GitHub - shanraisshan/claude-code-best-practice: practice made claude perfect
2 hours ago
- #AI Development Workflow
- #Claude Best Practices
- #Context Management
- Skills and sub-agents should be used for progressive disclosure to keep main context clean.
- Commands should be created for repetitive tasks, stored in .claude/commands/ and version-controlled.
- CLAUDE.md must include essential setup/build/test commands for reproducible environments.
- Hooks can enforce permissions, format code, measure usage, and manage destructive commands.
- Context forking isolates subagent tasks, preventing intermediate steps from cluttering main context.
- Model selection is flexible: use Opus for planning, Sonnet for coding, and enable thinking mode for transparency.
- Regularly compact context, reset with /clear for task switching, and monitor usage to optimize workflow.