jnbno1163

jnbno1163 / LG-token-saver

Public

Save 87% token usage for Claude Code. Zero install. 8 rules covering input+output+context. 6 months verified.

ai-tools anthropic claude claude-code claude-code-skill
10
3
69% credibility
Found Jun 01, 2026 at 10 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
AI Summary

LG-token-saver is a productivity tool for Claude Code that claims to reduce AI usage costs by up to 87%. It works by modifying how the AI assistant behaves: making it read files more efficiently (showing summaries instead of full content), filtering command-line output to show only relevant results, and encouraging more concise responses. The tool offers three intensity levels and requires no additional dependencies or complicated setup. The author previously built tools to help developers access Claude Code in regions where it wasn't officially available. The benchmarks show significant token reduction, though this is achieved by reducing the detail and context the AI works with rather than improving efficiency in a traditional sense.

How It Works

1
💬 You hear about saving money on AI

A friend mentions they've cut their AI assistant costs by nearly 90% and want to share how.

2
📦 You discover the tool

You learn there's a simple tool that claims to make your AI assistant work more efficiently without losing quality.

3
✨ You install it in one step

With just one command, the tool is added to your AI assistant. No complicated setup or extra accounts needed.

4
🤖 Your AI assistant learns new habits

The tool teaches your AI assistant to be more concise: shorter replies, smarter file reading, and filtered command output.

5
You choose your savings level
🟢
Lite mode

Subtle savings, barely noticeable difference in behavior

🟡
Full mode

Balanced savings and quality, the recommended starting point

🔴
Ultra mode

Maximum savings, every trick enabled for biggest reduction

6
💰 You work normally

You continue asking your AI assistant questions and helping it with tasks, but now it responds more efficiently.

🎉 Your bills shrink dramatically

The same work gets done with a fraction of the cost. Tasks that used to cost hundreds of dollars now cost dozens.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 10 to 10 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is LG-token-saver?

LG-token-saver is a Claude Code Skill that reduces token consumption by optimizing how the AI processes input, generates output, and manages conversation context. It works as a single configuration file you add to Claude Code, implementing three optimization layers: input filtering through SubAgent isolation and grep-first patterns, output compression via bash filtering and terse responses, and context management through deduplication and compact summaries. The project offers three modes (Lite, Full, Ultra) targeting different savings levels from 65% to 87%.

Why is it gaining traction?

The main draw is simplicity. Unlike ECC which requires 64 skills and 27 agents, you install LG-token-saver with a single npx command and it auto-activates. The three-tier savings model lets developers tune aggressiveness based on use case, and the benchmarks show meaningful real-world differences across scenarios like article builds (92% savings) versus simple Q&A (60%). Six months of production data backs the claims, which is more than most token optimization tools offer.

Who should use this?

Developers running heavy Claude Code workloads who are watching API costs will get the most value. If you're building articles, debugging across multiple directories, or running parallel agents, the savings compound quickly. Teams in regions without official Claude Code support might find it especially relevant since it works with proxy solutions like cc-switch. For casual users asking occasional questions, the overhead might not justify the savings.

Verdict

The concept is solid and the single-file approach is genuinely clever. However, with only 10 stars and a credibility score of 0.7%, this is early-stage software without community validation. The Chinese documentation in the secondary README adds some friction for non-Cantonese speakers. Worth trying in Lite mode to test against your specific workflow before committing, but don't bet critical pipelines on it until adoption picks up.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.