Skip to content

quickstart.md

Latest commit

 

History

History
252 lines (177 loc) · 8.19 KB

quickstart.md

File metadata and controls

252 lines (177 loc) · 8.19 KB
Raw

Quick Start Guide

Get productive with CC-Best in 5 minutes.

1. Install (30 seconds)

# Add marketplace and install
/plugin marketplace add xiaobei930/cc-best
/plugin install cc-best@xiaobei930

# Verify installation
/cc-best:status

You should see a summary of all installed components (44 commands, 19 skills, 8 agents, 43 rules).

2. Your First Iterate (2 minutes)

/cc-best:iterate "add a login page with email and password"

What Happens Next?

Claude automatically runs a full development pipeline:

1. 📋 /cc-best:pm       → Analyzes requirements, creates REQ-001
2. 🔍 /cc-best:clarify  → Clarifies any ambiguous points (if needed)
3. 🏗️ /cc-best:lead     → Designs technical solution, creates DES-001 + TSK-001
4. 🎨 /cc-best:designer → Generates UI guidance (frontend tasks only)
5. 💻 /cc-best:dev      → Writes code, runs self-tests
6. ✅ /cc-best:verify   → Build + type-check + lint + test + security scan
7. 🧪 /cc-best:qa       → Functional acceptance testing
8. 📦 /cc-best:commit   → Creates conventional commit

You just watch. Intervene only when needed (Ctrl+C to pause).

How Iterate Selects Roles

The iterate engine reads memory-bank/progress.md and selects the appropriate role:

Current State Role Selected Action
No requirements doc /cc-best:pm Requirement analysis
REQ has low-confidence items /cc-best:clarify Requirement clarification
Has REQ, no design /cc-best:lead Technical design
Has design, frontend tasks /cc-best:designer UI design guidance
Has tasks to implement /cc-best:dev Coding implementation
Code ready for verification /cc-best:verify Comprehensive verification
Verification passed /cc-best:qa Functional acceptance

When to Intervene

  • Ctrl+C: Pause immediately. Progress is saved to memory-bank/progress.md.
  • Type anything: Claude pauses and waits for your input.
  • Resume: Continue the conversation — Claude picks up where it left off.

Stop Conditions

Iterate stops only when:

  1. All tasks are completed
  2. You interrupt (Ctrl+C or Escape)
  3. A fatal error occurs that cannot be auto-resolved
  4. An external dependency requires your decision

3. Understanding the Role Pipeline

Each role has clear boundaries:

PM (what to build)
 └→ Lead (how to build)
     └→ Designer (how it looks — frontend only)
         └→ Dev (build it)
             └→ QA (verify it works)
                 └→ Verify + Commit (ship it)

Role Boundaries

Every role follows MUST/SHOULD/NEVER rules:

Role MUST NEVER
PM Analyze requirements autonomously Guess APIs, skip context reading
Lead Review PM decisions, create design Start coding, skip decomposition
Dev Follow tech design, self-test Modify unassigned modules
QA Test all acceptance criteria Modify source code

Document Traceability

Roles create numbered documents that link together:

REQ-001 (PM creates)
 └→ DES-001 (Lead creates, references REQ-001)
     └→ TSK-001, TSK-002 (Lead creates, reference DES-001)

Downstream Correction (A3)

  • Lead reviews PM: Can adjust requirement priority or scope
  • QA distinguishes bugs: Implementation bugs → back to Dev; Requirement assumption errors → flagged for PM review

4. Try Pair Programming

When you want step-by-step collaboration instead of full autonomy:

/cc-best:pair "help me refactor this authentication module"

iterate vs pair

Aspect /cc-best:iterate /cc-best:pair
Control Fully autonomous Confirm each step
Best for Clear tasks, batch work Learning, sensitive operations
Risk Medium (post-check) Low (pre-confirm)

5 Confirmation Checkpoints

In pair mode, Claude always asks before:

  1. Understanding — "I understand you need X. Correct?"
  2. Design choice — "Option A or B? I recommend A because..."
  3. Destructive action — "About to delete X. Confirm?"
  4. External call — "Will call production API. Proceed?"
  5. Commit — "Commit message: '...'. OK?"

Learning Mode

/cc-best:pair --learn "teach me how to write unit tests"

Claude adjusts its behavior:

  • Explains why before each step
  • Shows intermediate results
  • Explains errors and debugging process
  • Encourages you to ask questions

5. Top 10 Commands

Command What It Does
/cc-best:iterate Autonomous full-pipeline development
/cc-best:pair Step-by-step collaborative development
/cc-best:pm Analyze requirements, create REQ document
/cc-best:lead Design technical solution, create DES/TSK
/cc-best:dev Write code following the tech design
/cc-best:qa Test and validate implementation
/cc-best:commit Create conventional commit
/cc-best:status Check plugin installation status
/cc-best:checkpoint Save current context for later recovery
/cc-best:catchup Restore context from previous session

6. Setting Up Your Project

Minimal CLAUDE.md Template

Your CLAUDE.md should define project-specific constraints. Keep it under 100 lines:

# Project Name

## Core Constraints

- **Language**: TypeScript + React
- **Testing**: Vitest, coverage > 80%
- **API**: REST, OpenAPI spec in docs/api/

## Current State

See `memory-bank/progress.md`

## Forbidden

- No `any` types
- No committing .env files

Initialize Memory Bank

# Plugin users: automatic on first session
# Clone users: run init script
bash scripts/shell/init.sh

This creates:

  • memory-bank/progress.md — Rolling task tracker
  • memory-bank/architecture.md — System design decisions
  • memory-bank/tech-stack.md — Technology choices

Cross-Session Recovery

# Before ending a session — save context
/cc-best:checkpoint

# Starting a new session — restore context
/cc-best:catchup

# Context getting too long — compress it
/cc-best:compact-context

7. Quick FAQ

How do I stop iterate? Press Ctrl+C. Progress is automatically saved to memory-bank/progress.md. Resume by continuing the conversation or running /cc-best:catchup.

What if QA fails? Claude automatically sends it back to Dev for fixing (up to 3 attempts). After that, it asks for your input.

How do I customize rules? Add .md files to rules/. Use paths frontmatter to scope them:

---
paths:
  - "**/*.py"
---

# Your Python Rule

Can I use only some roles? Yes. Run individual role commands directly: /cc-best:dev "fix this bug" skips PM/Lead.

How do I add a new language? Create a new directory under rules/ (e.g., rules/rust/) with style, testing, security, and performance rule files.

How do I switch model strategy? Run /cc-best:model to interactively select between quality (all Opus), balanced (design Opus + execution Sonnet), or economy (core Sonnet + rest Haiku). Use /cc-best:model --show to see current configuration.

What is Lite mode? Lite mode simplifies the iterate pipeline to Dev → Verify → Commit, skipping PM/Lead/Designer/QA roles. Configure it in memory-bank/config.json or during /cc-best:setup --interactive.

How do I make a quick fix? Use /cc-best:hotfix "description" for urgent bug fixes. It goes straight to Dev → Verify → Commit without the full pipeline.

Next: Read the Advanced Guide for deep dives into the methodology, decision principles, and knowledge pipeline.