Genspark Super Agent Playbook: Modes, Credits, Skills, Workflows

Genspark's Super Agent takes one plain-language request, plans the steps, runs them on a cloud computer and hands back files: a report, a deck, a spreadsheet, a web page or images. A long job can run for up to 2 hours.^[1] Cost is the part to plan around. A Standard message typically costs 10 to 60 credits and an Ultra message 50 to 500, and a text job shows its cost only after the turn ends. Super Agent cannot run anything on a schedule. Recurring work belongs to Workflows, reusable instructions to Skills and Custom Agents, and shared projects to a Hub. Use it for one-off deliverables you will check before they leave your hands.

New or Classic: pick the mode before you type

Super Agent runs in two modes, and you choose before the chat starts. New Super Agent gives each project its own cloud computer; Classic is the original mode with no cloud computer. New is the default on paid plans, while most free accounts start in Classic and can switch New on from the plus menu. Skills do not exist in Classic, attachments there stop at 100 MB a file, and continuing an old Classic chat in New is open only to paid accounts on the web.

The project computer is temporary. It is a per-project sandbox, separate from the Cloud Computer subscription that Genspark Claw needs, and once Genspark recycles it the conversation cannot continue. Delivered files stay, while scratch files are lost. Treat every project as a session with an end date. Once you connect sources to SecondBrain, Super Agent can also act on that data, from triaging email to querying HubSpot or Salesforce records and drafting weekly reports; our Genspark SecondBrain guide explains which sources it reads.

What each message costs, and when you find out

Credits measure the work behind a turn. Super Agent defaults to Standard, which typically runs 10 to 60 credits a message; Ultra typically runs 50 to 500. Long chats, retries and big files land above those ranges, because every message in a long chat re-processes the whole context and each failed agent retry is billed again. Genspark compresses very long chats on its own, and Super Agent offers a manual Compress chat option. Use it before a thread grows expensive.

The bill arrives at two different moments. New images, video and music are confirm-first: the agent shows a form or price card with the model and its price before it generates anything. A text job is different. Its cost appears only after the turn, in your credit history, and that log can lag a few hours. New Super Agent also checks your balance before a turn and refuses one it cannot cover. A turn that runs dry midway stops and keeps the files made so far.

Free Quota changes this on personal plans only. Plus and Pro get a weekly balance for Standard-mode work that covers Super Agent turns at no credit cost, on a rolling 7-day window that starts with your first covered use; Genspark says Pro's balance is ten times Plus's. Free accounts and Team or Enterprise seats have none, so every turn there bills credits. When the quota runs out, turns move to credits automatically. Ultra, custom models, video and music always bill credits.

When a run stalls, type continue

Long jobs hit limits, and the fix is usually one word. When a turn reaches its time limit, type continue in the same chat and Super Agent resumes from the files it saved. A question the agent asks and nobody answers for 10 minutes ends the turn, so stay near the screen for the opening minutes of a big job. If the project computer has already been recycled, continuing is no longer possible, and you start a new project instead.

Watch what you are handed, too. Decks come out of the viewer as PPTX or PDF, and Super Agent produces real docx and xlsx files when you ask. A web page arrives as a single self-contained HTML file that you must host somewhere else. Links the agent writes into the chat can stop working after a couple of hours, while the file itself does not expire. Save the file, not the link.

Skills and Custom Agents save you from re-prompting

A Skill is a saved workflow that applies the same structure, tone and reasoning across Super Agent, AI Slides and AI Docs. The library holds 100+ community Skills from publishers including Anthropic, OpenAI and Genspark, and a free account is enough to use them.^[2] You can build your own with an AI-guided creator or upload a .zip, .skill or .md file of up to 200 MB. Some Skills will not run until you connect a Connector such as Gmail or Slack.

Sharing has expiry dates. An email invite lasts 30 days, and a public link is also good for 30 days but installs a snapshot that never receives your later edits. Publishing a Skill to the whole organization needs Team or Enterprise, with admin review. Your library is full at 80 Skills, so prune the ones nobody runs before you reach that ceiling.

A Custom Agent is narrower. You describe it in one prompt, test it in a Preview window and call it with @ inside a Super Agent chat. You can keep it private, share it by link or list it in the Custom Agent Store; sharing with your organization needs Team or Enterprise. Sharing passes only the name and description, and each user's conversations stay isolated. Deleting one is permanent. Unlike New Super Agent, a Custom Agent runs no balance check before a turn.

Workflows and Hubs carry what Super Agent cannot

Super Agent cannot run scheduled jobs. Recurring tasks belong to Workflows, which you describe in plain language and which run on a schedule or when a Gmail or Outlook email arrives.^[3] Each workflow supports one trigger, so a morning report and an inbox reaction become two workflows. Integrations include Gmail, Slack, Microsoft Teams, Salesforce, Pipedrive, Stripe, GitHub and Custom Agents. The Workflows builder always uses credits, even on plans with Free Quota.

A Test Run is a real run. Emails go out, data changes and credits are spent, so point the first version at a test inbox or a copy of the sheet. A GenTeam step posts as the workflow owner and pauses for approval while Approvals is on. Claw is the other route for recurring work, with schedules and Heartbeat check-ins that each draw credits; cloud mode needs a separate Cloud Computer subscription, covered in our Genspark Claw guide.

A Hub is the shared folder for a project team. It holds up to 100 files of up to 10 MB each, a limit Genspark sets to save credits, and takes no audio, video or compressed files. Members see every file and the project history and can continue any project, while only the creator invites or removes people. Super Agent is the default agent inside. The owner cannot leave a Hub, only delete it, which erases everything.

Where a person must stay in the loop

Genspark builds several stops into its agents and leaves a few gaps. The Terms set the baseline: users cannot rely on generated content.^[4] Before you hand an agent anything that leaves your company, know which checkpoint applies to it and which step has none, because the agents behave differently. The widest gap sits in GenMail, where a Skill set to auto-run can send mail without a fresh confirmation each time. The stops look like this:

• Media generation shows a price card before images, video or music are made.
• Workflow Approvals, on by default in AgentBase workflows, wait for a person before sensitive sends; the run fails if nobody approves.
• GenTeam cloud agents ask before deleting, spending money or sending something outside; see our GenTeam guide.
• Claw needs permission for external actions such as sending email or posting.
• Chat-app agents are instructed to confirm before actions that affect others, send-type tools such as email, SMS and calls run only on turns you started, and there is no single undo.
• The Chrome extension clicks buttons, fills forms and extracts data for you; Genspark advises testing automation on a non-critical page first and confirming intent before sensitive actions such as submitting orders or sending emails.

Reviewers describe where it goes wrong. Complaint themes include credits spent on failed or unwanted outputs and on regenerations, with no running cost meter, and many credits consumed while iterating on slide formatting. Some reviewers call it weaker for creative work than for office work. Runs that fail on a system error get credits back, but quality disputes get no automatic refund. A precise first prompt costs less than a string of regenerations.

The first job to hand Super Agent

Start with one deliverable you would otherwise build by hand this week: a research brief, a short deck, a spreadsheet pulled from scattered notes. Run it in Standard mode, where the typical message range is lowest and Free Quota applies on personal Plus and Pro. Ask for the files early, save them rather than the chat links, and read every claim before it travels, since the Terms leave that responsibility with you. When the same request comes back a third time, turn it into a Skill. When it needs a clock, move it to Workflows.