Orchestrate (command)
Contributed by Rohit Ghumare (awesome-claude-code-toolkit)
Source: original file · added 2026-10-29
Run a multi-step workflow by breaking a complex task into coordinated sub-tasks.
Steps
- Parse the workflow specification from the argument:
- Accept a natural language description of the end goal.
- Or accept a structured plan with explicit steps.
- Decompose into ordered sub-tasks:
- Identify dependencies between tasks (which must complete before others start).
- Determine which tasks can run in parallel.
- Estimate complexity of each task (small, medium, large).
- For each sub-task, define:
- Clear objective and success criteria.
- Input requirements (files, data, prior task outputs).
- Expected output (files created, changes made, results).
- Verification method (test, manual check, build success).
- Execute tasks in dependency order:
- Mark each task as pending, in-progress, or complete.
- Capture output and errors from each step.
- If a task fails, determine if downstream tasks should be skipped or can proceed.
- After all tasks complete, run a final verification:
- Build passes.
- Tests pass.
- No regressions introduced.
- Report the full execution summary.
Format
Workflow: <description>
Tasks: <total> (<completed>/<total>)
| # | Task | Status | Duration | Notes |
|---|------|--------|----------|-------|
| 1 | <task> | done | 2m | <notes> |
| 2 | <task> | done | 5m | <notes> |
| 3 | <task> | failed | 1m | <error> |
Overall: <success/partial/failed>
Duration: <total time>Rules
- Never execute destructive operations (delete, force push) without explicit confirmation.
- If a critical task fails, stop and report rather than continuing blindly.
- Keep each sub-task focused and independently verifiable.
- Save progress after each completed task so work is not lost on failure.
- Limit workflow to 10 tasks maximum; break larger workflows into phases.
Original source file (unmodified)
Run a multi-step workflow by breaking a complex task into coordinated sub-tasks. ## Steps 1. Parse the workflow specification from the argument: - Accept a natural language description of the end goal. - Or accept a structured plan with explicit steps. 2. Decompose into ordered sub-tasks: - Identify dependencies between tasks (which must complete before others start). - Determine which tasks can run in parallel. - Estimate complexity of each task (small, medium, large). 3. For each sub-task, define: - Clear objective and success criteria. - Input requirements (files, data, prior task outputs). - Expected output (files created, changes made, results). - Verification method (test, manual check, build success). 4. Execute tasks in dependency order: - Mark each task as pending, in-progress, or complete. - Capture output and errors from each step. - If a task fails, determine if downstream tasks should be skipped or can proceed. 5. After all tasks complete, run a final verification: - Build passes. - Tests pass. - No regressions introduced. 6. Report the full execution summary. ## Format ``` Workflow: <description> Tasks: <total> (<completed>/<total>) | # | Task | Status | Duration | Notes | |---|------|--------|----------|-------| | 1 | <task> | done | 2m | <notes> | | 2 | <task> | done | 5m | <notes> | | 3 | <task> | failed | 1m | <error> | Overall: <success/partial/failed> Duration: <total time> ``` ## Rules - Never execute destructive operations (delete, force push) without explicit confirmation. - If a critical task fails, stop and report rather than continuing blindly. - Keep each sub-task focused and independently verifiable. - Save progress after each completed task so work is not lost on failure. - Limit workflow to 10 tasks maximum; break larger workflows into phases.