Memory and project instructions
Teach the agent how your project works with HARNES.md, and let it remember what matters.
Every new chat starts from scratch. Two tools save the agent from rediscovering everything each time.
The HARNES.md file
If the project folder contains HARNES.md (or AGENTS.md or CLAUDE.md), the agent reads it at the start of every chat. Put in it:
- what the project is and its parts
- key commands: run, test, build
- rules: “always reply in Persian”, “ask me before deploying”
- server addresses and paths (but never passwords)
You don’t have to write it yourself. Say:
Look through this project and write a good HARNES.md for it.
Memory
When you mention something important — “the site runs on port 8080”, “always install with yarn” — the agent can save it to the project memory (.harnes/memory.md) or your personal memory. You can also ask directly:
Remember that our test server is 1.2.3.4.
Long chats
When a chat gets very long, the agent summarises the older parts so it can keep going without the cost growing.