Definition: Custom agent commands are saved prompts you trigger with a slash (/) so an AI agent runs the same task the same way every time. Instead of retyping instructions, you type /summarize or /draft-reply and the agent fires a reusable command you defined once. They turn your best prompts into one-keystroke actions across every project.
TL;DR: Custom commands save your best prompts as
/shortcutsan AI agent runs on demand. Define once, trigger anywhere, get consistent output. Past three commands per agent they auto-promote to Agent Skills loaded on demand. Built into every Taskade agent. Build an agent free →
You are already doing a version of this. You have a "magic prompt" pasted in a notes file, a reply template you copy every Monday, or a checklist you repeat to a chatbot. A custom command is that habit, made permanent and shareable. You define the prompt once, and the agent runs it identically for you and your whole team.
What Are Custom Agent Commands?
Custom commands are saved prompts triggered by / + a name you choose, so an AI agent performs a defined task on demand instead of improvising. They range from a one-line text transform (/shorten) to a multi-step analysis the agent plans and works through. Because the instructions live in the command, the output stays consistent across people, projects, and days.
Each command runs in one of two modes:
- Default mode for a direct, single-pass response (rewrite, summarize, classify, extract).
- Plan & Execute mode for a structured task where the agent sets goals and works through them step by step.
Authoring a command takes about a minute: open an agent, go to the Commands tab, click New command, name it, write the prompt behavior, pick the mode, and save. From then on it shows up in the slash menu.
How a Custom Command Runs
A custom command turns one keystroke into a repeatable action. You type the slash trigger, the agent loads your saved prompt, runs it against the current task or selection, and returns output in the same shape every time. The flow below shows the path from definition to instant result.
The payoff is consistency. A saved /qualify-lead command scores every prospect on the same five questions, so the answer does not drift based on how you phrased the request that morning.
Built-In Tools vs Custom Commands
Built-in tools are the 34 capabilities every Taskade agent ships with (web search, code execution, file analysis, image generation, and more). Custom commands are the shortcuts you add on top, made of your own prompts. Tools are the agent's hands; custom commands are the saved plays you call. The table maps the difference.
| Aspect | Built-in tools | Custom commands |
|---|---|---|
| Origin | Ship with every agent (34 total) | You author them |
| What they do | Capabilities (search, code, files, images) | Saved prompts that orchestrate those capabilities |
| Trigger | Agent calls them automatically when needed | You trigger with /name on demand |
| Best for | General reasoning and research | Repeated, team-specific tasks |
| Consistency | Varies with the request | Identical output every run |
| Scaling | Always available | Past 3 per agent, auto-promote to Agent Skills |
A custom command often uses built-in tools under the hood. A /research-competitor command can call web search, read the results, and write a structured brief, all from one keystroke. You get the power of the 34 built-in tools packaged as a one-word shortcut.
What Custom Commands Do in Taskade
Custom commands plug into the whole workspace, not a single chat. Once saved, a command works on selected tasks, drives automation workflows, and travels with the agent when you share it. The capabilities below cover the common ways teams put them to work.
- Slash activation. Type
/+ the command name to fire it instantly in any project. - Two modes. Default for fast single-pass output, Plan & Execute for structured multi-step work.
- Bulk operations. Apply one command to many tasks or paragraphs at once with bulk commands.
- Automation integration. Run commands inside reliable automation workflows so they fire on a schedule or a trigger.
- Team sharing. Share an agent and its commands so everyone gets the same assistant.
- Multi-agent coordination. Give different agents different command sets and run them together as a multi-agent team.
When an agent collects more than three custom commands, Taskade automatically promotes the set into Agent Skills: the agent sees a catalog of skill names and loads only the relevant one into context on demand. That keeps the agent fast even as your command library grows into the dozens.
A command library, at a glance
Picture an agent's command menu the way a team actually uses it. Each line is a saved prompt one keystroke away.
┌─────────────────────────────────────────────┐
│ AGENT: Front Desk Commands tab │
├─────────────────────────────────────────────┤
│ /qualify-lead Score on 5 intake answers │
│ /draft-reply Reply in our brand voice │
│ /summarize 3 bullets + next step │
│ /research-co Web search → brief │
│ ────────────────────────────────────────── │
│ 4 commands · Skills mode ON (auto >3) │
└─────────────────────────────────────────────┘
Getting Started
To create your first command, open an agent, go to the Commands tab, click New command, name it (this becomes the slash trigger), describe the prompt behavior, choose Default or Plan & Execute, then save. The command appears in the slash menu across your projects, ready to run.
Start with the task you repeat most. If you rewrite the same kind of email daily, that is your first /draft-reply. The command pays for itself the second time you use it. For step-by-step screenshots, see custom agents in Learn Taskade and bulk commands.
Custom Commands FAQ
What is a custom agent command?
A custom command is a saved prompt you trigger with / + a name, so an AI agent runs a defined task the same way every time. You author it once in the agent's Commands tab, then call it on demand across any project.
How do I create a custom command?
Open an agent, go to the Commands tab, click New command, name it, write the prompt behavior, pick Default or Plan & Execute mode, and save. The command then appears in the slash menu in your projects.
What is the difference between Default and Plan & Execute mode?
Default mode returns a direct single-pass response, ideal for rewriting, summarizing, or extracting. Plan & Execute mode has the agent set structured goals and work through them step by step, ideal for research, analysis, or multi-stage tasks.
How are custom commands different from built-in tools?
Built-in tools are the 34 capabilities every agent ships with, like web search and code execution. Custom commands are shortcuts you author from your own prompts, and they often call those tools under the hood. Tools are automatic; commands are the saved plays you trigger.
What happens when I have many custom commands?
Once an agent has more than three custom commands, Taskade promotes them into Agent Skills. The agent sees a catalog of names and loads only the relevant skill on demand, so it stays fast even with dozens of commands.
Can my team share custom commands?
Yes. Share an agent and its commands travel with it, so every teammate gets the same assistant with the same shortcuts. Pair this with multi-agent teams to give different agents different command sets.
Can custom commands run inside automations?
Yes. You can call commands inside reliable automation workflows, so a saved prompt fires on a schedule or a trigger. Triggers pull events in, actions push results out, across 100+ integrations.
What You'd Build in Taskade
Custom commands stop being a clever trick the moment they live inside a real workspace. Picture a small operations team running on a Taskade Ops Dashboard. The dashboard pulls in new requests across connected projects, and a Front Desk agent carries four saved commands: /qualify-lead, /draft-reply, /summarize, and /research-co. A coordinator opens a new request, types /qualify-lead, and the agent scores it on the same five questions every time. A reliable automation workflow fires /draft-reply overnight so replies are queued before anyone logs in. The team sees one clean board, logs in with their own accounts, and the routine work runs on its own.
That is the shape: one prompt becomes one app, with agents that remember, reason over 15+ frontier models from OpenAI, Anthropic, Google, and open-weight providers, and run across 100+ integrations. Describe the operation you repeat, and Taskade EVE, the meta-agent behind Taskade Genesis, builds the foundation under it. Start building free →