efficient-opus

A Claude Code slash command that runs a task efficiently: it decomposes the task into independent subtasks, runs each on a parallel Haiku subagent, then synthesizes the results.

Your main Claude Code session (Opus/Sonnet) does the planning; cheap, fast Haiku subagents do the parallel grunt work. Everything runs on your Claude Code subscription — no API key, no separate billing.

Requirements

• Claude Code with access to the Task tool (used to spawn subagents)
• Node.js 18+ (only to run the one-time installer)

Install

npx efficient-opus

This copies the /efficient-opus command into ~/.claude/commands/. Restart Claude Code, then confirm it appears when you type /.

Usage

/efficient-opus Write unit tests for every file in src/utils

1. Claude Code breaks the task into independent subtasks (max 10) and shows you the plan.
2. It asks: "Run these N parallel Haiku subagents? yes or no".
3. On yes, it spawns all subtasks in parallel as Haiku subagents.
4. It synthesizes their outputs into one final result, flagging any subtask that failed.

Caveats

• Cost / quota: each subtask spawns a separate subagent run. An N-subtask plan does N parallel runs and consumes proportional quota — the gate step lets you decline before any spawn. Best for coarse, independent work (one file/module/component each), not fine-grained or sequential tasks.
• Haiku quality: subagents run on Haiku, tuned for speed and cost. For work needing deep reasoning, prefer running it directly in your main session.

How it works

The whole tool is a single markdown command (commands/efficient-opus.md). There is no server and no API key — the command instructs Claude Code to use its built-in Task tool to fan work out to Haiku subagents on your subscription.

Development

npm test          # installer unit tests (vitest)
npm run typecheck # TypeScript strict check
npm run build     # compile installer to dist/