Stop Claude agents from wasting tokens with pointless terminal checks
Claude AI agents sometimes run terminal commands repeatedly or check things they already know, burning through tokens and slowing work down. Adding rules to a CLAUDE.md file can curb this behavior.
When an AI agent like Claude works autonomously on a coding task, it can fall into a habit of firing off terminal commands more than necessary — re-checking file paths, pinging the system, or confirming information it already has. Each of these actions consumes tokens, which directly translates to higher API costs and slower task completion.
CLAUDE.md is a plain text file you place in your project folder; Claude reads it at the start of each session and follows the instructions inside. By writing explicit rules such as 'do not re-check files you have already read' or 'avoid unnecessary terminal pings', you train the agent to skip redundant steps. This is a lightweight, free way to make Claude cheaper and faster on any project.
Key points
- Claude agents can waste tokens by running unnecessary terminal commands repeatedly
- CLAUDE.md is a text file in your project that gives Claude standing instructions
- Add rules like 'don't re-confirm already-known info' to reduce redundant actions
- Example rules: 'no repeated file-existence checks', 'skip echo/ping unless required'
- Result: lower API costs and faster task execution
Quick term guide
- AI agents
- AI agents are AI tools that can carry out steps toward a goal, not just answer once.
- AI agent
- An AI program that can inspect information and suggest what to do next.
- terminal
- A text-based way to use a computer by typing commands.
- CLAUDE.md
- A project instruction file that tells Claude Code how to work with the codebase.
- autonomous
- The ability of an AI to complete tasks or make decisions without constant human guidance.
- API costs
- Fees paid when software calls an online service programmatically.
- session
- A continuous period of interaction between a user and a computer program.
- terminal ping
- A short command an agent runs just to check system status, often unnecessarily