Stop Claude Code from spawning three agents to fix a one-line bug.
You ask Claude to fix a null check. It spawns a spec reviewer, a code quality agent, and a test agent. Five minutes later, you have a review. The fix took 30 seconds.
You ask Claude to fix a null check. It fixes the null check. You move on.
Claude Code's 2026 default behavior spawns multiple subagents for most tasks: a spec reviewer, a quality reviewer, sometimes a test agent. For a 5-file feature, that's useful. For a one-line fix, you're paying 3x the input tokens and waiting 3-5 minutes instead of 30 seconds.
There's no built-in way to tell Claude "just do it" without configuring this explicitly. Lean Mode is that configuration.
The skill defines three complexity tiers:
The default is Tier 1. You override per-task ("Treat this as Tier 3") or toggle for the session ("Lean mode off").
mkdir -p .claude/skills cp lean-mode/skills/lean-mode.md .claude/skills/
Then in Claude Code: /lean-mode
Or skip the skill: copy one of the three CLAUDE.md blocks from claude-snippets/ into your project's CLAUDE.md. That's all it takes.
On a one-line bug fix, the multi-agent default runs three separate contexts. Each inherits your full conversation history. At 10k tokens of conversation, you're billed 30k instead of 10k in input tokens. The bigger cost is time: 2-5 minutes per task instead of 30-60 seconds lean.
Questions: zac@builtbyzac.com