      * Fix background process detection for both "cmd &" and "cmd&" patterns

    These changes prevent false positives on commands like "cdbuild", ensure
    relative scripts work in worktrees, and handle paths with spaces correctly.

    * docs: fix a path in the instructions for how to enable the bash worktree fix hook

    * fix: harden bash worktree hook against security vulnerabilities

    Security and robustness improvements to prevent shell injection and edge cases:

    - Prevent pathname expansion attacks by disabling globbing during tokenization
      * Commands like "echo *" no longer expand before parsing
    - Fix absolute path detection for environment-prefixed commands
      * Now correctly handles "VAR=1 /usr/bin/make" patterns
    - Implement secure path quoting using single quotes with proper escaping
      * Handles paths containing quotes, spaces, and special characters
    - Optimize .git file reading using shell built-ins instead of cat
      * Avoids subshell and only reads first line for efficiency
    - Improve debug mode to accept multiple truthy values (true/TRUE/1/yes/YES)
    - Add ShellCheck compliance with consistent variable bracing and default cases

    The hook now safely handles edge cases like:
    - Paths with single quotes: /path/to/joe's-project/
    - Commands with wildcards that shouldn't expand
    - Mixed environment variables and absolute paths

    All changes maintain POSIX compatibility while significantly improving
    security posture and reliability.

    * refactor: improve worktree hook robustness and portability

      Major improvements to the bash worktree fix hook for better reliability:

      Validation & Safety:
      - Add gitdir path validation to verify the directory actually exists
      - Normalize relative gitdir paths to absolute for proper resolution
      - Add pwd directory check to handle edge cases
      - Use dirname with -- flag for safer path handling

      Portability:
      - Replace echo with printf throughout for consistent behavior
      - Use literal space/tab characters instead of [:space:] for POSIX compatibility
      - Rename shell_quote to shell_squote for clarity
      - Maintain set -f protection during pattern matching

      Simplification:
      - Streamline skip detection logic with cleaner patterns
      - Consolidate builtin command checks into single case statement
      - Remove unnecessary comments while keeping essential documentation
      - Cleaner whitespace detection using portable pattern

      The hook now properly validates that gitdir paths exist before treating
      them as worktrees, preventing false positives. Pattern matching is more
      portable across different POSIX shells, and printf usage eliminates
      echo's platform-specific quirks.

      All security improvements are retained: pathname expansion protection,
      proper single-quote escaping, and safe handling of background processes.

    * fix: restore [:space:] patterns and handle CRLF line endings

    Critical fixes and improvements to the bash worktree hook:

    Bug Fixes:
    - Handle CRLF line endings in .git files (Windows compatibility)
      * Strip trailing CR characters that break gitdir path parsing
    - Restore [:space:] character class for better Unicode support
      * Literal space/tab was too restrictive for international locales
    - Remove absolute path skip logic that broke many commands
      * Commands like `/usr/bin/npm install` still need correct cwd
      * Only skip builtins and commands that truly don't need directory context

    Performance:
    - Add fast-path exit when not in a worktree
      * Avoids unnecessary skip checks for common case
    - Remove redundant set -f globbing protection
      * Pattern matching doesn't trigger pathname expansion in case statements

    Clarity:
    - Streamline comments to be more concise
    - Quote original_command consistently
    - Better explain why absolute paths aren't skipped

    The hook now correctly handles Windows-created .git files with CRLF
    endings and properly supports all absolute-path commands that depend
    on the current working directory (which is most of them).

    * revert: remove dirname -- flag for broader compatibility

      Remove the '--' end-of-options flag from dirname command as it's not
      universally supported across all POSIX implementations. While '--' is
      good practice for handling filenames starting with dashes, dirname
      rarely encounters such edge cases in practice, and the compatibility
      issue is more pressing.

      Some older or minimal POSIX shells don't recognize '--' with dirname,
      causing the script to fail. Since we're already in a controlled loop
      walking up directory paths that come from pwd, the risk of dash-prefixed
      paths is negligible.

      Trade-off: Prioritizing compatibility over theoretical edge case protection.

    * Enhance context creation with comprehensive language detection

    Expand project detection capabilities to support 20+ programming languages and frameworks:

    - Add detection for JavaScript/TypeScript, Python (pyproject.toml), Java (Maven/Gradle),
      C#/.NET, Ruby, PHP, Dart/Flutter, Swift (including Xcode projects), C/C++, and Docker
    - Extend file type detection to include .ts, .jsx, .tsx, .kt, .cs, .rb, .php, .swift,
      .dart, .c, .h, .cpp, .hpp, and .sh files
    - Improve test detection to find both test directories and test files with common
      naming patterns (*test.*, *spec.*, test_*.*, *_test.*)
    - Exclude hidden directories in test detection for better performance

    This enhancement enables more accurate project analysis and better context creation
    across diverse development environments and technology stacks.

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    * Expand comprehensive programming language support across all project components

    Complete systematic enhancement of language detection and build system support,
    following the comprehensive approach established in context/create.md:

    **Testing Framework Detection (.claude/commands/testing/prime.md):**
    - Add detection for PHP (PHPUnit, Pest), C#/.NET (MSTest, NUnit, xUnit),
      Java (JUnit/Maven/Gradle), Kotlin, Swift (XCTest), Dart/Flutter,
      C/C++ (GoogleTest, Catch2), Ruby (RSpec, Minitest)
    - Expand framework-specific configurations with proper commands and options
    - Add comprehensive dependency checking for all package managers
    - Include installation commands for all build systems (composer, mvn, gradle,
      dotnet, bundle, flutter, swift, ctest)
    - Add test file discovery patterns for all languages

    **Context Management Enhancement:**
    - Update dependency change detection in context/update.md for 11 build systems
    - Enhance context/prime.md project file analysis beyond basic detection
    - Maintain consistency with context/create.md comprehensive language support

    **Build System Integration:**
    - Expand pm/epic-merge.md test commands from 2 to 12 build systems
    - Add Maven, Gradle, Composer, .NET, Cargo, Go, Flutter, Swift, CMake support
    - Enable automated testing across all major development ecosystems

    **Infrastructure Support:**
    - Add bash permissions for 10 additional build tools in settings.local.json
    - Update process cleanup in test-execution.md and testing/run.md for all frameworks
    - Ensure consistent cross-language development workflow support

    This enhancement enables full development lifecycle support for 20+ programming
    languages and frameworks, providing consistent project detection, dependency
    management, testing, and build capabilities across diverse technology stacks.

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    * Complete comprehensive programming language expansion with final enhancements

    Add final layer of language support improvements across infrastructure components:

    **Multi-Language Test Script (.claude/scripts/test-and-log.sh):**
    - Transform Python-only script to support 10+ languages and frameworks
    - Add intelligent test framework detection: Python (pytest), JavaScript/TypeScript
      (npm test, jest), Java (Maven, Gradle), C# (.NET), Ruby (RSpec), PHP (PHPUnit),
      Go, Rust, Swift, Dart/Flutter
    - Automatic file extension detection with appropriate test runner selection
    - Enhanced usage examples showing multi-language support

    **AST-Grep Language Reference (.claude/rules/use-ast-grep.md):**
    - Expand supported languages list from basic to comprehensive 50+ languages
    - Organize by categories: Primary Languages, Web & Markup, Configuration & Data,
      Shell & Scripts, and specialty languages
    - Add specific language codes for JavaScript/TypeScript variants, C/C++/C#,
      web frameworks (Vue, Svelte), and configuration formats (YAML, TOML, XML)

    **Issue Analysis Enhancement (.claude/commands/pm/issue-analyze.md):**
    - Generalize package.json references to support all project configuration files
    - Include build system coordination for Maven, Gradle, CMake, etc.

    **Test Runner Agent (.claude/agents/test-runner.md):**
    - Replace pytest-specific fallback with comprehensive multi-language fallback
    - Support framework detection and execution for all major testing ecosystems

    These enhancements complete the comprehensive language support initiative,
    ensuring consistent multi-language development workflow capabilities across
    all system components, from project detection to testing execution.

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    * Fix ast-grep supported languages list to match official documentation

    Correct the ast-grep language support documentation to accurately reflect
    the actual 25 languages supported according to ast-grep.github.io:

    **Removed unsupported languages:**
    - Dart (not actually supported by ast-grep)
    - Vue, Svelte (framework-specific, not core languages)
    - SCSS, SASS, LESS (CSS preprocessors not supported)
    - TOML, XML (not in official support list)
    - PowerShell, shell variants beyond bash
    - OCaml and other languages incorrectly listed

    **Corrected to official 25 languages:**
    - System: C, C++, C#, Go, Rust
    - Web: JavaScript (js/jsx), TypeScript (ts), TSX, HTML, CSS, PHP
    - General: Python, Java, Kotlin, Ruby, Swift, Scala, Lua, Elixir, Haskell
    - Config: JSON, YAML
    - Other: Bash, Nix, Solidity

    **Updated examples:**
    - Use correct language aliases (js instead of javascript, ts instead of typescript)
    - Align with actual ast-grep command patterns

    This ensures accurate guidance for developers using ast-grep integration
    and prevents errors from referencing unsupported language identifiers.

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    * Fix spelling error in README.md (#608)

    Bit of an inane PR but 🤷

    * Fix path to install script in README.md (#570)

    * Add Chinese documentation and fix typo (#676)

    - Add Chinese documentation links in README.md, COMMANDS.md, and AGENTS.md
    - Fix typo: 'mutiple' -> 'multiple' in README.md
    - Include new Chinese documentation files in doc/ directory

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-authored-by: Claude <noreply@anthropic.com>

    * Fix: Prevent local file paths exposure in GitHub issue comments (#631)

    * Fix privacy issue: Add path standards validation and normalization

    - Add .claude/rules/path-standards.md with comprehensive path usage guidelines
    - Add .claude/scripts/check-path-standards.sh for automated validation
    - Add .claude/scripts/fix-path-standards.sh for automatic path normalization
    - Add documentation for path standards tools

    Addresses issue #630: Prevents local file paths from being exposed
    in GitHub issue comments, maintaining privacy and portability.

    Features:
    - Automatic detection of absolute path violations
    - Safe path normalization with backup support
    - Comprehensive validation with colored output
    - CI/CD integration ready
    - Cross-platform support (Unix/Windows paths)

    Author: Zephyr <zephyrTang@aliyun.com>

    * Fix PR feedback: Improve scripts and documentation

    - Enhanced shell error handling with 'set -Eeuo pipefail'
    - Fixed Windows path patterns in regex expressions
    - Added exclusions for rules and backup files to prevent false positives
    - Converted all Chinese text to English in documentation
    - Corrected path examples in standards documentation
    - Improved script robustness and cross-platform compatibility

    Addresses feedback from CodeRabbit AI and GitHub Copilot reviews.

    Author: Zephyr <zephyrTang@aliyun.com>

    * Polish: Final micro-optimizations for PR feedback

    - Fixed grammar issues in path-tools-README.md per LanguageTool feedback
    - Corrected path transformation examples in path-standards.md to match actual implementation
    - Ensured consistency between documentation examples and script behavior
    - Minor improvements for production-ready code quality

    All CodeRabbit AI and GitHub Copilot feedback addressed.

    Author: Zephyr <zephyrTang@aliyun.com>

    ---------

    Co-authored-by: dengzefeng <zephyr@shopline.com>

    * Fix: Ensure GitHub issues are created in correct repository (#532)

    - Add explicit --repo flag to all gh issue create commands
    - Create repository detection configuration file
    - Handle both SSH and HTTPS git remote formats
    - Add validation and error handling for repository detection

    This prevents issues from being created in automazeio/ccpm instead of the user's actual repository.

    Co-authored-by: jodagreyhame <jodagreyhame@users.noreply.github.com>

    * Major Cleanup & Issue Resolution - 2025-01-24 (#953)

    * feat: Major cleanup and issue resolution - 2025-01-24 release

    ✨ NEW FEATURES:
    - Add comprehensive LOCAL_MODE.md guide for offline workflow
    - Auto-create GitHub labels (epic, task) during init
    - Modernize PM command syntax with !bash pattern
    - Add context accuracy safeguards with self-verification
    - Create comprehensive CHANGELOG.md documentation

    🔧 IMPROVEMENTS:
    - Update 4 core PM commands to use efficient !bash syntax
    - Enhance init.sh with automatic GitHub label creation
    - Document context accuracy improvements in CONTEXT_ACCURACY.md
    - Streamline command definitions for better token efficiency

    📊 IMPACT:
    - Resolves 10+ GitHub issues with comprehensive solutions
    - Enables true local-only development workflow
    - Reduces AI hallucination in context generation
    - Improves Claude Code compatibility and performance

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    * fix: improve GitHub detection and error handling in init.sh

    Address GitHub Copilot feedback for better robustness:

    🔧 IMPROVED REPOSITORY DETECTION:
    - Replace string matching with `gh repo view` for robust detection
    - Handles SSH URLs, GitHub Enterprise, and edge cases properly
    - Graceful fallback for non-GitHub repositories

    🛡️ ENHANCED ERROR HANDLING:
    - Replace silent failures with meaningful user feedback
    - Track label creation success/failure individually
    - Check for existing labels before reporting errors
    - Clear status messages: success, partial, or failure

    🧪 TESTED:
    - Bash syntax validation passed
    - Repository detection logic verified
    - Error handling scenarios tested

    Addresses high and medium priority feedback from Copilot review.

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    ---------

    Co-authored-by: Claude <noreply@anthropic.com>

    * Complete Directory Restructure & PM Command Modernization

    - Restructure from .claude/ to ccpm/ directory
    - Update all PM commands to use !bash execution syntax
    - Modernize command frontmatter for better token efficiency
    - Complete the missing directory restructuring from previous PRs

    This completes the architectural shift to the ccpm/ template structure.

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    * fix: correct script paths from .claude/ to ccpm/ in all PM commands

    All PM commands now properly reference ccpm/scripts/pm/ instead of
    the old .claude/scripts/pm/ paths. This completes the directory
    restructure that was partially implemented.

    - Fixed 28 script path references across all PM command files
    - Commands now correctly point to ccpm/scripts/pm/ location
    - Resolves issue where PM commands couldn't find their scripts

    🤖 Generated with [Claude Code](https://claude.ai/code)

    Co-Authored-By: Claude <noreply@anthropic.com>

    * Update installation script URLs in README files

    - Changed installation script URLs from GitHub raw links to direct links at automaze.io for improved reliability.
    - Updated both English and Chinese README files to reflect the new URLs.
    - Ensured consistency across Unix/Linux/macOS and Windows installation instructions.

    This change enhances the user experience by providing a more stable source for installation scripts.

    🤖 Generated with [Claude Code](https://claude.ai/code)

    * Update installation script URLs in README files to point to the new install path

    - Changed installation script URLs from the old path to the new `/install` path for improved clarity and reliability.
    - Updated both English and Chinese README files to reflect the new URLs.
    - Ensured consistency across Unix/Linux/macOS and Windows installation instructions.

    This change enhances the user experience by providing a clear and stable source for installation scripts.

    ---------

    Co-authored-by: James Ross <james@flyingrobots.dev>
    Co-authored-by: coderabbitai[bot] <136622811+coderabbitai[bot]@users.noreply.github.com>
    Co-authored-by: Steven <tseven@gmail.com>
    Co-authored-by: Claude <noreply@anthropic.com>
    Co-authored-by: Connor Adams <connorads@users.noreply.github.com>
    Co-authored-by: Stefan Legg <legg.stefan@gmail.com>
    Co-authored-by: cyxer000 <48980829+cyxer000@users.noreply.github.com>
    Co-authored-by: Aka mo <56784262+ZephyrDeng@users.noreply.github.com>
    Co-authored-by: dengzefeng <zephyr@shopline.com>
    Co-authored-by: jodagreyhame <66655312+jodagreyhame@users.noreply.github.com>
    Co-authored-by: jodagreyhame <jodagreyhame@users.noreply.github.com>

2025-09-24 18:50:01 +01:00

16 KiB

Raw Blame History

Claude Code 项目管理

使用规范驱动开发、GitHub issues、Git worktrees和并行运行的多个AI代理来交付更快更好的 Claude Code工作流程。

停止丢失上下文。停止在任务上阻塞。停止交付bug。这个经过实战检验的系统将PRD转化为史诗任务，将史诗任务分解为GitHub issues，并将issues转化为生产代码——每一步都有完整的可追溯性。

背景

每个团队都面临同样的问题：

会话之间上下文丢失，迫使不断重新发现
并行工作在多个开发者接触相同代码时产生冲突
需求偏离，口头决定覆盖书面规范
进度在最后之前不可见

这个系统解决了所有这些问题。

工作流程

graph LR
    A[PRD创建] --> B[史诗任务规划]
    B --> C[任务分解]
    C --> D[GitHub同步]
    D --> E[并行执行]

实际操作演示（60秒）

# 通过引导式头脑风暴创建全面的PRD
/pm:prd-new memory-system

# 将PRD转化为技术史诗任务并进行任务分解
/pm:prd-parse memory-system

# 推送到GitHub并开始并行执行
/pm:epic-oneshot memory-system
/pm:issue-start 1235

与众不同之处

传统开发	Claude Code PM系统
会话之间丢失上下文	跨所有工作的持久上下文
串行任务执行	并行代理处理独立任务
从记忆中"凭感觉编码"	规范驱动，全程可追溯
进度隐藏在分支中	GitHub中的透明审计轨迹
手动任务协调	智能优先级排序，使用`/pm:next`

为什么选择GitHub Issues？

大多数Claude Code工作流程在孤立环境中运行——单个开发者在本地环境中与AI协作。这产生了一个根本问题：AI辅助开发变成了孤岛。

通过使用GitHub Issues作为我们的数据库，我们解锁了强大的功能：

🤝 真正的团队协作

多个Claude实例可以同时处理同一项目
人类开发者通过issue评论实时查看AI进度
团队成员可以随时加入——上下文始终可见
管理者获得透明度而无需中断流程

🔄 无缝的人机交接

AI可以开始任务，人类可以完成任务（反之亦然）
进度更新对每个人可见，不会困在聊天记录中
代码审查通过PR评论自然发生
无需召开"AI做了什么？"会议

📈 超越个人工作的可扩展性

添加团队成员无需繁琐的入职流程
多个AI代理并行处理不同issues
分布式团队自动保持同步
与现有的GitHub工作流程和工具兼容

🎯 单一真相来源

无需单独的数据库或项目管理工具
Issue状态即项目状态
评论即审计轨迹
标签提供组织结构

这不仅仅是一个项目管理系统——它是一个协作协议，让人类和AI代理能够大规模协作，使用团队已经信任的基础设施。

核心原则：拒绝凭感觉编码

每一行代码都必须可追溯到规范。

我们遵循严格的5阶段纪律：

🧠 头脑风暴 - 深入思考
📝 文档化 - 编写不留任何解释空间的规范
📐 规划 - 通过明确的技术决策进行架构设计
⚡ 执行 - 精确构建规范中指定的内容
📊 跟踪 - 在每一步保持透明进度

不走捷径。不做假设。不留遗憾。

系统架构

.claude/
├── CLAUDE.md          # 始终在线的指令（将内容复制到项目的CLAUDE.md文件中）
├── agents/            # 面向任务的代理（用于上下文保存）
├── commands/          # 命令定义
│   ├── context/       # 创建、更新和准备上下文
│   ├── pm/            # ← 项目管理命令（此系统）
│   └── testing/       # 准备和执行测试（编辑此部分）
├── context/           # 项目范围的上下文文件
├── epics/             # ← PM的本地工作区（放入.gitignore中）
│   └── [epic-name]/   # 史诗任务和相关任务
│       ├── epic.md    # 实现计划
│       ├── [#].md     # 单个任务文件
│       └── updates/   # 进行中的更新
├── prds/              # ← PM的PRD文件
├── rules/             # 将任何规则文件放在此处以供引用
└── scripts/           # 将任何脚本文件放在此处以供使用

工作流程阶段

1. 产品规划阶段

/pm:prd-new feature-name

启动全面的头脑风暴，创建产品需求文档，捕捉愿景、用户故事、成功标准和约束条件。

输出： .claude/prds/feature-name.md

2. 实现规划阶段

/pm:prd-parse feature-name

将PRD转化为技术实现计划，包含架构决策、技术方法和依赖映射。

输出： .claude/epics/feature-name/epic.md

3. 任务分解阶段

/pm:epic-decompose feature-name

将史诗任务分解为具体的、可操作的任务，包含验收标准、工作量估算和并行化标志。

输出： .claude/epics/feature-name/[task].md

4. GitHub同步

/pm:epic-sync feature-name
# 或对于自信的工作流程：
/pm:epic-oneshot feature-name

将史诗任务和任务作为issues推送到GitHub，带有适当的标签和关系。

5. 执行阶段

/pm:issue-start 1234  # 启动专门代理
/pm:issue-sync 1234   # 推送进度更新
/pm:next             # 获取下一个优先任务

专门代理实现任务，同时保持进度更新和审计轨迹。

命令参考

Tip

输入/pm:help获取简洁的命令摘要

初始设置

/pm:init - 安装依赖并配置GitHub

PRD命令

/pm:prd-new - 为新产品需求启动头脑风暴
/pm:prd-parse - 将PRD转换为实现史诗任务
/pm:prd-list - 列出所有PRD
/pm:prd-edit - 编辑现有PRD
/pm:prd-status - 显示PRD实现状态

史诗任务命令

/pm:epic-decompose - 将史诗任务分解为任务文件
/pm:epic-sync - 将史诗任务和任务推送到GitHub
/pm:epic-oneshot - 一次性分解和同步命令
/pm:epic-list - 列出所有史诗任务
/pm:epic-show - 显示史诗任务及其任务
/pm:epic-close - 标记史诗任务为完成
/pm:epic-edit - 编辑史诗任务详情
/pm:epic-refresh - 从任务更新史诗任务进度

Issue命令

/pm:issue-show - 显示issue和子issues
/pm:issue-status - 检查issue状态
/pm:issue-start - 开始工作并启动专门代理
/pm:issue-sync - 将更新推送到GitHub
/pm:issue-close - 标记issue为完成
/pm:issue-reopen - 重新打开已关闭的issue
/pm:issue-edit - 编辑issue详情

工作流程命令

/pm:next - 显示下一个优先issue及史诗任务上下文
/pm:status - 整体项目仪表板
/pm:standup - 每日站会报告
/pm:blocked - 显示被阻塞的任务
/pm:in-progress - 列出进行中的工作

同步命令

/pm:sync - 与GitHub的双向同步
/pm:import - 导入现有的GitHub issues

维护命令

/pm:validate - 检查系统完整性
/pm:clean - 归档已完成的工作
/pm:search - 搜索所有内容

并行执行系统

Issues并非原子性的

传统思维：一个issue = 一个开发者 = 一个任务

现实：一个issue = 多个并行工作流

单个"实现用户认证"issue不是一个任务。它是...

代理1：数据库表和迁移
代理2：服务层和业务逻辑
代理3：API端点和中间件
代理4：UI组件和表单
代理5：测试套件和文档

所有这些都在同一工作树中同时运行。

速度的数学计算

传统方法：

包含3个issues的史诗任务
串行执行

本系统：

同样的史诗任务包含3个issues
每个issue分解为约4个并行流
12个代理同时工作

我们不是将代理分配给issues。我们是利用多个代理来更快交付。

上下文优化

传统的单线程方法：

主对话承载所有实现细节
上下文窗口填满了数据库模式、API代码、UI组件
最终达到上下文限制并失去连贯性

并行代理方法：

主线程保持干净和战略性
每个代理独立处理自己的上下文
实现细节从不污染主对话
主线程保持监督而不会淹没在代码中

你的主对话成为指挥家，而不是管弦乐队。

GitHub vs 本地：完美分离

GitHub看到的内容：

干净、简单的issues
进度更新
完成状态

本地实际发生的事情：

Issue #1234分解为5个并行代理
代理通过Git提交进行协调
复杂的编排对视图隐藏

GitHub无需知道工作是如何完成的——只需知道工作已完成。

命令流程

# 分析可以并行化的内容
/pm:issue-analyze 1234

# 启动集群
/pm:epic-start multi-agent-collaboration

# 观看奇迹发生
# 12个代理在3个issues上工作
# 全部在：../epic-memory-system/中

# 完成时进行一次干净的合并
/pm:epic-merge memory-system

主要功能和优势

🧠 上下文保存

永不丢失项目状态。每个史诗任务维护自己的上下文，代理从.claude/context/读取，并在同步前本地更新。

⚡ 并行执行

通过多个代理同时工作来更快交付。标记为parallel: true的任务支持无冲突的并发开发。

🔗 GitHub原生

与团队已使用的工具兼容。Issues是真相来源，评论提供历史，不依赖Projects API。

🤖 代理专业化

每项工作都有合适的工具。不同的代理处理UI、API和数据库工作。每个代理自动读取需求并发布更新。

📊 全程可追溯

每个决策都有文档记录。PRD → 史诗任务 → 任务 → Issue → 代码 → 提交。从想法到生产的完整审计轨迹。

🚀 开发者生产力

专注于构建，而非管理。智能优先级排序，自动上下文加载，准备就绪时增量同步。

已验证的结果

使用此系统的团队报告：

89%的时间不再因上下文切换而丢失——你将很少使用/compact和/clear
5-8个并行任务 vs 之前的1个——同时编辑/测试多个文件
bug率降低75%——由于将功能分解为详细任务
功能交付速度提升3倍——基于功能大小和复杂度

示例流程

# 开始新功能
/pm:prd-new memory-system

# 审查和完善PRD...

# 创建实现计划
/pm:prd-parse memory-system

# 审查史诗任务...

# 分解为任务并推送到GitHub
/pm:epic-oneshot memory-system
# 创建issues：#1234（史诗任务），#1235，#1236（任务）

# 开始任务开发
/pm:issue-start 1235
# 代理开始工作，在本地维护进度

# 同步进度到GitHub
/pm:issue-sync 1235
# 更新作为issue评论发布

# 检查整体状态
/pm:epic-show multi-agent-collaboration

立即开始

快速设置（2分钟）

将此仓库安装到你的项目中：

Unix/Linux/macOS
```
cd path/to/your/project/
curl -sSL https://automaze.io/ccpm/install | bash
# 或：wget -qO- https://automaze.io/ccpm/install | bash
```
Windows（PowerShell）
```
cd path/to/your/project/
iwr -useb https://automaze.io/ccpm/install | iex
```
⚠️ 重要：如果你已有.claude目录，请将此仓库克隆到不同目录，然后将克隆的.claude目录内容复制到你项目的.claude目录中。

查看完整的/其他安装选项在安装指南 ›中
初始化PM系统：
```
/pm:init
```
此命令将：
- 安装GitHub CLI（如需要）
- 与GitHub进行身份验证
- 安装gh-sub-issue扩展以建立正确的父子关系
- 创建所需目录
- 更新.gitignore
创建包含仓库信息的CLAUDE.md
```
/init include rules from .claude/CLAUDE.md
```
如果你已有CLAUDE.md文件，运行：/re-init来用.claude/CLAUDE.md中的重要规则更新它。
准备系统：
```
/context:create
```

开始你的第一个功能

/pm:prd-new your-feature-name

观看结构化规划如何转化为交付的代码。

本地 vs 远程

操作	本地	GitHub
PRD创建	✅	—
实现规划	✅	—
任务分解	✅	✅（同步）
执行	✅	—
状态更新	✅	✅（同步）
最终交付物	—	✅

技术说明

GitHub集成

使用gh-sub-issue扩展建立正确的父子关系
如果未安装扩展则回退到任务列表
史诗任务issues自动跟踪子任务完成情况
标签提供额外组织（epic:feature，task:feature）

文件命名约定

任务在分解期间以001.md，002.md开始
GitHub同步后，重命名为{issue-id}.md（例如，1234.md）
便于导航：issue #1234 = 文件1234.md

设计决策

故意避免GitHub Projects API的复杂性
所有命令首先在本地文件上操作以提高速度
与GitHub的同步是明确且受控的
Worktrees为并行工作提供干净的git隔离
GitHub Projects可以单独添加用于可视化

支持这个项目

Claude Code PM由Automaze开发，为交付产品的开发者，由交付产品的开发者。

如果Claude Code PM帮助你的团队交付更好的软件：

⭐ 给这个仓库点赞 来表达你的支持
🐦 在X上关注@aroussi 获取更新和提示

Tip

使用Automaze更快交付。 我们与创始人合作，将他们的愿景变为现实，扩展他们的业务，并优化成功。 访问Automaze与我预约通话 ›

16 KiB Raw Blame History Unescape Escape