Skip to main content

​
CLAUDE.md Best Practices

Keep CLAUDE.md under 150 lines for reliable adherence. HumanLayer uses 60 lines as their guideline, though this isn’t 100% guaranteed.

​
Multiple CLAUDE.md for Monorepos

For large monorepo projects, use multiple CLAUDE.md files with ancestor + descendant loading.
See Memory Best Practices for detailed guidance on CLAUDE.md loading behavior in monorepos.

​
Commands vs Agents

Use commands for your workflows instead of agents. Commands provide better entry points for workflows.
Create feature-specific subagents with skills (progressive disclosure) instead of general-purpose agents like “qa” or “backend engineer.”

​
Context Management

​
Manual Compacting

Avoid the “agent dumb zone” — perform manual /compact at around 50% context usage for best results.

​
Memory and Rules Limitations

/memory, /rules, and constitution.md do not guarantee anything. These are guidance mechanisms, not strict enforcement.

​
Planning and Execution

​
Always Start with Plan Mode

Always start complex tasks with plan mode for better task breakdown and execution.

​
Task Size Guidelines

  • Vanilla Claude Code is better than complex workflows for smaller tasks
  • Break subtasks small enough to complete in under 50% context usage
  • Use human-gated task list workflow for multi-step tasks

​
Advanced Workflows

​
Parallel Development

Use agent teams and git worktrees for parallel development across isolated branches.

​
Long-Running Tasks

Use the Ralph Wiggum plugin for long-running autonomous tasks that iterate until completion.

​
Daily and Hourly Habits

​
Daily Updates

Update Claude Code daily and start your day by reading the changelog.

​
Commit Often

Commit often — as soon as a task is completed, commit. This creates better checkpoints for rewinding and helps with context management.

Build docs developers (and LLMs) love

Get started for freeTalk to us