AI skills

Run a workflow from an agent

Let an agent invoke a published, permitted on-demand workflow instead of improvising its steps.

Before you start

Use a test channel or sample conversation first. Confirm the visible result in ValueChat before applying the setup to production traffic.

How it works

  1. 01

    Prepare

    Prepare. Create an On demand workflow, choose its permitted inboxes, configure its steps and publish it. Event-only or draft workflows are not arbitrary callable tools.

  2. 02

    Agent setup

    Agent setup. Add Run workflow, select the actual workflow and describe the request conditions that justify using it. References identify the saved workflow rather than relying on an unlinked name.

  3. 03

    Execution

    Execution. The application validates workspace/channel permissions, runs the graph and returns its real execution result. Connected app steps wait for their provider response before later confirmations.

  4. 04

    Ownership

    Ownership. An explicit assignment inside the flow uses the ordinary handoff path. After handoff or closing, the previous agent must not continue acting as owner. Calling a workflow does not automatically authorize another account or channel.

  5. 05

    Failure and preview

    Failure and preview. A failed or uncertain action is not a successful flow; earlier external writes may already exist. Preview simulates actions without real provider calls. Cyclic parent/child calls are rejected.

What to verify

✓The expected channel, conversation, contact or execution is visible.

✓The assigned person or agent matches the routing rule.

✓No customer-facing message claims an action that failed.

✓Activity or execution history contains enough context to investigate.

Need help?

Collect the workspace, channel, conversation and execution identifiers, then contact support@valuechat.app. Never send API keys or passwords.