Skip to main content

Troubleshooting Automations

This guide shows a reliable workflow for debugging an Automation (Workflow) in Attive and improving it quickly.

The main idea: use the Automation agent as much as possible. When something looks off, feed the agent the symptoms (what you expected vs what happened) and ask it to fix the workflow.

The troubleshooting loop​

1) Build (or update) the workflow​

  • Start from a clear goal: what should happen, for which records, and when.
  • Keep inputs reusable (avoid hardcoded IDs).
  • Prefer query → extract → action so later steps rely on simple, named outputs.

2) Test the workflow​

Test early and often.

  • If it’s a manual/on-demand run, run it with a realistic set of inputs.
  • If it’s an event-triggered workflow, test with a specific record (e.g., the trigger’s testRecordId) and confirm the workflow is using the right entity and properties.
  • If it’s scheduled, verify required inputs have defaults so the schedule can run safely.

3) Verify the outputs (and the behavior)​

When you review a run, look for three things:

  • Did the right steps run?
    • Were steps skipped due to conditions?
    • Did dependencies execute in the order you expected?
  • Are outputs shaped correctly?
    • Are downstream steps reading simple variables (e.g. extracted fields), not raw query results?
  • Did the final action do the right thing?
    • Slack message posted to the right channel/user?
    • CRM writes updated the intended object type/fields?

Use the Automation agent to fix issues​

The fastest way to troubleshoot is to “close the loop” with the Automation agent.

What to send the agent​

Copy/paste (or describe) these items:

  • What you expected vs what happened
  • Which run you’re looking at (and whether it was scheduled, manual, or event-triggered)
  • Inputs used (or defaults, for scheduled runs)
  • The step(s) that look wrong (step IDs are ideal)
  • Observed outputs (missing, empty, wrong shape, wrong values)
  • Errors (any error message you see, plus which step it came from)

Example prompts you can use​

  • “This run completed but the Slack message is missing fields. The ai_extract outputs are empty. Please fix the workflow so the extracted variables are always present (or skip Slack when they’re missing).”
  • “The workflow is too noisy: it triggers too often. Please tighten the event trigger conditions and add step-level conditions so Slack only posts when the threshold is met.”
  • “The CRM update wrote to the wrong field. Please use CRM field search to find the correct API name, update the step, and re-run.”

Common symptoms and what to do​

“The workflow didn’t trigger”​

  • Confirm you’re using the intended trigger type (schedule vs event).
  • For event triggers, confirm:
    • The entity name matches your graph entity.
    • The watched properties are correct (for property change triggers).
    • The conditions match the new values you expect.

Then ask the Automation agent:

  • “This workflow didn’t trigger when I expected. Here’s the record and the change I made. Please adjust the trigger config to match and explain what changed.”

“Steps ran but outputs are missing / unusable”​

  • If downstream steps are referencing complex objects, add an ai_extract step to turn them into simple variables.
  • If the input is large/text-heavy, use ai_summarize before extraction or generation.

Then ask the Automation agent:

  • “Outputs are missing / wrong shape. Please add the right extraction step(s) so downstream steps only consume simple variables.”

“Slack message is wrong / noisy”​

  • Add step conditions so Slack only posts when the workflow has enough signal.
  • Make the message include the most important context and a clear next step.

Then ask the Automation agent:

  • “Please tighten Slack posting rules and improve the message formatting. Here’s an example run output and what I want the message to contain.”

“CRM update/create failed”​

  • Verify object type and field names are the system/API names (not display names).
  • Ensure the workflow is using the right record IDs.

Then ask the Automation agent:

  • “CRM write failed. Please search the correct object/field API names, update the step config, and rerun the workflow.”

Final review (so you can speak about it confidently)​

Before publishing (or after a big fix), do one last pass:

  • Explain the workflow in one minute: trigger → key steps → outcome.
  • Confirm each step’s purpose and what it outputs.
  • Confirm conditions match your intent (what gets skipped, and why).
  • Confirm the workflow is reusable via inputs.

If anything feels unclear, ask the Automation agent to:

  • “Rewrite step names/descriptions to match what they do, and simplify the workflow where possible.”