Leanctx Integration

Automatically compresses OpenClaw tool outputs to reduce token usage by 60-99%

LeanContext Integration

LeanContext integration for OpenClaw that automatically compresses tool outputs to reduce token usage by 60-99%.

External Endpoints

Endpoint	Data Sent	Purpose
None	N/A	This skill operates entirely locally

No external API calls are made. All processing happens on your local machine.

Security & Privacy

• Local-only operation: All compression and caching happens locally
• No data leaves your machine: No files, commands, or content is sent to external services
• Cache location: Stored in memory (optional filesystem cache can be configured)
• No telemetry: No usage data, metrics, or analytics are transmitted

Model Invocation

This skill operates autonomously during OpenClaw's normal operation. It intercepts read and exec tool calls transparently without requiring explicit user approval for each invocation. This is standard behavior for OpenClaw skills.

To disable automatic invocation, set enabled: false in your OpenClaw config.

Trust Statement

By using this skill, your file contents and command outputs are processed locally for compression purposes. No data is sent to external services. Only install if you are comfortable with this local processing.

What it does

Intercepts OpenClaw's read and exec tool calls and applies intelligent compression:

• File reads: AST-aware compression removes comments, whitespace, collapses function bodies
• Shell exec: Pattern-matching compresses common commands (git, npm, docker, etc.)
• Session caching: Re-reads cost ~13 tokens instead of thousands

Installation

bash

openclaw skills install leanctx-integration

Or manually:

bash

cd ~/.openclaw/workspace/skills
git clone https://github.com/your-repo/leanctx-integration.git
cd leanctx-integration
npm install
npm run build

Configuration

Add to your openclaw.json:

json

{
  "skills": {
    "leanctx-integration": {
      "enabled": true,
      "config": {
        "threshold": 100,
        "cacheEnabled": true,
        "excludedPaths": ["node_modules", ".git", "dist"],
        "excludedCommands": ["cat", "echo"]
      }
    }
  }
}

Usage

Once installed, LeanContext works automatically - no code changes required!

Check Metrics

Note: Metrics tracking has been temporarily disabled. The skill automatically compresses tool outputs, but session-level metrics reporting is not currently available. This feature may be revisited in a future update.

Clear Cache

Via your agent: Ask: "Clear the LeanCTX cache"

Direct CLI:

bash

npx leanctx clear-cache

Compression Examples

TypeScript File (800 lines)

Before: ~2000 tokens
After: ~50 tokens
Savings: 97.5%

Git Log

Before: ~500 tokens (full commit info)
After: ~30 tokens (hashes + messages)
Savings: 94%

NPM Install

Before: ~200 tokens (all download messages)
After: ~10 tokens (summary only)
Savings: 95%

Supported Languages

• TypeScript / JavaScript
• Python
• Svelte
• Go
• Rust
• Java / Kotlin
• Generic (fallback)

Supported Commands

• git log → Commit hashes + first line
• git status → Changed files only
• git diff → Diff markers + changes
• npm install → Package count
• npm test → Test results + summary
• cargo build → Errors only
• docker ps → Header + first 10 rows
• kubectl → Header + first 15 rows

Performance

• Cache hit: <5ms overhead
• First read: ~10ms (AST parse)
• Compression: 60-99% token reduction

Troubleshooting

Not working?

1. Check if skill is enabled in config
2. Verify threshold isn't too high
3. Check excluded paths aren't matching

Want to disable for specific files?

Add to excludedPaths in config:

json

{
  "excludedPaths": ["node_modules", ".git", "**/*.spec.ts"]
}

Cache too large?

Ask your agent: "Clear the LeanCTX cache"

Or via CLI:

bash

npx leanctx clear-cache

How it works

1. OpenClaw makes a tool call (read or exec)
2. LeanContext intercepts the call
3. Checks cache (instant return if hit)
4. Compresses based on file type or command pattern
5. Caches result
6. Returns compressed output to OpenClaw
7. Metrics updated

License

MIT