ccpm-claude-code-pm/COMMANDS.md

Commands

中文文档 (Chinese Documentation)

Complete reference of all commands available in the Claude Code PM system.

Note

: Project Management commands (/pm:*) are documented in the main README.md.

Table of Contents

• Context Commands
• Testing Commands
• Utility Commands
• Review Commands

Context Commands

Commands for managing project context in .claude/context/.

/context:create

• Purpose: Create initial project context documentation
• Usage: /context:create
• Description: Analyzes the project structure and creates comprehensive baseline documentation in .claude/context/. Includes project overview, architecture, dependencies, and patterns.
• When to use: At project start or when context needs full rebuild
• Output: Multiple context files covering different aspects of the project

/context:update

• Purpose: Update existing context with recent changes
• Usage: /context:update
• Description: Refreshes context documentation based on recent code changes, new features, or architectural updates. Preserves existing context while adding new information.
• When to use: After significant changes or before major work sessions
• Output: Updated context files with change tracking

/context:prime

• Purpose: Load context into current conversation
• Usage: /context:prime
• Description: Reads all context files and loads them into the current conversation's memory. Essential for maintaining project awareness.
• When to use: At the start of any work session
• Output: Confirmation of loaded context

Testing Commands

Commands for test configuration and execution.

/testing:prime

• Purpose: Configure testing setup
• Usage: /testing:prime
• Description: Detects and configures the project's testing framework, creates testing configuration, and prepares the test-runner agent.
• When to use: Initial project setup or when testing framework changes
• Output: .claude/testing-config.md with test commands and patterns

/testing:run

• Purpose: Execute tests with intelligent analysis
• Usage: /testing:run [test_target]
• Description: Runs tests using the test-runner agent which captures output to logs and returns only essential results to preserve context.
• Options:
  • No arguments: Run all tests
  • File path: Run specific test file
  • Pattern: Run tests matching pattern
• Output: Test summary with failures analyzed, no verbose output in main thread

Utility Commands

General utility and maintenance commands.

/prompt

• Purpose: Handle complex prompts with multiple references
• Usage: Write your prompt in the file, then type /prompt
• Description: Ephemeral command for when complex prompts with numerous @ references fail in direct input. The prompt is written to the command file first, then executed.
• When to use: When Claude's UI rejects complex prompts
• Output: Executes the written prompt

/re-init

• Purpose: Update or create CLAUDE.md with PM rules
• Usage: /re-init
• Description: Updates the project's CLAUDE.md file with rules from .claude/CLAUDE.md, ensuring Claude instances have proper instructions.
• When to use: After cloning PM system or updating rules
• Output: Updated CLAUDE.md in project root

Review Commands

Commands for handling external code review tools.

/code-rabbit

• Purpose: Process CodeRabbit review comments intelligently
• Usage: /code-rabbit then paste comments
• Description: Evaluates CodeRabbit suggestions with context awareness, accepting valid improvements while ignoring context-unaware suggestions. Spawns parallel agents for multi-file reviews.
• Features:
  • Understands CodeRabbit lacks full context
  • Accepts: Real bugs, security issues, resource leaks
  • Ignores: Style preferences, irrelevant patterns
  • Parallel processing for multiple files
• Output: Summary of accepted/ignored suggestions with reasoning

Command Patterns

All commands follow consistent patterns:

Allowed Tools

Each command specifies its required tools in frontmatter:

• Read, Write, LS - File operations
• Bash - System commands
• Task - Sub-agent spawning
• Grep - Code searching

Error Handling

Commands follow fail-fast principles:

• Check prerequisites first
• Clear error messages with solutions
• Never leave partial state

Context Preservation

Commands that process lots of information:

• Use agents to shield main thread from verbose output
• Return summaries, not raw data
• Preserve only essential information

Creating Custom Commands

To add new commands:

1. Create file: commands/category/command-name.md

2. Add frontmatter:

   ---
   allowed-tools: Read, Write, LS
   ---
   

3. Structure content:

   • Purpose and usage
   • Preflight checks
   • Step-by-step instructions
   • Error handling
   • Output format

4. Follow patterns:

   • Keep it simple (no over-validation)
   • Fail fast with clear messages
   • Use agents for heavy processing
   • Return concise output

Integration with Agents

Commands often use agents for heavy lifting:

• test-runner: Executes tests, analyzes results
• file-analyzer: Summarizes verbose files
• code-analyzer: Hunts bugs across codebase
• parallel-worker: Coordinates parallel execution

This keeps the main conversation context clean while doing complex work.

Notes

• Commands are markdown files interpreted as instructions
• The / prefix triggers command execution
• Commands can spawn agents for context preservation
• All PM commands (/pm:*) are documented in the main README
• Commands follow rules defined in /rules/