mcpField
T
task-tracker

task-tracker

github.com/breverdbidderProductivity47 installs
Repository

About task-tracker

task-tracker is an ADHD-oriented task-tracking skill that models work through states and detects possible abandonment, then applies interventions and records completion streaks.

Key features

  1. 01Track tasks through defined states
  2. 02Detect context switches and stalled tasks
  3. 03Apply micro-commitment and chunking strategies
  4. 04Record completion streaks and task patterns

Use cases

  1. 01Following a task from request to completion
  2. 02Noticing when work stalls after a context switch
  3. 03Managing tasks blocked by external dependencies

Install

Copy the command and run it in a terminal.

$npx skills add https://github.com/breverdbidder/life-os --skill task-tracker

After installing, type/task-trackerin Claude Code to confirm it loaded.

Details

From the project README

Task Tracker - ADHD State Machine

Overview

Systematic task tracking designed for ADHD patterns. Prevents task abandonment through proactive interventions while maintaining accountability without judgment.

State Machine

                    ┌──────────────────────────────────────────┐
                    │                                          │
                    ▼                                          │
┌──────────┐   ┌──────────────────┐   ┌─────────────┐         │
│ INITIATED│──▶│ SOLUTION_PROVIDED│──▶│ IN_PROGRESS │─────────┤
└──────────┘   └──────────────────┘   └─────────────┘         │
     │                  │                    │                 │
     │                  │                    ▼                 │
     │                  │              ┌───────────┐          │
     │                  │              │ COMPLETED │          │
     │                  │              └───────────┘          │
     │                  │                                      │
     │                  ├─────────────────┐                   │
     │                  │                 │                    │
     │                  ▼                 ▼                    │
     │           ┌───────────┐     ┌──────────┐               │
     └──────────▶│ ABANDONED │     │ BLOCKED  │               │
                 └───────────┘     └──────────┘               │
                        │                │                     │
                        ▼                ▼                     │
                 ┌───────────┐     ┌──────────┐               │
                 │  DEFERRED │◀────│ (Retry)  │───────────────┘
                 └───────────┘     └──────────┘

State Definitions

StateTriggerNext Actions
INITIATEDTask request receivedAssess complexity, provide solution
SOLUTION_PROVIDEDSolution givenWait for execution signal
IN_PROGRESSUser confirms workingMonitor for completion
COMPLETEDUser confirms doneLog success, update streak
ABANDONEDNo response + context switchLog pattern, intervention
BLOCKEDExternal dependencyIdentify blocker, schedule retry
DEFERREDConscious postponementSet reminder, capture reason

Task Metadata

Every task tracks:

task = {
    "id": "task_20251219_001",
    "description": "Review Q4 insurance renewals",
    "state": "INITIATED",
    "domain": "BUSINESS",  # BUSINESS | MICHAEL | FAMILY | PERSONAL
    "complexity": 6,       # 1-10 scale
    "clarity": 8,          # 1-10 scale (how clear were instructions)
    "estimated_minutes": 45,
    "actual_minutes": None,
    "initiated_at": "2025-12-19T10:30:00Z",
    "solution_provided_at": None,
    "completed_at": None,
    "abandonment_count": 0,
    "intervention_level": 0
}

Abandonment Detection Triggers

Monitor for these patterns:

  1. Context Switch - New topic without closing current task
  2. Session End - Chat ends with task in SOLUTION_PROVIDED or IN_PROGRESS
  3. Time Decay - >30 minutes since solution with no update
  4. Topic Drift - User asks unrelated question while task pending
Detection Algorithm
def check_abandonment(task, current_message, elapsed_minutes):
    if task["state"] not in ["SOLUTION_PROVIDED", "IN_PROGRESS"]:
        return False
    
    # Time-based trigger
    if elapsed_minutes > 30:
        return True
    
    # Context switch detection
    if not is_related_to_task(current_message, task):
        return True
    
    return False

Intervention Protocol

Level 1: Gentle Check (0-30 minutes)
📌 Quick check: [task description] - still on it?
Level 2: Pattern Observation (30-60 minutes)
🔄 I notice [task] from earlier. This is pattern #[N] this week.
Continue or consciously defer?
Level 3: Direct Accountability (>60 minutes)
⚠️ ACCOUNTABILITY: [task] started [time] ago.
Current status? Be honest - no judgment, just facts.

ADHD Intervention Strategies

When abandonment detected, apply these techniques:

Micro-Commitment

Break to smallest possible action:

Just step 1? [tiny specific action] That's it.
Nothing else required right now.
Body Doubling

Create virtual co-working presence:

Let's do this together. You: [specific action]
Me: ⏱️ Waiting... (I'll check back in 5 minutes)
Chunking

Focus tunnel vision:

Step 1 only. Confirm when done.
Don't think about step 2 yet.
Energy Matching

Align with daily patterns:

It's [time] - your [peak/dip] period.
[Suggest appropriate task complexity]

Workflow: Task Creation

Trigger: Any request that implies work to be done

  1. Extract task description from message
  2. Assess complexity (1-10) based on:
    • Number of steps
    • External dependencies
    • Decision-making required
    • Time estimate
  3. Assess clarity (1-10) based on:
    • Specificity of request
    • Known vs unknown elements
    • Ambiguity level
  4. Assign domain (BUSINESS/MICHAEL/FAMILY/PERSONAL)
  5. Set state to INITIATED
  6. Provide solution → state to SOLUTION_PROVIDED

Workflow: State Transitions

INITIATED → SOLUTION_PROVIDED
# After providing solution
task["state"] = "SOLUTION_PROVIDED"
task["solution_provided_at"] = now()
SOLUTION_PROVIDED → IN_PROGRESS

Trigger: User says "working on it", "starting now", "doing this"

task["state"] = "IN_PROGRESS"
IN_PROGRESS → COMPLETED

Trigger: User says "done", "finished", "completed"

task["state"] = "COMPLETED"
task["completed_at"] = now()
task["actual_minutes"] = elapsed_time()
update_streak()
Any → ABANDONED

Trigger: Abandonment detection + no response to intervention

task["state"] = "ABANDONED"
task["abandonment_count"] += 1
log_pattern()
Any → BLOCKED

Trigger: User identifies external dependency

task["state"] = "BLOCKED"
task["blocker"] = identified_dependency
schedule_retry()
Any → DEFERRED

Trigger: Conscious postponement with reason

task["state"] = "DEFERRED"
task["defer_reason"] = reason
task["defer_until"] = scheduled_time

Streak Tracking

Maintain completion streak for motivation:

streak = {
    "current": 5,  # Days with at least 1 completion
    "longest": 12,
    "total_completions": 47,
    "abandonment_rate": 0.23  # 23% tasks abandoned
}

# On completion
"✅ Done. Streak: 5 days. Total: 47 tasks."

# On first task of day
"Day 6 starts now. Keep the streak alive."

Domain-Specific Behavior

BUSINESS (Everest Capital)
  • Higher urgency weighting
  • Deadline awareness
  • Revenue impact assessment
MICHAEL (D1 Swimming)
  • Connect to swim schedule
  • Nutrition/training alignment
  • Recruiting deadlines
FAMILY (Orthodox Observance)
  • Shabbat/holiday awareness
  • No work pressure evenings
  • Family event priority
PERSONAL (Health/Learning)
  • Energy level consideration
  • Learning capture integration
  • Health metric correlation

LangGraph Integration

State Schema
class TaskState(TypedDict):
    task_id: str
    description: str
    state: str
    domain: str
    complexity: int
    clarity: int
    elapsed_minutes: int
    intervention_level: int
    streak_current: int
Nodes
  • create_task - Initialize from user message
  • provide_solution - Generate response, update state
  • check_abandonment - Periodic abandonment check
  • intervene - Apply intervention strategy
  • complete_task - Log completion, update streak

Scripts

  • scripts/task_state_machine.py - State transition logic
  • scripts/abandonment_detector.py - Pattern detection
  • scripts/streak_calculator.py - Streak maintenance

References

  • references/intervention_templates.md - Full intervention scripts
  • references/adhd_patterns.md - Common patterns and responses

Comments

Markdown supported