Migrate an automation
You can use an existing automation as the starting point for a qbash task. The new-task screen accepts material from n8n, OpenClaw, Claude Code, or Codex, and the development agent helps rebuild it as Python, prompts, and task settings you can edit.
Bring an example input and a result you are happy with. For a company-brief automation, that could be a company name, meeting notes, and a brief that includes the right details. Explain what starts it, which systems it uses, and where people need the result.
Importing an automation
Choose Import an automation on the new-task screen and provide the workflow definition, code, or instructions. The agent uses this material to create a draft in the editor.
You can also choose Write code and use Iterate with AI to work from a description. This is useful for a small script or a procedure that lives in someone's notes or conversation history.
Connect accounts and Secrets under Resources when the task needs them. Credentials from the original system are not transferred with its code.
Files, connections, and triggers
The imported automation uses the same parts as any other qbash task:
| Existing automation | In the task editor |
|---|---|
| Inputs or form fields | Task inputs |
| Code and transformations | main.py and Python scripts |
| Model instructions | Prompts |
| An agent that uses tools | A prompt with Agent mode enabled |
| Reference documents or a rubric | Markdown documents or memory |
| Connected accounts and credentials | Resources |
| A reusable workflow | A linked task |
| A schedule, event, API, or webhook | Triggers |
| A returned result | qbash.output(...) |
The structure may change during the move. Several transformation nodes in a visual workflow might become one Python function. The useful comparison is whether the task follows the same rules and produces the result you need.
Source material
n8n
Export the workflow JSON and include any expressions, environment variables, or community nodes it depends on. The agent can use the workflow to understand the sequence, while qbash connections replace its saved credentials.
OpenClaw
Bring the instructions, tool definitions, and examples that describe the job. Reusable tools can become Python scripts or linked tasks, with the prompt calling them when needed.
Claude Code and Codex
Bring the relevant source files and explain how you currently run them. If the code reads local files or depends on a command installed on your machine, that part needs adapting: qbash runs the task in the cloud. Files can become task inputs or documents, and connected services can supply data that previously came from a local environment.
Cowork
A procedure can be rebuilt from its standing instructions, source references, and an example result. Include corrections that matter to the job, such as which figures to trust or how a report should be formatted. Conversation history, local files, and schedules do not transfer automatically.
Testing the imported task
Use Test run with an example you know. For a company brief, compare the facts and requested sections rather than expecting identical wording from the model. Inspect the console to see what was read, which prompts ran, and what the task tried to write.
Writes are blocked by default, so you can inspect a proposed message or update before sending it. Try an empty result or a missing input too, especially if the previous system handled those cases for you.
Iterate with AI can help with differences. Give it the input, the expected result, and what happened instead. For agents, the tool calls in the console can explain a difference that is not obvious from the final answer.
Switching over
Save the task when you are satisfied with the draft. Configure its timeout, retries, and concurrent-run settings, then make it available where the original automation was used. Distribution covers API callers, AI assistants using MCP, schedules, and events.
When replacing a live automation, pause its old trigger before enabling the new one so both systems do not deliver the same work. Check the first run and its destination, and keep the old definition available while you verify the replacement.
The task editor explains files, settings, versions, and run history. Testing and debugging covers test files, draft Test runs, and the editor console.