omer-faruk-g

ClaudeLean — Productivity skill for Claude Code

Productivity community

Claude Code skill that cuts agentic token waste: a Read guard for whole-file reads, an Edit-over-Write note, a reminder after verbose replies, and a transcript report.

How to install ClaudeLean

This entry records only its repository, not the path inside it, so there is no exact command to give. Open omer-faruk-g/ClaudeLean and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

What ClaudeLean does

Claude Code skill that cuts agentic token waste: a Read guard for whole-file reads, an Edit-over-Write note, a reminder after verbose replies, and a transcript report. GPL-3.0.

Alternatives in Productivity

  • Performance Optimization Skill — Systematic performance analysis and optimization for RTK CLI tool, focusing on startup time (<10ms), memory us 11.9k ★
  • Jobs — Create, list, edit, or delete cron jobs 1.3k ★
  • Maestro Overlay — Create or edit command overlays from natural language, or auto-generate them from workflow deficiency signals 530 ★

README

ClaudeLean

ClaudeLean is a [Claude Code](https://code.claude.com) skill that cuts the tokens agentic coding wastes on both sides of the conversation.

  • Input. While solving a problem, Claude runs Read, Grep, ls and shell commands. Every file and log it pulls in stays in the context and is re-sent with every later request. Reading twenty whole files to find one bug fills the context with hundreds of thousands of tokens.
  • Output. Output tokens cost about five times as much as input tokens. Long explanations of why a solution was chosen, unrequested alternatives, walkthroughs of how the code works, visible self-correction, and reprinted imports and unchanged classes all spend them on text you did not need.

ClaudeLean gives Claude rules for both sides, and three small hooks that enforce or remind where rules alone slip.

What it does

Part What happens
Rules (SKILL.md) Locate with Grep before reading and read ranges, not whole files. Follow the call chain from the error, not twenty files. Don't re-read unchanged files. Cap command output; give wide searches to a subagent. Lead with the result plus at most 3 short lines. Explain, compare or walk through only when asked. Keep self-correction internal. Use Edit for existing files and show diffs in chat; never reprint unchanged imports, types or classes. Your explicit request always wins.
Read guard (PreToolUse, Read) Blocks a whole-file read of a text file over 400 lines that has no limit. The message tells Claude to Grep for the part it needs and read that range, or to call Read again with an explicit limit if it really needs everything. Images, PDFs, notebooks and binaries pass.
Write note (PreToolUse, Write) When a Write repeats most of an existing file (60%+ of its lines, files of 80+ lines), Claude gets a note to use Edit next time. The write still happens: blocking it would only spend the tokens twice.
Reply note (UserPromptSubmit) If Claude's last