Skip to main content

Complete Agent Workflow

This page describes the recommended workflow for AI agents using Tracer. Follow these steps in every session to maintain context and coordinate work effectively.

1. Find Ready Work

At the start of any session, check for unblocked work:
Always start your session with tracer ready to orient yourself. This shows you exactly what can be worked on without blockers.

Parsing Ready Work

2. Claim and Start Work

Once you’ve selected an issue, claim it by updating the status:
When you update an issue to in_progress, Tracer automatically assigns it to you (based on the TRACE_ACTOR environment variable) if no assignee is set.

Example: Claiming Work

3. Create Issues During Work

As you discover new work, file it immediately to ensure nothing gets lost:

Example: Discovered Work

File issues liberally - it’s better to have too many tracked issues than to lose important work.

4. Add Dependencies

When you realize work blocks other work, model it explicitly:

Dependency Types

  • blocks: “bd-2 cannot start until bd-1 is done”
  • parent-child: “bd-2 is a subtask of epic bd-1”
  • discovered-from: “bd-2 was found while working on bd-1”
  • related: “bd-2 and bd-1 are connected but don’t block”

5. Update Status Regularly

Keep the tracker in sync with reality:
1

Starting Work

2

Hit a Blocker

3

Back to Work

4

Complete

6. Complete Work

When you finish a task, close it with a meaningful message:
Provide meaningful completion messages in --reason. This helps with audit trails and understanding what was accomplished.

7. Check Status

Regularly review the overall state of your work:

Complete Integration Example

Here’s a complete agent workflow script:

Example Session

Here’s what a typical agent session looks like:

Advanced Workflows

Next Steps

JSON Output

Learn about JSON response formats

Best Practices

Follow proven patterns for success