The Ultimate Open‑Source Companion for Claude Code & Codex in IDEA (3.3k Stars)

Project Differentiation

The claude-code-best-practice repository does not merely explain what Claude Code commands do; it records community‑validated practices for using Claude Code effectively. For example, while the official docs describe the purpose of /compact, the repository notes that Claude’s comprehension degrades once the total context reaches roughly 300 k–400 k tokens. The recommended practice is to keep context utilization below 40 % and invoke /compact proactively.

Four Main Sections

Section 1 – Core Concepts Covers the ten foundational concepts of Claude Code: Subagents, Commands, Skills, Workflows, Hooks, MCP Servers, Plugins, Settings, Status Line, and Memory. Each concept links to a concrete “best‑practice” example. The MCP Servers subsection lists the most useful MCPs, such as chrome-devtools-mcp (browser debugging), github-mcp (direct PR manipulation), and sequential-thinking (forced step‑by‑step reasoning).

Section 2 – Popular Features Tracks Anthropic’s latest beta features (Ultrareview, Devcontainers, Channels, Ultraplan, Auto Mode, Computer Use, Agent SDK, Code Review, GitHub Actions, Agent Teams, Scheduled Tasks, etc.) and aggregates early‑adopter experiences before official documentation catches up.

Section 3 – Development Workflows Compares 13 well‑known Claude Code workflows (e.g., Superpowers, Everything Claude Code, Spec Kit, gstack, Get Shit Done). For each workflow the repository records star count, uniqueness, presence of a planning stage, and the number of Subagents/Commands/Skills used. A single table lets teams select the workflow that matches their needs. All successful workflows converge on the five‑step pattern: Research → Plan → Execute → Review → Ship .

Section 4 – 83 Practical Tips Organizes 83 tips into 15 categories (Prompting, Planning, Context, Session Management, Memory, Agents, Commands, Skills, Hooks, Workflows, Advanced Workflows, Git/PR, Debugging, Utilities, Daily Practices). Notable tips include using screenshots when Claude struggles, and leveraging chrome-devtools-mcp to let Claude read console logs.

Three Most Actionable Tips

Keep Context Utilization Below 40 % – When the status line shows utilization >40 %, execute /compact immediately. The repository records “Context rot at ~300‑400 k tokens”; staying under 40 % prevents degradation.

Separate Planning and Execution – Follow the pattern Research → Plan → Execute → Review → Ship . After the Plan stage, commit the plan to a markdown file, start a new Claude session, and have Claude read that markdown before executing. Do not carry intermediate drafts from the Plan stage into Execute.

Use Hooks for Discipline, Not Features – Define a PostToolUse hook that runs mvn test automatically whenever Claude modifies files under src/. In multi‑module Spring Boot projects this prevents committing code without running the corresponding module’s tests.

Accessing the Chinese Version

The repository’s default language is English. Chinese translations are contributed via pull requests. To view a Chinese version, locate a PR that adds a translation and check out its branch.

Key Technical Observations

Context size beyond 300 k–400 k tokens causes “context rot,” reducing Claude’s ability to follow earlier instructions.

The five‑step workflow (Research → Plan → Execute → Review → Ship) emerged from community analysis of successful projects.

Combining Claude Code for code generation with Codex for code review yields more objective feedback than a single model self‑review.

Repository URL: https://github.com/shanraisshan/claude-code-best-practice