How to Prompt Miraxa: Writing Effective Instructions

Learn how to phrase questions and instructions so Miraxa, the intelligent layer across your automation, builds, debugs, and explains your Spojit workflows accurately the first time.

Overview

Miraxa is the intelligent layer across your automation in Spojit. It is the chat surface available on every page: it can build and edit workflows from a sentence, explain how a feature works in the context of the page you are on, investigate why a run failed, and suggest ways to make a workflow run faster and cheaper. In the Workflow Designer it can add, delete, update, and connect nodes on the canvas for you. Because it acts on real nodes, connectors, and variables, the quality of what it does depends almost entirely on how precisely you ask.

The mental model is simple: Miraxa knows the Spojit vocabulary you can see in the app, so the closer your prompt is to that vocabulary, the more deterministic the result. A vague request makes Miraxa guess (and it will often ask a clarifying question before acting); a specific request that names real node types, connectors, exact tool names, and variables lets it act immediately and correctly. The most reliable pattern is to scaffold with Miraxa, then fine-tune the details in the node properties panel. This guide shows you how to write prompts for the three things you will ask Miraxa to do most: building, debugging, and explaining.

Steps

Open Miraxa from the chat surface on any page in Spojit, then apply these techniques.

Step 1: Be specific about the outcome

State exactly what you want to happen, not just the topic. Replace a broad goal with the concrete nodes and conditions involved.

Weak: do something when an order comes in
Strong: Add a Condition node that checks if {{ order.total }} is over 100, and connect the true branch to a Send Email node.

The strong version names a real node type (Condition), the exact comparison, the variable, and where the branch should connect, so Miraxa can place and wire the nodes without guessing.

Step 2: Name real node types, connectors, and tools

Use the exact names you see in the designer and the connector catalog. Node types include Trigger, Connector, Transform, Condition, Loop, Parallel, Human, Knowledge, Send Email, Attachment, Subworkflow, and Response. When you mean a specific action, name the connector and its tool verbatim, such as the Slack connector's send-message tool or the Shopify connector's get-order tool.

Weak: post the result to chat
Strong: Add a Connector node in Direct mode that uses the Slack send-message tool to post a summary to the #orders channel.

Step 3: Reference variables with handlebars

Spojit passes data between steps as variables, written with handlebars such as {{ order.total }}, {{ input.subject }}, or {{ step.field }}. When you tell Miraxa exactly which value to use, it maps the field instead of inventing a placeholder.

Weak: email the customer their order info
Strong: In the Send Email node, set Recipients to {{ order.customer.email }} and put {{ order.number }} in the Subject.

Step 4: Match the prompt to the task: build, debug, or explain

Miraxa does different jobs, and each wants a different shape of prompt.

Building: describe the trigger, the steps in order, and the connections. Example: Build a workflow that watches a mailhook, extracts the PDF invoice with an Attachment node, and posts a summary to Slack.
Debugging: point Miraxa at the run or node and ask what went wrong. Example: Why did my last run fail? or The Shopify get-order step returned nothing. What is wrong? On a workflow page, Miraxa can read the execution history and the node configuration to answer.
Explaining: ask a focused question about a feature or choice. Example: What is the difference between Agent mode and Direct mode? or How does the Timeout field on the Human node behave?

Step 5: Scaffold first, then refine in the properties panel

For larger workflows, ask Miraxa for the structure first, then open each node and adjust the fields yourself. For example, start with Add a Loop node set to ForEach over {{ attachment.attachments }}, and inside it a Knowledge node in Embed mode. Once the nodes exist, open the Knowledge node's properties panel to set the Collection, Document Type, and Document Input precisely. This keeps the conversation short and gives you exact control over the details.

Tips

Put one clear instruction per message when you want a predictable canvas edit. A single message asking for five unrelated changes is harder for Miraxa to wire correctly than five focused ones.
When you want structured output from a Connector node in Agent mode or a Knowledge node in Query mode, say so: ask Miraxa to "force JSON output with a Response Schema" so the result is reliable to use downstream.
If Miraxa asks which node to connect a new node to, answer with the node's label rather than re-describing the whole workflow. It already knows the canvas you are editing.
For cost-conscious changes, ask Miraxa to "find ways to make this workflow run faster and cheaper": it can suggest moving a predictable single-tool call from Agent mode to Direct mode to avoid AI credits.

Common Pitfalls

Using generic words like "the spreadsheet tool" or "the email thing" instead of real names. If you are not sure a tool exists, name the connector and the action one level higher (for example, "your CRM connector") and let Miraxa propose the exact tool.
Forgetting handlebars on values. Writing order.total as plain prose may produce a literal string; write {{ order.total }} so it resolves as a variable.
Asking Miraxa to "fix everything" on a failed run without context. Open the workflow first so it can see the execution and the node, then ask about the specific step that failed.
Expecting Miraxa to invent data it cannot see. The Send Email node only resolves upstream variables, so a prompt that references the workflow name or status will not produce a usable value.