Agent Slimmer

Optimize your AGENTS.md to its minimum effective form.

Research shows that bloated context files actively hurt AI coding agent performance. Every unnecessary instruction adds cognitive load, increases inference cost, and can reduce task success rates.

Agent Slimmer is a reusable skill file that any AI coding agent can follow to evaluate and rewrite your AGENTS.md, CLAUDE.md, or similar context file — cutting what doesn't help and keeping what does.

How it works

Point your AI coding agent at the skill file:

Use the skill in slim-agents.md to optimize my AGENTS.md file.

The agent will inventory your repo, classify every block in your context file against research-backed criteria, and produce a slimmed-down version with a changelog explaining what was cut and why.

What gets removed

• Codebase descriptions the agent could discover on its own
• Content that duplicates existing repo documentation
• Generic best practices any competent agent already follows
• Vague guidance that adds load without directing behavior

What stays

• Specific tool and command requirements
• Behavioral constraints not inferable from the code
• Essential project knowledge not documented elsewhere

Research basis

Built on findings from two 2026 studies:

• Lulla et al. — focused context files significantly improve agent efficiency
• Gloaguen et al. — comprehensive files degrade accuracy and increase cost

View on GitHub