What information does the skill remove?

It removes 'discoverable' information such as tech stack descriptions (found in package.json), directory maps, and product descriptions already present in README files.

Will I lose my custom build commands?

No. The skill specifically keeps build/run commands, non-obvious naming conventions, and environment variable requirements as these cannot be easily inferred from the codebase.

Does it delete my instructions automatically?

No. The skill generates a report of proposed removals with one-line justifications and token saving estimates, requiring your approval before applying changes.

Why should I trim my CLAUDE.md file?

Research indicates that LLM-generated context files can reduce task success by ~3% because agents treat every instruction as a constraint. Trimming ensures the agent only sees what it can't find itself.

Context File Trimmer

Name: Context File Trimmer
Author: denisraison

bydenisraison

•

生産性とワークフロー

Optimizes CLAUDE.md and AGENTS.md files to reduce token costs and improve agent task success through research-backed pruning.

The Context File Trimmer applies empirical research-backed pruning to project context files, specifically targeting the 'less is more' approach to agentic coding. By identifying and removing redundant information that Claude can autonomously discover—such as tech stacks, directory structures, and project descriptions found in READMEs—this skill reduces inference costs by over 20% and prevents the AI from becoming over-constrained by unnecessary instructions. It provides a detailed audit and reporting phase before applying changes, ensuring only non-obvious constraints, specific commands, and essential environment variables remain in your context files.

主な機能

01Cross-file duplication detection between local and global configurations

02Preservation of non-obvious tooling quirks and build commands

032 GitHub stars

04Research-based classification of redundant vs. essential information

05Automated discovery of project-level and global context files

06Pre-execution reports with token saving estimates and justifications

ユースケース

01Auditing instruction files for bloat after significant project evolution

02Reducing LLM inference costs by trimming bloated CLAUDE.md files

03Improving agent reasoning by removing redundant project constraints

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add denisraison/claude-plugins trim-context

For use in Claude.ai and ChatGPT

主な機能

01Cross-file duplication detection between local and global configurations

02Preservation of non-obvious tooling quirks and build commands

032 GitHub stars

04Research-based classification of redundant vs. essential information

05Automated discovery of project-level and global context files

06Pre-execution reports with token saving estimates and justifications

ユースケース

01Auditing instruction files for bloat after significant project evolution

02Reducing LLM inference costs by trimming bloated CLAUDE.md files

03Improving agent reasoning by removing redundant project constraints