wb-flow

Appearance

Practical Guide: /wbWork (the agent) ​

What is /wbWork? ​

/wbWork is the execution arm of your planning process. While /wbPlan maps out what needs to be done, /wbWork actually does it. It transforms your AI from a Strategist into a Worker.

How to Use It ​

1. State & Sanity Check ​

bash
/wbWork plan_core2_20260503.md

If you pass no execution flags (like --id), the command acts as a status check. It reads the specific plan you linked, tells you which tasks are pending (⬜), and ensures the plan file formatting isn't corrupted. It will not execute tasks in this mode.

2. Execute Specific Tasks ​

Use the --id (or -i) flag to target specific tasks from the plan's table.

bash
/wbWork plan_core2_20260503.md --id=2 # Execute Task 2
/wbWork plan_core2_20260503.md --id=1,2,3 # Execute Tasks 1, 2, and 3
/wbWork plan_core2_20260503.md --id>2&&--id<=5 # Execute Tasks 3, 4, and 5 (AND logic)
/wbWork plan_core2_20260503.md --id<2||--id>5 # Execute Tasks 1, 6, 7... (OR logic)
/wbWork plan_core2_20260503.md --id!=2 # Execute all pending tasks EXCEPT 2

3. Execute Everything ​

bash
/wbWork plan_core2_20260503.md *

Executes all tasks currently marked as pending (⬜ Done).

4. Inline Tasks (Auto-Triage) ​

Instead of planning first, you can pass an issue directly to /wbWork. The AI will assess its complexity, add it to today's plan, and either execute it immediately (if simple) or break it into sub-tasks first (if complex).

bash
/wbWork packages/wb-core "Fix TypeError on submit" # Assessed as P3 -> One-shot execution
/wbWork packages/wb-core "Build new Auth System" # Assessed as P0 -> Breaks into sub-tasks, then executes

What Happens During Execution? ​

  1. The AI reads the task description from the plan.
  2. The AI writes code, creates files, or runs commands to complete the task.
  3. The AI generates a formal report in tasks/task_<ID>/task_<ID>_report_<TARGET>.md.
  4. The AI updates the main plan file, turning ☐ Done into ✅<br>ModelName.

Always pair /wbWork with a highly capable coding model (like the agent the AI agent or the AI agent). Once /wbWork is finished, switch to a "thinker" model (like the AI agent or the AI agent) and run /wbValid to verify the work.