Day 7 of 28 · OpenClaw Challenge
What Just Happened?
⏱ 7 min
📊 Beginner
Yesterday you had your first real conversation with OpenClaw. It searched the web, summarized articles, and created files on your computer.
But how do you know what it actually did? How do you trust that it didn't make things up, skip steps, or access something it shouldn't have?
Today you'll learn to read the agent's logs — a complete record of every thought and action it took.
Why logs matter
Every time OpenClaw does something, it writes it down. Every tool it called. Every decision it made. Every result it got back.
This isn't just for developers. This is how you build trust with your agent. You can always look behind the curtain and see exactly what happened.
Think of it like a receipt. When a contractor finishes work on your house, you want to see what they did, what materials they used, and how long it took. Agent logs are your receipt.
Knowledge Check
Why should you care about agent logs?
A
They let you verify exactly what the agent did — building trust and transparency
B
They make the agent run faster
C
They're required by law
D
They're only useful for developers debugging code
Logs are your window into the agent's actions. You don't have to read them every time, but knowing you can is what makes it safe to let the agent act on your behalf.
Reading a log
Let's look at what happened when you asked OpenClaw to check the weather yesterday. Here's a simplified version of the log:
🔍 **OBSERVE**: User asked "What's the weather in San Francisco?"
🧠 **THINK**: I need current weather data. Best tool: web_search. Query: "current weather San Francisco." No user preferences conflict with this action.
⚡ **ACT**: Calling web_search("current weather San Francisco")
→ Result: 62°F, partly cloudy, wind 12mph W, humidity 68%, sunset 6:14 PM
📤 **REPORT**: Formatting weather data for user. Presenting temperature, conditions, wind, humidity, sunset time. Offering to show weekly forecast.
↻ Replay conversation
See how clean that is? Four entries, matching the four steps of the agent loop. You can trace exactly what happened:
The agent understood your question (Observe). It decided to use the web search tool (Think). It searched and got real results (Act). It formatted those results for you (Report).
No mystery. No black box.
Knowledge Check
In the log, the agent wrote "No user preferences conflict with this action." What does that mean?
A
The agent checked if you have a weather app preference
B
The agent asked for your permission silently
C
It's a meaningless system message
D
The agent verified that nothing in your custom instructions would prevent or modify this action
Before acting, the agent checks your preferences and custom instructions. If you had a rule like "never use web search" or "always ask before accessing the internet," it would have adapted. This check is logged so you can see the agent is respecting your rules.
When something goes wrong
Logs become even more valuable when things don't work. Instead of the agent just saying "sorry, I couldn't do that," you can see why:
🔍 **OBSERVE**: User asked "Send the report to Sarah."
🧠 **THINK**: I need the email tool. Recipient: Sarah. But which Sarah? User has 3 contacts named Sarah. I should ask for clarification.
📤 **REPORT**: Asking user to specify which Sarah.
↻ Replay conversation
I found 3 contacts named Sarah:
1. Sarah Chen (Engineering)
2. Sarah Miller (Marketing)
3. Sarah Lopez (Client)
Which Sarah should I send the report to?
Sent ✓ — Report delivered to Sarah Chen at s.chen@company.com
↻ Replay conversation
Knowledge Check
Why did the agent ask for clarification instead of just picking a Sarah?
A
Because it can only handle one contact at a time
B
Because it doesn't have access to your contacts
C
Because it found multiple matches and chose to confirm rather than risk sending to the wrong person
D
Because it's programmed to always ask before sending email
The agent made a judgment call — sending a report to the wrong person could be a real problem. Instead of guessing, it asked. This is the kind of thoughtful behavior you want from an agent handling your real work.
Week 1 — what you've learned
Let's take a step back. In seven days, you've gone from zero to running your own AI agent:
Day 1: Chatbots talk, agents do
Day 2: OpenClaw is open-source, runs locally, connects to anything
Day 3: The agent loop — Observe, Think, Act, Report
Day 4: Tools give the agent hands — email, calendar, web, files, code
Day 5: You installed OpenClaw on your own machine
Day 6: Your first real conversation — the agent searched, summarized, and created files
Day 7: You learned to read the agent's logs and understand its reasoning
Next week, we connect the agent to your real world — email, calendar, memory, and more. That's where the magic really starts.
Final Check
What was the purpose of this first week?
A
To become an expert in AI agents
B
To build a production-ready AI system
C
To understand the fundamentals — what agents are, how they think, and how to run one yourself
D
To learn programming and computer science
This week was about building a foundation. You now understand what AI agents are, how they work (the agent loop), what makes them useful (tools), and you've installed and used one yourself. Next week, you'll start connecting it to your real tools and workflows.
🏁
Day 7 Complete — Week 1 Done!
"Foundation built. Next week, we connect your agent to the real world."
Tomorrow — Day 8
Connecting to Email
Week 2 begins — let's connect OpenClaw to your inbox and watch it handle real email.