Example4 min read

Example 03 — Two Concurrent Agents

A feature Run and a bug-fix Run work on one repository without sharing a checkout.

Architecture

      workspace-root driver
               │
           repos/api-service
                /           \
feature Run / worktree   bug Run / worktree
       ↓ Candidate A       ↓ Candidate B
                \         /
                 validation + integration

The driver coordinates from the control repository root. Both bounded Runs target the same independent nested repository, but each receives its own Git worktree and branch.

Workspace config

runs:
  - id: feature-12
    write_scope: api-service
    repository_path: repos/api-service
  - id: bug-44
    write_scope: api-service
    repository_path: repos/api-service

Expected lifecycle

Both Runs publish Intent, use different branches and worktrees, validate independently, and hand off Candidates. Integration rebases each Candidate against current canonical history.

Example commands

ws run coder "Implement feature 12" --repo api-service --no-integrate &
ws run coder "Fix bug 44" --repo api-service --no-integrate &
wait
ws runs
ws inspect <candidate-a-or-run-id>
ws inspect <candidate-b-or-run-id>

Trade-offs

Wall-clock time improves, but merge and semantic conflicts remain possible. Add Runtime Resource Isolation when either Run starts local services.