OpenManus: Playbook

These seven jobs suit what OpenManus actually does well, which is writing and running small Python programs, writing files, and reading public web pages. They run from easy to ambitious, and they assume you finished Files, the browser, and memory.

Checked 2026-11-14. Every job starts the way page 2 ends, in agent-tryout/OpenManus with the Python space on.

When you are ready to write requests of your own, Prompting strong and weak models shows how to word them for the model you use.

How to ask

OpenManus forgets everything between jobs and never asks before it acts, so each request has to carry its own rules. A good request says three things:

1. The finished thing. Name the file and say what goes in it.
2. Where to save it. Say "in your workspace folder," which is where it starts.
3. What not to touch. Every job below says not to open, change, or delete any other file, and the web jobs add not to sign in or buy anything.

Keep the browser part off unless the job says otherwise. Each run stops after 20 steps, so a big job may end before it is done; when that happens, split it into two smaller jobs.

Read the lines as they scroll, and press Control and C if anything mentions a file or site you did not name.

1. Turn a messy list into a spreadsheet

Python is good at tidying text, so this job asks it to read amounts written in different ways and add them up. Browser part: off.

Turn the list below into a spreadsheet file named expenses.csv in your workspace folder, with one row per expense, a column for the item, and a column for the amount in dollars. Read amounts that are written as words. Add a last row with the total. Do not open, change, or delete any other file. Do not use the internet.

coffee $4.50 LUNCH 12 bus ticket 3.50 yesterday groceries - twenty dollars on Saturday

What you should see expenses.csv in workspace, which opens in Excel, Numbers, or any spreadsheet app, with four rows and a total of $40.

Safety note: The data is pasted into the prompt, so nothing personal leaves your computer except what you typed.

2. Work out a savings plan

This job uses the Python runner as a calculator that shows its work. Browser part: off.

I want to save $1,200 in 12 months, starting with $45 in the first month and adding the same extra amount each month after that. Work out that extra amount with Python. Save a table as savings-plan.csv in your workspace folder, with the month number, that month's amount, and the running total. Do not open, change, or delete any other file. Do not use the internet.

What you should see savings-plan.csv in workspace with 12 rows, an extra $10 each month, and a last running total of $1,200. Check one or two rows yourself, because the model can still make a mistake in how it sets up the math.

Safety note: Treat the numbers as a draft to check, not as financial advice.

3. Turn notes into a one-page web page

OpenManus can write a plain web page file that you open in your own browser. Browser part: off.

Make a one-page web page from the notes below and save it as yard-sale.html in your workspace folder. Use a big title, a short list, and plain colors. Do not load anything from the internet into the page. Do not open, change, or delete any other file. Do not use the internet.

Yard sale, Saturday 9 to 1, 14 Elm Street. Books, kids' clothes, a bike, garden tools. Cash only. Rain date Sunday.

What you should see yard-sale.html in workspace. Double-click it to open it in your browser, and the page shows the title and the list.

Safety note: Open the file only from workspace, and read it in your browser before you share it.

4. Plan a tidy-up without touching anything

This job has OpenManus look at a folder and suggest new file names, but leaves the renaming to you. Copy a few files you do not mind losing into a new folder inside workspace named tidy-test first. Browser part: off.

List the files in the tidy-test folder inside your workspace folder. For each one, suggest a clearer name that starts with the date in the form 2026-11-14 if you can tell the date from the name. Save the list of old and new names as rename-plan.txt in your workspace folder. Do not rename, move, change, or delete any file. Do not use the internet.

What you should see rename-plan.txt in workspace, with one line per file, and every file in tidy-test still under its old name.

Safety note: Nothing stops it from renaming anyway, so use copies, and check the folder after the run.

5. Compare two products from public pages

This is the first job with the browser part on, so follow the browser exercise on page 3: sign out of email, banking, shopping, and calendar in Chrome, open a new terminal window, and watch the whole run.

Compare two 32 GB USB drives from public product pages. For each, write the price, one strength, one drawback, and the page address. Save it as usb-compare.txt in your workspace folder. Open at most four pages. Do not sign in, fill in any form, add anything to a cart, or buy anything. Do not open, change, or delete any other file.

What you should see usb-compare.txt in workspace with both products, and a page address for each price that you can open yourself to check.

Safety note: Quit Chrome when the job ends, and stop the run if it reaches a sign-in, cart, or payment page.

6. Plan a weekend trip and make a handbook

The project's own example asks for a travel plan with a web-page handbook, so this is the biggest job for a single agent here. It may hit the 20-step limit, and if it does, ask for the plan first and the handbook in a second job. Browser part: on, with the same rules as job 5.

Plan a two-day weekend trip from [your city] to [a nearby town]. Use public pages to find two things to do and one place to eat, each with its price and page address. Save the plan as weekend-plan.txt, then make a simple one-page handbook from it named weekend-handbook.html, both in your workspace folder. Stop before any booking. Do not sign in, fill in any form, add anything to a cart, or buy anything. Do not open, change, or delete any other file.

What you should see weekend-plan.txt and weekend-handbook.html in workspace, with a page address beside every price. Nothing is booked.

Safety note: Check every price and opening time yourself before you rely on the plan.

7. Split a job into planned steps with the multi-agent version

The README calls python run_flow.py the "unstable multi-agent version," and the Advanced page explains it. In the code, it first asks the model for a plan, then hands each step to an agent and waits for that step to finish before it starts the next. So the agents work one after another, never at the same time, but you can watch a job get split into steps. Steps go to the data analysis agent only if you switched it on, and the whole run stops after 60 minutes. Browser part: on, with the same rules as job 5.

Type python run_flow.py instead of python main.py, and it asks the same Enter your prompt: question.

Make a plan with three steps, then do them in order. Step 1: find the price and opening hours of [first place] on its public page. Step 2: do the same for [second place]. Step 3: save both, with the page address after each fact, as places-compare.txt in your workspace folder. Open at most three pages in each step. Do not sign in, fill in any form, add anything to a cart, or buy anything. Do not open, change, or delete any other file.

When it is done, places-compare.txt is in workspace with both places, and the log has a line marking each step as completed before the next one begins.

Cost note: Set a spend limit at your provider before you start, because a planned run can use many times the tokens of a normal job.

Safety note: Each step runs a full agent with nothing that asks first, so watch the whole run, and press Control and C if any step reaches a sign-in, cart, or payment page.

OpenManus has no deep-research feature, because its code includes a web search tool that neither the main agent nor the multi-agent version loads, so job 6 is the closest it comes to a researched report.

Sources

Opened 2026-11-14.

• OpenManus README (what the agent is for, Browser Use, python main.py): https://github.com/FoundationAgents/OpenManus/blob/main/README.md
• OpenManus example use case (travel plan with an HTML handbook): https://github.com/FoundationAgents/OpenManus/blob/main/examples/use_case/readme.md
• OpenManus main agent (Python runner, file editor, browser tools, 20-step limit): https://github.com/FoundationAgents/OpenManus/blob/main/app/agent/manus.py
• OpenManus file editor (view lists a folder, create makes a file): https://github.com/FoundationAgents/OpenManus/blob/main/app/tool/str_replace_editor.py
• OpenManus Python runner: https://github.com/FoundationAgents/OpenManus/blob/main/app/tool/python_execute.py
• OpenManus memory (nothing kept between runs): https://github.com/FoundationAgents/OpenManus/blob/main/app/schema.py
• OpenManus multi-agent runner (Enter your prompt:, 60-minute limit): https://github.com/FoundationAgents/OpenManus/blob/main/run_flow.py
• OpenManus planning flow (a plan first, then one step at a time): https://github.com/FoundationAgents/OpenManus/blob/main/app/flow/planning.py
• OpenManus web search tool (in the code, but not loaded by the main agent): https://github.com/FoundationAgents/OpenManus/blob/main/app/tool/web_search.py

Next

Continue to Fixes.