Claude Code Best Practice Tips Compendium

中文 | English

Summary

This source page groups the new tip-style clippings imported alongside claude-code-best-practice: Boris Cherny threads on parallelism, plan mode, code review, PR size, hidden features, and customization, plus Thariq's longer note on how Anthropic uses skills.

Source

• Raw family:
  • 10 tips | 中文
  • 12 customization tips | 中文
  • 13 usage tips | 中文
  • 15 hidden features | 中文
  • Code review and test-time compute | 中文
  • Squash merge and PR size | 中文
  • How Anthropic uses skills | 中文
• Source family URL: https://github.com/shanraisshan/claude-code-best-practice
• Ingest date: 2026-04-15

Key Contributions

• Reinforces parallel worktrees and multiple isolated sessions as one of the biggest practical productivity levers.
• Makes plan mode, review mode, and independent context windows sound like operating defaults rather than occasional tricks.
• Surfaces a high-throughput merge philosophy: many small PRs, squash by default, and cheap follow-up fixes.
• Adds a stronger ecosystem-level story around skills: not single markdown files, but discoverable folders with scripts, assets, and clearer category boundaries.

Practical Implications

• Claude Code practice is converging on explicit test-time compute, not only “one chat, one task.”
• Review quality depends on structural independence between contexts, not just more attention inside one thread.
• Teams adopting agent-heavy development need merge and PR hygiene that match the new throughput level.

Related Pages

• Claude Code Best Practice Repository
• Agent Harness Design
• Agent-First Repositories
• Codex Operating Practices