Skip to main content

Tips, Tricks & Troubleshooting

You are almost done with the tutorial — congrats. If you made it this far, you already understand the Taurus model much better than most people do on first contact.

Memory matters a lot​

If you only keep one power-user habit from this manual, keep this one: pay attention to MEMORY.md.

When an agent keeps forgetting a durable fact, a standing rule, or the shape of an ongoing project, MEMORY.md is usually where that belongs.

If MEMORY.md gets big, tell the agent to refactor or condense it. Taurus loads the beginning first, so once the file grows enough, the tail matters less and can eventually get cut off from what the agent sees at run start.

A few habits that pay off fast​

  • Start a new run when the task changes a lot.
  • Ask for a plan first when the task is messy or high-risk.
  • Stop the agent early if it is obviously going the wrong way.
  • Make specialized agents instead of turning one agent into a giant everything-bot.
  • Use dashboards and scheduled agents when you want recurring outputs instead of repeating yourself.

When something feels off​

Check these first:

  1. Is the agent actually running, or is it waiting on you?
  2. Did you give it the right tools, model, and access?
  3. Did the run simply get too bloated and need a new run or compaction?
  4. Should that missing instruction really live in MEMORY.md instead?

A surprising number of Taurus problems are just one of those four.

Final note​

You do not have to use every advanced feature immediately. Taurus gets more fun as you grow into it.

Start simple, notice what repeats, and then teach your agent tree better habits over time.