Repolode
Submit
JayPokale

JayPokale/Chisle

A zero-dependency plugin that trims an AI coding agent's token use by tightening its prose, code style, and oversized tool output.

View on GitHubWebsite
Stars
740
Forks
50
Open issues
0
Last commit
13 days ago

What is it?

Chisle is a plugin for AI coding agents, mainly Claude Code, that reduces the tokens they consume without changing answer quality. It works on three axes: a ruleset that makes the model write terser prose, a YAGNI-first rule set that favors the smallest working code, and a PostToolUse hook that shrinks oversized tool output before it re-enters the context window.

Installation is a single command, npx chisle, which detects supported agents and wires each one. The same ruleset is distributed to Pi, Cursor, Codex, Gemini, Copilot, OpenCode, Antigravity and several others. An MCP wrapper command extends tool-output compression to stdio MCP servers.

The project publishes benchmark transcripts, including runs where it performed worse than alternatives. Its results report answer length relative to a bare model, and accuracy tests on maths, coding and agentic repositories.

Key features

  • 01Shortens answer prose by removing filler and hedging.
  • 02Favors the smallest working code and avoids speculative abstractions.
  • 03Compresses oversized tool output in Claude Code, Pi, OpenCode and Copilot CLI hooks.
  • 04Wraps stdio MCP servers to compress their tool results.
  • 05Installs with one command and supports 13 agents, with an uninstall option.

Use cases

  • Reducing token costs for daily Claude Code sessions with long tool output.
  • Keeping agent answers short when reviewing or generating code.
  • Applying one consistent ruleset across several AI coding agents in a team.
  • Compressing verbose MCP server responses in Cursor or Gemini CLI.

Quick start

  1. 1

    Install the plugin with one command

    npx chisle
  2. 2

    Preview the install first (optional)

    npx chisle --dry-run

Things to know

  • Benchmarks are reported for Claude Code and Pi only, with results depending on the specific models and tasks tested.
  • Lazy code rules skip features like abstractions and boilerplate, so you may need to add them back later.
  • Input-side compression covers only some agents; many agents get the ruleset alone or need an MCP proxy.

In collections

Claude Code tools, skills & pluginsOpen-source AI agent frameworks & appsMCP serversAI agent skill libraries

Topics

#ai-agents#token-optimization#claude-code#developer-tools#prompt-engineering#mcp#cost-reduction#agent-skills#anthropic#caveman

FAQ

What is Chisle?
A zero-dependency plugin that trims an AI coding agent's token use by tightening its prose, code style, and oversized tool output. Chisle is a plugin for AI coding agents, mainly Claude Code, that reduces the tokens they consume without changing answer quality. It works on three axes: a ruleset that makes the model write terser prose, a YAGNI-first rule set that favors the smallest working code, and a PostToolUse hook that shrinks oversized tool output before it re-enters the context window.
Is Chisle free to use? What license does it have?
Chisle is open source under the MIT license. Check the license terms before using it in commercial products.
Is Chisle still maintained?
Actively maintained. The last commit to Chisle was on Oct 24, 2026.
What can I use Chisle for?
Reducing token costs for daily Claude Code sessions with long tool output. Keeping agent answers short when reviewing or generating code. Applying one consistent ruleset across several AI coding agents in a team. Compressing verbose MCP server responses in Cursor or Gemini CLI.
How do I install Chisle?
Install the plugin with one command: npx chisle Preview the install first (optional): npx chisle --dry-run
Where can I get Chisle?
The source code and installation instructions are on GitHub: https://github.com/JayPokale/Chisle

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects