Manus Tutorial: From Your First Task to Saving Workflows with Agent Skills for Reuse

At 1:00 AM, an e-commerce founder friend in Taiwan sent me a message saying he had spent the entire weekend organizing a price and specification sheet for 50 competing products, practically wearing out his Excel skills. I asked him why he didn't just hand it over to an AI, and he replied, "I tried, but ChatGPT just gives me a bunch of text, and I still have to copy and paste it back into a table myself."

This complaint is quite typical and neatly illustrates the reason tools like Manus exist.

What is Manus?

Manus is an AI agent platform. The fundamental difference between it and a chatbot is simple: it takes action.

The official documentation makes its execution environment very clear—skills run in a secure cloud virtual machine with a complete Ubuntu file system, the ability to execute shell commands, and support for native scripts. In practice, this means it can open a web browser to click through web pages, write and run a Python script to check the results, and save the output as an Excel or presentation file for you to download.

You are delegating a "goal," not just "a conversation."

What Can It Do For You?

• Multi-step data collection: For example, "Find the store counts and pricing for the top 30 chain coffee shops in Taiwan and organize them into a table."
• Web automation: Logging in, filling out forms, scraping, and comparing data—actions that originally required manual clicking.
• Code and file generation: Writing scripts, running analyses, and outputting .xlsx, .pptx, or .pdf files.
• Scheduled tasks: Running automatically at fixed times, such as updating competitor prices every Monday morning.
• Turning completed tasks into skills: Re-running the exact same workflow next time with a single sentence.

Plans and Credits: Understand Before You Begin

This is the most crucial step to figure out first, because your credit balance will dictate how you prompt the tool. According to data from the official help center:

Plan	Price	Credits	Concurrent Tasks	Scheduled Tasks	Models
Free	$0	300 credits/day	1	2	Chat Mode, Manus 1.6 Lite in Agent Mode
Pro Tier 1	From $20/month (~17% off annually)	From 4,000 credits/month	20	20	Plus Manus 1.6, 1.6 Max
Pro Tier 2	From $40/month (includes 7-day trial)	From 8,000 credits/month	20	20	Same as above
Team	From $20/user/month	Shared credit pool	—	—	Also includes SSO, data training opt-out, team usage analytics

The official notes also state that existing members will continue to renew automatically at their original prices.

Note that the free version provides credits daily (300 credits), while paid versions grant them monthly—a distinction that impacts your usage rhythm. The free version is great for doing one small task every day, whereas the paid version is suited for concentrated bursts of work on specific days.

How to Use It: Six Steps

Step 1: Register and Select Your Mode

After registering, you will see two options: Chat Mode and Agent Mode. Chat Mode is for standard conversation, while Agent Mode is the one that actually gets things done. Choose your agent before starting a new task to avoid wasting credits in the wrong mode from the start.

Step 2: Clearly Articulate Your Task (The Most Important Step)

The most common mistake beginners make is giving vague prompts. Saying something like "Help me research the electric vehicle market" will cause Manus to wander aimlessly searching for data, burning through a bunch of credits and returning results you don't even want.

A good task description includes four elements:

1. Goal: The specific result you are looking for.
2. Scope: Which subjects, what time frame, which region.
3. Output format: Table? Presentation? Markdown document? What specific columns?
4. Constraints: What to exclude, and requirements for data sources.

Let's compare these two side-by-side:

❌ Help me look at which companies in Taiwan are doing AI.

✅ Find AI startups in Taiwan founded after 2024 with public fundraising records, and organize them into a table. Columns must include: Company Name, Founding Year, Main Products, Most Recent Funding Amount and Date, and Official Website. Only include entries with verifiable official websites, and please attach links for the data sources.

Based on my own tests, the second approach consumes about half the credits of the first, and the results are ready to use immediately.

Step 3: Watch It Work and Interrupt When Necessary

When Manus is executing a task, it shows you what it is doing—which web page it opened, what command it ran. This isn't just for show; it's your brake pedal.

If you notice it stubbornly searching for the exact same keyword in a loop by the third minute, stop it immediately and provide clarification. Doing this saves far more credits than letting it finish and starting over. My habit is to watch closely for the first two minutes to confirm it's heading in the right direction before moving on to other work.

Step 4: Use "Wide Research" for High-Volume Tasks

When faced with tasks involving high homogeneity but large scale, such as "analyze 100 products" or "research 50 companies," Manus features a mode called Wide Research.

Its approach works like this: the main agent breaks the request down into independent sub-tasks, assigns each sub-task to an agent with an isolated context window, runs them all in parallel, and then has the main agent assemble the results into your desired format.

This design solves the physical limitations of context windows—data for 50 companies cannot fit into a single conversation, but splitting it into 50 independent micro-tasks solves the problem. According to official documentation, this feature triggers automatically at appropriate moments, and from Manus 1.6 onward, all sub-agents run on the Max architecture.

Ideal use cases: competitor audits, market research, literature reviews, lead generation, and generating high-volume similar content.

Step 5: Save Completed Work as Agent Skills

This is, in my opinion, the most underrated feature.

Agent Skills are reusable custom workflows. The creation method described in official documentation is so simple it feels anticlimactic:

1. Complete a task you want to repeat using Manus.
2. Ask it to package the process into a skill; Manus will automatically generate the documentation.
3. Trigger it directly the next time you need it.

No coding is required—you can build them with Markdown instructions, and only advanced automations require scripts. Skills use progressive loading: metadata is loaded first (around 100 tokens), instructions are triggered on demand (under 5k tokens), and resources are only loaded when utilized, meaning storing a mountain of skills won't slow things down.

A practical example: If you need to produce a monthly "Competitor New Features Summary," finish it once and save it as a skill. Afterward, a single sentence each month—"Run the competitor monthly report"—finishes the job.

Step 6: Set Schedules to Run Autonomously

The free version allows you to set 2 scheduled tasks, while Pro allows 20. Offload routine work like "update keyword rankings every Monday morning" or "run the competitor monthly report on the 1st of every month" into scheduled tasks so you don't even have to think about them.

Advanced Tips: Three Habits to Save Credits

1. Ask before you act. For complex tasks, use Chat Mode first to ask, "How do you plan to break down this task?" Once you confirm its planning direction is correct, switch to Agent Mode to execute. Credit consumption during the planning phase is far lower than during execution.

2. Lock down the output format upfront. If you don't specify a format, it will decide for itself, and asking for revisions means running the whole thing over again. Specify what columns the table needs, whether links should be included, and what file format you want—all in your initial prompt.

3. Split large tasks into two stages. Instead of demanding it "find data and make a presentation" all at once, first ask for the data table. Once you confirm the data is correct, ask it to build the presentation. If an error occurs in the middle, you only need to re-run the second half.

Cautions: Three Points Taiwanese Users Should Note Especially

The risk of data leaks is real. When Manus executes tasks, it sends the data you provide into a cloud sandbox for processing. If you ask it to organize customer lists, financial statements, or unreleased product plans, that data leaves your machine. The Team plan includes a "data training opt-out" option, but that doesn't mean your data never passes through their systems. For tasks involving business secrets or personal information, verify your company's security policies first—under Taiwan's Personal Data Protection Act, this isn't something you can brush off with a "it's probably fine."

It makes mistakes, and it makes them with absolute confidence. During my tests, I encountered instances where it grabbed the wrong fundraising amount for a company. The numbers looked entirely reasonable, but clicking through to the source revealed no such figure. Therefore, your task description must require source links to be attached, and you should spot-check three to five entries before delivery. This is basic discipline when using any AI agent.

Prices and model codenames change. Products in this space evolve at lightning speed, and this article is based on data verified in August 2026. Before deciding to subscribe, check the official pricing page directly.

When NOT to Use Manus

To be completely honest, using Manus is actually slower in the following scenarios:

• Simple Q&A: Use ChatGPT or Claude directly; it's much faster and doesn't require waiting for a sandbox to spin up.
• Programming that requires precise control: Use editor-tier tools like Cursor or Zed where you can review code line by line.
• Automation with fixed rules: If the workflow is completely deterministic, workflow tools like n8n are much more stable and cheaper.
• Highly sensitive data processing: Just don't.

To gain a more complete understanding of the AI agent category, you can check out our previously published AI Agent Build Guide and n8n Automation Tutorial. If you want to compare and switch prices across multiple models, the OpenRouter Tutorial is also worth a read. To find more similar tools, browse the AI Automation Category or check out the Task-Oriented Tool List.

TheAI Academy Review

After using it for a period of time, my verdict on Manus is this: it turns "AI helps me do things" from a marketing slogan into something verifiable. You delegate a task, it hands you a file, and whether the result is good or bad is immediately obvious—this kind of clarity is actually quite rare among AI tools.

However, don't think of it as a panacea. Its strongest scenarios are work that is "highly repetitive, multi-step, but with straightforward individual steps," such as organizing data for 50 companies. When it comes to things that genuinely require judgment and creativity, it still falls short.

Review: What makes Manus valuable isn't how smart it is, but that it actually finishes the job. For the sort of sorting and compiling work in Taiwanese SMEs that people "always knew they should do but never had time for," this is the real solution—provided you first learn how to articulate your instructions clearly.

Practical advice for Taiwanese readers: Start with the 300 daily credits on the free version. Pick a routine task you have to repeat every month (competitor prices, industry news roundups, or customer list updates), complete it once, and save it as an Agent Skill. Run it for a full month, and you'll naturally know whether it's worth upgrading to Pro.

Sources

• Manus Official Help Center: Membership Plans and Pricing
• Manus Official Documentation: Agent Skills
• Manus Official Documentation: Wide Research

Compiled based on public information; official sources prevail. Prices and features are subject to change at any time, so please check the official page before subscribing.