> ## Documentation Index
> Fetch the complete documentation index at: https://docs.gocromo.com/llms.txt
> Use this file to discover all available pages before exploring further.

# Build and run a Workflow

> Draft a recurring process in chat, test it safely, activate a trigger, and investigate runs.

This guide creates a weekly renewal-review Workflow. Adapt the pattern to any repeatable process with clear inputs and outputs.

## 1. Prove the process manually

Before automating, run the task once in chat.

Attach the Renewal Policy page and Renewals table, then ask:

> Find renewals due in the next 90 days. Group them by account owner, flag high-risk records and records without a next action, and produce a CSV plus a concise summary.

Review the result with the process owner. Automation should preserve an agreed process, not hide an unresolved one.

## 2. Describe the Workflow contract

Open **Workflows → New workflow**, or ask from chat:

> Create a workflow named Weekly renewal review. It should accept an optional as-of date, read the Renewals table, select records due within 90 days, group them by account owner, flag high-risk or incomplete records, write a dated CSV to Project files, and return a Markdown summary. It must fail without updating records if the table cannot be read.

Include:

* trigger and timezone;
* input types and defaults;
* data and Knowledge references;
* output files or records;
* side effects;
* retry and failure rules.

## 3. Inspect the definition

Open the Workflow and use the graph view to follow the execution order. Click each node and check:

* what input it receives;
* where mapped values come from;
* what output it produces;
* whether it can call an external system;
* what happens when it fails;
* whether fan-out branches are aggregated correctly.

Use the raw JSON view when you need to verify exact argument names or mappings.

## 4. Ask for changes

Use **Ask to edit** instead of changing the definition by hand.

For example:

> Change the workflow so an empty result still creates a summary saying that no renewals are due, but does not create a CSV. Add the table name and as-of date to every failure message.

Reopen the graph and verify the changed branches.

## 5. Run a safe test

Select **Run now** and provide representative inputs.

If the Workflow accepts a file or Project reference, choose it in the generated input form. File inputs can be up to 10 MB each.

For the first test:

* use a copy or a non-critical record set;
* disable unnecessary external writes;
* choose a date that produces known results;
* tell the process owner when the run will happen.

## 6. Inspect the run

Open the run detail and verify:

* the run status;
* resolved arguments;
* each node's input and output;
* retries or skipped branches;
* generated files;
* external side effects;
* aggregate results after fan-out.

A failed node should expose an error that identifies the failing operation and relevant input. Fix the definition or dependency, then run a new test.

## 7. Activate a trigger

After a successful manual test, configure the required trigger:

* manual only;
* schedule;
* webhook;
* supported event.

For a schedule, confirm the timezone and intended day. For a webhook, store the token securely and rotate it if exposed. Update the calling system after rotation.

## 8. Monitor and pause

Review runs regularly, especially after changing:

* a Source or Knowledge policy;
* a Data model or table;
* connector credentials;
* Workspace or Project membership;
* the Workflow definition.

Pause the Workflow when its owner, input, or dependency is not trustworthy. Resume it after a successful manual test.

## Release checklist

* [ ] The process owner approved the expected result.
* [ ] Inputs and outputs are typed and named clearly.
* [ ] Side effects are limited to the intended systems.
* [ ] Failure and retry behavior were tested.
* [ ] A manual run completed with representative data.
* [ ] The trigger and timezone are correct.
* [ ] The owner remains a member of the Project.
* [ ] Someone is responsible for reviewing failed runs.