Skip to main content

Run a Workflow

Open the workflow (BPMN Workflows → pick it) and fill in the Execution request panel:

  • Goal / Description — what this run should accomplish.
  • Ticket Key — optional, e.g. VLC-123, shown throughout the run for traceability.
  • Acceptance Criteria — one per line; the agent checks its own work against these before finishing.
  • Repository — shown when an assigned profile declares workspace.needs_repository: true. If the project has exactly one repository, AgentMesh selects it automatically. For a multi-repository project, select the required repository role instead of assuming an alias named primary exists.
  • Delivery Branch / Base Branch — leave blank to auto-generate a branch name, or set your own.
  • Task Instruction / Additional Text — any extra detail or context the agent needs.
  • Attachments — files relevant to the task, if any. The Backend stores them as references; the selected worker downloads and materializes them under the run workspace's input/ directory before the provider starts. Binary files are never embedded in the RunSpec as base64.

Click Run Workflow.

Watching it run​

The instance appears under Workflow Instances, and each dispatched step under Runs (or the main Dashboard), with live status. Click a run to see its detail: which agent profile executed it, the branch, and (once finished) its result.

If a step fails​

A failed step shows up as an incident — the workflow stays running but that step is stuck until you act. Open the failed run and click Retry task once you've fixed the underlying problem (usually a missing connection binding — see Connect a Git Repository — or a missing required resource on the agent profile). AgentMesh keeps the same workflow instance, resolves the incident, and schedules the failed task for another attempt. The new agent execution may start from the beginning; it does not resume an interrupted CLI session.

Before you run: a quick checklist​

  • At least one Worker shows Online for your organization, with free capacity.
  • Every connection alias an assigned agent profile needs (memory provider, a plugin's connection) is bound on this project — see Connect a Git Repository.
  • Every required Skill/Hook/Plugin/MCP/Knowledge dependency attached to an assigned agent profile is published and active in Registry.

If dispatch cannot resolve a required dependency or connection, the affected task fails with an incident and a diagnostic message in the run detail.

Next​

You're now running governed agents end to end. If you or your team also want to drive AgentMesh's own tools directly from your editor (not through a workflow), see Connect Your Editor via MCP.