Common troubleshooting steps for Pylon's Agentic features

Last updated: September 9, 2026

Use these checks when an Agentic workflow does not start, cannot use a tool, returns an incomplete result, or does not complete the expected action. Work through the checklist in order so you can identify whether the problem is with the tool, connector, skill, trigger, or agent run.

Quick troubleshooting checklist

  1. Check that the required tool exists and is enabled for the agent.

  2. Check that the connector is connected with the right account and permissions.

  3. Check that the skill is enabled, applies to the request, and references available tools.

  4. For automatic workflows, check that the trigger matched and launched the intended agent.

  5. Review the run to find the selected skill, attempted tools, and any errors.

  6. Verify the expected action in its destination instead of relying only on the agent's message.

1. Check tool access

First, identify the exact capability the workflow needs. Examples include searching a knowledge base, reading an external system, updating a field, or preparing a draft.

  • Open Settings > AI Controls > Agents and select the affected agent.

  • Review the agent's available tools and confirm that the required tool is enabled.

  • Confirm that the tool supports the requested action. A similarly named tool may support reading data but not changing it, or it may prepare a draft instead of sending it.

  • If the required tool is not available, changing the prompt or skill will not add that capability.

If the agent attempted the tool and received an error, continue to the connector checks.

2. Check the connector setup

Tools that access another system depend on a working connector and valid credentials.

  • Open Settings > AI Controls > Connectors and confirm that the connector is connected.

  • Confirm that the connection used by the agent is authenticated and has not expired.

  • Confirm that the connected account has permission to access the required data or perform the requested action in the external system.

  • Confirm that the required tools from the connector are enabled for the affected agent.

Human-driven and autonomous workflows can use different connection paths. Confirm the connection used by the specific agent you are troubleshooting.

3. Check the skill

If the tool and connector are available, review the skill that should guide the workflow.

  • Confirm that the skill is enabled for the agent.

  • Review the skill description. It should clearly state the situations in which the skill applies.

  • Confirm that every tool and source required by the skill is available to the agent.

  • Check for conflicting instructions between the skill and the agent's general instructions.

  • Keep detailed, repeatable procedures in the skill rather than in broad agent instructions.

When testing with Assist Agent, invoke the skill directly with /skill-name. If the skill works when invoked directly but is not selected automatically, improve its description. If it starts but cannot complete the workflow, check its instructions, tools, and connector access.

4. Check the trigger or assignment

Use this step when the agent never started.

  • Confirm that the issue or event matched the trigger conditions at the time the trigger evaluated it.

  • Confirm that the trigger is enabled and launches or assigns the intended agent.

  • Confirm that the relevant issue fields, team, tags, or channel values match the trigger conditions.

If the trigger did not launch the agent, fix the trigger or assignment before changing the skill.

5. Check the agent's context and resources

An agent can only use the context and resources available in that workflow.

  • Run the test from the relevant issue or account when that context is important.

  • Confirm that the required knowledge base, article, issue history, or other resource is available to the agent.

  • Check any visibility or access settings on the source content.

  • Make the request specific and include the expected outcome.

6. Review the run

If the agent started but the result is missing, slow, or incomplete, review its run details under Settings > AI Controls > Agents.

Check logs to look for:

  • Whether the run is still processing, completed, stopped, or failed

  • Which skill was selected

  • Which tools were attempted

  • Tool errors, permission errors, or authentication errors

  • The last successful step before the workflow stopped

External tools, attachment processing, and other delegated work can add time. Use the run status to distinguish active work from a failed or stopped run.

7. Verify the final action

An agent may research successfully but fail during the final action. It may also prepare an action that still requires human review.

  • Check whether the result is a suggestion, a draft, or a completed action.

  • If a draft was accepted, confirm whether it was placed in an editor and still needs to be sent.

  • Check the destination system or Pylon issue to confirm that the change actually occurred.

  • Do not treat the agent's written confirmation by itself as proof that an action completed.

8. Test one change at a time

After identifying the likely failure point, change one setting and rerun the same representative test. This makes it easier to tell whether the tool, connector, skill, or trigger change resolved the problem.

When to contact Pylon Support

If the problem continues, contact Pylon Support with the affected issue link and the error or unexpected behavior you observed.