workflow-orchestration

name	workflow-orchestration
description	Execute non-trivial work using plan-first, verify-first, and lessons-fed defaults. Use when running 3+ step tasks, architectural work, or anything that may require re-planning.

Workflow Orchestration

Executable checklist for plan-first, verify-first execution. Use for any non-trivial task (3+ steps, architectural decisions, multi-stage verification).

Canonical Sources of Truth

Purpose	Location	Role
Active work	.cursor/plans/*.plan.md	Primary plan ledger; mark todos here
Durable completion	docs/{TITLE}_{MMMDD}_{YYYY}.md	Completion docs; add to MASTER_DOCUMENT_INDEX
Optional todo mirror	tasks/todo.md (MAS repo)	Lightweight operational summary; can be generated from .cursor/plans
Optional lessons	tasks/lessons.md (MAS repo)	Lessons learned; appended by pattern; pruned periodically

Relationship: .cursor/plans and docs/ are primary. tasks/todo.md and tasks/lessons.md are optional mirrors—use them only if they add operational value. Do not replace the stronger systems.

Executable Checklist

Workflow Progress:
- [ ] 1. Plan (or load plan)
- [ ] 2. Verify the plan
- [ ] 3. Execute
- [ ] 4. Mark progress
- [ ] 5. Verify
- [ ] 6. Elegance review (non-trivial only)
- [ ] 7. Document results
- [ ] 8. Capture lessons (when applicable)

1. Plan (or load plan)

• For 3+ step tasks: create or load a plan. Use plan-tracker when invoking.
• Read .cursor/CURSOR_DOCS_INDEX.md and task-relevant docs. For new plans: read last 72 hours of docs per read-recent-docs-before-planning.mdc.
• Identify sub-agents; read their .cursor/agents/<name>.md and apply their protocols.

2. Verify the plan

• Confirm scope, dependencies, and acceptance criteria are clear.
• Check for re-plan triggers: blockers, scope drift, or a cleaner approach available.
• If destabilized: stop and re-plan. Do not continue blindly.

3. Execute

• Work through plan todos in order. Mark in_progress as you start each.
• Use sub-agents for focused work. Keep one tack per subagent.
• Execute operations yourself—never ask the user to run commands.

4. Mark progress

• Update plan file: - [x] for completed todos.
• Update status docs if they list the task.
• If tasks/todo.md exists, sync completed items.

5. Verify

• Never mark complete without evidence. Run tests, check logs, verify runtime behavior.
• Use terminal-watcher for build/test output; test-engineer for test design; regression-guard for pre-deploy validation.
• See subagents-must-test-before-complete.mdc.

6. Elegance review (non-trivial only)

• For non-trivial changes: ask whether there is a more elegant approach.
• Do not over-engineer obvious, small fixes.

7. Document results

• Create dated completion doc: docs/{PLAN_NAME}_COMPLETE_{MMMDD}_{YYYY}.md.
• Add to MASTER_DOCUMENT_INDEX (and CURSOR_DOCS_INDEX if vital).
• Update plan source status to Complete. Link to completion doc.
• See plan-and-task-completion-docs.mdc and create-dated-document skill.

8. Capture lessons (when applicable)

• After user corrections or repeated mistakes: capture the lesson.
• Feed into rules, agent guidance, or tasks/lessons.md so the same mistake is less likely to recur.
• Use myca-autonomous-operator for lesson extraction and rule hardening.

Key Rules

• Plan mode by default for 3+ step tasks.
• Stop and re-plan when execution destabilizes.
• No task marked complete without proof.
• Use subagents aggressively for focused work.
• Execute operations yourself. Never hand work back to the user.