Plan Executor

Use when: User says execute plan and has a written plan to execute, either in current session or via subagents.

Trigger Phrases

• execute plan
• run the plan
• implement the plan

Execution Modes

Mode 1: BATCH (Sequential in Current Session)

Use when:

• Tasks have dependencies
• Need tight coordination
• Plan is small (< 10 tasks)

For each task in plan:
  1. Mark task in-progress (TodoWrite)
  2. Implement task
  3. Verify task
  4. Mark task complete
  5. Continue to next

Mode 2: SUBAGENT (Parallel via Task Tool)

Use when:

• Tasks are independent
• Plan is large (10+ tasks)
• Tasks touch different parts of codebase

1. Identify independent task groups
2. Dispatch Task() for each group
3. Collect results
4. Handle failures
5. Merge and verify

Workflow

Phase 1: Load Plan

1. Find plan file (check in order):

   • User-specified path
   • conductor/tracks/<id>/plan.md
   • docs/plans/*.md (most recent)
   • .beads/ database via bd ready

2. Parse plan structure:

   • Phases/sections
   • Individual tasks
   • Dependencies between tasks

Phase 2: Classify Execution Mode

IF plan has sequential dependencies:
  mode = BATCH
ELIF tasks are independent AND count > 5:
  mode = SUBAGENT  
ELSE:
  mode = BATCH

Ask user to confirm mode if ambiguous.

Phase 3: Execute

BATCH Mode:

for task in tasks:
  todo_write(task, status="in-progress")
  implement(task)
  verify(task)  # tests, lint, typecheck
  todo_write(task, status="completed")

SUBAGENT Mode:

groups = partition_independent_tasks(tasks)
for group in groups:
  Task(
    description: "Implement: {group.summary}",
    prompt: """
      Context: {plan_context}
      Tasks: {group.tasks}
      
      For each task:
      1. Implement with TDD
      2. Verify (tests, lint)
      3. Report status
      
      Return: summary, files changed, issues encountered
    """
  )

Phase 4: Verify All

1. Run full test suite
2. Run typecheck/lint
3. Check for conflicts between parallel work
4. Report overall status

Phase 5: Update Tracking

# If using beads
bd update <epic-id> --notes "Plan executed. Completed: X/Y tasks"

# If using conductor
# Update plan.md with [x] markers and commit SHAs

Output Format

PLAN: <plan-name>
MODE: BATCH | SUBAGENT
TASKS: <completed>/<total>

[Per-task status...]

RESULT: Success | Partial | Failed
ISSUES: <any problems encountered>
NEXT: <recommendation>

Error Handling

• If task fails, mark as blocked with reason
• Continue with independent tasks if in SUBAGENT mode
• Stop and report if in BATCH mode (unless user overrides)
• Always run verification before claiming success