track-manager

Track Manager Skill

Manage the lifecycle of Conductor tracks including status updates, task completion, and phase transitions. All operations follow the Evaluate-Loop process defined in conductor/workflow.md.

MANDATORY: Evaluate-Loop Integration

Every track operation must follow the Evaluate-Loop:

PLAN → EVALUATE PLAN → EXECUTE → EVALUATE EXECUTION → COMPLETE/FIX

Key rules:

1. ALWAYS update plan.md after completing any task (prevents duplicate work across sessions)
2. ALWAYS evaluate before marking anything complete
3. NEVER skip the pre-execution plan evaluation

Trigger Conditions

Use this skill when:

• Checking track status or progress
• Marking tasks as complete
• Transitioning between phases
• Running pre/post execution evaluations
• User mentions: "track status", "mark complete", "next task", "update plan", "evaluate"

Track Structure

conductor/
├── tracks.md           # Master track list
├── authority-matrix.md # Lead Engineer decision boundaries
├── schemas/
│   └── track-metadata.v2.json  # Metadata schema definition
└── tracks/
    └── <track_id>/
        ├── spec.md         # Requirements
        ├── plan.md         # Phased tasks (MUST be kept updated)
        └── metadata.json   # v2 status with loop_state

Metadata v2 Protocol

All tracks use the v2 metadata schema with explicit loop state tracking.

Creating a New Track

Initialize metadata.json with v2 structure:

{
  "version": 2,
  "track_id": "feature-name_20260131",
  "type": "feature",
  "status": "new",
  "created_at": "2026-01-31T00:00:00Z",
  "updated_at": "2026-01-31T00:00:00Z",

  "loop_state": {
    "current_step": "PLAN",
    "step_status": "NOT_STARTED",
    "fix_cycle_count": 0,
    "max_fix_cycles": 5,
    "plan_revision_count": 0,
    "max_plan_revisions": 3,
    "checkpoints": {
      "PLAN": { "status": "NOT_STARTED" },
      "EVALUATE_PLAN": { "status": "NOT_STARTED" },
      "EXECUTE": { "status": "NOT_STARTED" },
      "EVALUATE_EXECUTION": { "status": "NOT_STARTED" },
      "FIX": { "status": "NOT_STARTED" },
      "BUSINESS_SYNC": { "status": "NOT_STARTED", "required": false }
    }
  },

  "lead_consultations": [],
  "discovered_work": [],
  "blockers": []
}

Updating Loop State

When a step completes, update the checkpoint:

{
  "loop_state": {
    "current_step": "EXECUTE",
    "step_status": "IN_PROGRESS",
    "checkpoints": {
      "PLAN": {
        "status": "PASSED",
        "completed_at": "2026-01-31T10:00:00Z",
        "agent": "loop-planner"
      },
      "EVALUATE_PLAN": {
        "status": "PASSED",
        "completed_at": "2026-01-31T10:30:00Z",
        "verdict": "PASS"
      },
      "EXECUTE": {
        "status": "IN_PROGRESS",
        "started_at": "2026-01-31T11:00:00Z",
        "tasks_completed": 3,
        "tasks_total": 10,
        "last_task": "Task 1.3"
      }
    }
  }
}

Migrating v1 to v2

If a track has v1 metadata (no version field or loop_state):

1. read_file current metadata fields
2. Infer loop state from plan.md content
3. Add v2 structure with inferred values
4. write_file back to metadata.json

Task Status Markers

Workflow Operations

Before Starting ANY Work

1. read_file tracks.md to see what's already complete
2. read_file the track's plan.md to see what tasks are done vs pending
3. read_file spec.md to understand requirements
4. Evaluate the plan — verify scope matches spec, no overlap with completed tracks

Start a Task

# Before
- [ ] Implement user authentication

# After (mark in progress)
- [~] Implement user authentication

Complete a Task (MANDATORY: update plan.md immediately)

# After completion (add commit SHA + summary of what was done)
- [x] Implement user authentication <!-- abc1234 -->
  - Created src/components/auth/signup-form.tsx
  - Added email/password validation
  - Integrated with mock API client

Update tracks.md

When completing a phase, update conductor/tracks.md:

## Active Tracks

| Track ID | Type    | Status      | Progress  |
| -------- | ------- | ----------- | --------- |
| auth-001 | feature | in_progress | Phase 2/3 |

Phase Transition Rules

1. All tasks in phase must be [x] before moving to next phase
2. Run post-execution evaluation (see Evaluate-Loop in conductor/workflow.md)
3. If evaluation fails → create fix tasks → execute → re-evaluate (loop)
4. If evaluation passes → update metadata.json with completion timestamp
5. Create commit for phase completion
6. Update tracks.md progress column
7. Update conductor/index.md current status

Post-Execution Evaluation Checklist

Before marking a track complete, verify:

Response Format

After track operations:

## Track Update

**Track**: [track_id]
**Operation**: [started/completed/updated/evaluated]
**Phase**: [phase number] - [phase name]
**Progress**: [completed]/[total] tasks
**Evaluation**: [PASS / FAIL - describe issues]
**Next**: [next task description]