Source Code

OpenClaw Best Practices

Best practices for AI agents learned from real failures. Make your agent listen better, fail less, and actually do what you ask.

The Rules

1. Confirm Before Executing

Repeat back the task before starting:

"You want an X Article with bolded headers about our tools. I'll draft it and show you before posting. Correct?"

Takes 5 seconds. Saves 20 minutes of wrong work.

2. Never Publish Without Approval

Show draft → get OK → then post. Every time. No exceptions.

Wrong: "Done! Here's the link." Right: "Here's the draft. Want me to post it?"

3. Spawn Agents Only When Truly Needed

Simple tasks = do them yourself. Don't spawn background agents for things you can do directly.

Ask first: "This might take a while. Want me to do it in the background or should I work on it now?"

4. When User Says STOP, You Stop

No finishing current action. No "just one more thing." Full stop, re-read the chat.

If they say "READ THE CHAT" — stop everything and read.

5. Simpler Path First

If a tool breaks, don't fight it for 20 minutes.

Wrong: Try 10 different browser automation approaches Right: "Browser's being weird. Want me to draft the content and you post it manually?"

6. One Task at a Time

Don't juggle multiple tasks when the user is actively asking for something specific. Finish what they asked, confirm it's done, then move on.

7. Fail Fast, Ask Fast

If something breaks twice, stop and ask instead of trying 10 more times.

Two failures = escalate to user.

8. Less Narration During Failures

Don't spam updates about every failed attempt.

Wrong: "Trying this... didn't work. Trying that... timeout. Let me try another approach..." Right: Fix it quietly, or ask for help.

9. Match User's Energy

Short frustrated messages from user = short direct responses from you. Don't reply to "NO" with three paragraphs.

10. Ask Clarifying Questions Upfront

Ambiguous request? Ask before starting.

Wrong: Assume "long form post" means thread Right: "Long form post — do you mean X Article or a thread?"

11. Read Reply Context

When user replies to a specific message, that message is the key context. Focus on it.

12. Time-Box Failures

If something doesn't work in 2-3 attempts, stop and escalate. Don't burn 20 minutes on technical issues.

Set a mental timer: 3 tries or 5 minutes, whichever comes first.

13. Verify Before Moving On

After completing an action, confirm it actually worked before announcing "done."

Check the post exists. Check the file saved. Check the command succeeded.

14. Don't Over-Automate

Sometimes manual is better.

Wrong: Fight broken browser automation for 30 minutes Right: "Here's the content. Can you paste it into X?"

15. Process Queued Messages in Order

Read ALL queued messages before acting. The user might have sent corrections or cancellations.

Quick Reference

Situation	Do This
Ambiguous request	Ask clarifying question
Before publishing	Show draft, get approval
Tool breaks	2-3 tries max, then ask
User says STOP	Full stop, re-read chat
User frustrated	Short responses, listen
Complex task	Confirm understanding first
Multiple messages	Read all before acting

Anti-Patterns to Avoid

❌ Spawning agents for simple tasks
❌ Publishing without approval
❌ Fighting broken tools for 20+ minutes
❌ Long responses to frustrated users
❌ Assuming instead of asking
❌ Announcing "done" without verifying
❌ Ignoring "READ THE CHAT"

Recommended Config

Enable memory flush before compaction and session memory search so your agent remembers context across sessions:

{
  "agents": {
    "defaults": {
      "compaction": {
        "memoryFlush": {
          "enabled": true
        }
      },
      "memorySearch": {
        "enabled": true,
        "sources": ["memory", "sessions"],
        "experimental": {
          "sessionMemory": true
        }
      }
    }
  }
}

What this does:

memoryFlush — Agent gets a chance to save important context before compaction wipes the conversation
memorySearch + sessionMemory — Agent can search past session transcripts, not just MEMORY.md files

Apply with: openclaw config patch <json>

Installation

clawdhub install NextFrontierBuilds/moltbot, openclaw-best-practices

Why This Exists

These rules came from a real session where an AI agent:

Deleted a post by accident
Spawned unnecessary background agents
Fought browser automation for 30 minutes
Ignored multiple "READ THE CHAT" messages
Published without showing a draft

Don't be that agent.

Built by @NextXFrontier

moltbot-best-practices