Task Intake

Scenario Demos
← Back to Hub
Key Files
backend/src/services/task-intake/TaskScopeClassifier.ts backend/src/services/task-intake/TaskIntakeOrchestrator.ts backend/src/services/task-intake/handlers/*.ts
Scenario A: Simple Task SMALL
Single action tasks under 2 hours. No decomposition needed - direct task creation with optional calendar scheduling.
Call mom
👤
User Input
"Call mom"
→
🧠
TaskScopeClassifier
LLM analysis
→
⚡
SmallTaskHandler
Direct creation
📅
CalendarService
Find slots
→
✅
Task Created
With proposed time
🎯
Classification Result
TaskScopeClassifier output
SMALL
Scope
lifestyle
Domain
15m
Est. Duration
0.95
Confidence
Classification Confidence 95%
📅
Proposed Time Slots
2:00 PM - 2:15 PM Today Recommended
4:30 PM - 4:45 PM Today
10:00 AM - 10:15 AM Tomorrow
⚡ Fast Path
Small tasks skip decomposition entirely. The SmallTaskHandler creates the task directly, estimates duration using LLM, and queries the CalendarService for available slots. Total processing time: ~800ms.
Scenario B: Study/Learning MEDIUM
Multi-step learning goals (2-8 hours). Decomposed into Epic with Features and Stories for structured study sessions.
Study for calculus exam
👤
User Input
"Study for calculus exam"
→
🧠
TaskScopeClassifier
Detects MEDIUM
→
📋
MediumTaskHandler
Epic decomposition
🤖
EpicAgent
Generate hierarchy
→
✅
DecompositionEvaluator
Quality check
🎯
Classification Result
TaskScopeClassifier output
MEDIUM
Scope
learning
Domain
6h
Est. Duration
0.88
Confidence
🌳
Decomposition Output
Epic Study for Calculus Exam ~6h total
Feature Review Core Concepts
Story Study limits and continuity 45m
Story Review derivatives rules 45m
Story Practice integration techniques 60m
Feature Practice Problems
Story Complete problem set A 60m
Story Review incorrect answers 30m
Feature Final Review
Story Take practice exam 90m
1
Epic
3
Features
6
Stories
6h
Total Est.
📚 Learning Domain Detection
The classifier detects "study", "exam", "learn" keywords and routes to the learning domain. This influences how stories are structured - shorter sessions with review breaks aligned with spaced repetition principles.
Scenario C: Event Planning MEDIUM
Multi-phase event planning (4-6 hours of work). Decomposed into discrete planning phases with actionable tasks.
Plan my friend's birthday party
👤
User Input
"Plan birthday party"
→
🧠
TaskScopeClassifier
Domain: event
→
🎉
MediumTaskHandler
Event planning
🎉
Classification Result
TaskScopeClassifier output
MEDIUM
Scope
event
Domain
5h
Est. Duration
0.91
Confidence
🎉
Event Planning Breakdown
Epic Plan Birthday Party ~5h total
Feature Core Planning
Story Pick date & confirm guest list 45m
Story Choose venue & make reservation 30m
Feature Logistics
Story Plan menu & create shopping list 60m
Story Order decorations 30m
Story Arrange entertainment/music 30m
Feature Communications
Story Send invitations 30m
Story Create day-of timeline 15m
📆 Calendar Integration
Event planning tasks are scheduled with dependency awareness. "Send invitations" depends on "Pick date" completing first. The system proposes calendar slots that respect these dependencies.
Scenario D: Large Project LARGE
Complex multi-week projects. Full 4-level hierarchy: Project → Epics → Features → Stories. Uses ProjectSplitterAgent for initial breakdown, then EpicAgent for each epic.
Start an Airbnb business
👤
User Input
"Start an Airbnb business"
→
🧠
TaskScopeClassifier
Detects LARGE
→
🏗
LargeProjectHandler
Full decomposition
📊
ProjectSplitterAgent
Define epics
→
🤖
EpicAgent (x3)
Sequential expansion
→
✅
Quality Evaluation
Validate structure
💼
Classification Result
TaskScopeClassifier output
LARGE
Scope
financial
Domain
65h+
Est. Duration
0.92
Confidence
🏗
Full Project Hierarchy
Project Start an Airbnb Business ~65h total
Epic 1 Research & Planning ~20h
Feature Market Research
Story Research local rental market 2h
Story Analyze competitor properties 1.5h
Feature Business Plan
Epic 2 Legal & Financial Setup ~15h
Feature Legal Registration
Feature Financial Setup
Epic 3 Property Preparation ~30h
Feature Renovations & Cleaning
Feature Photography & Listing
1
Project
3
Epics
12
Features
35+
Stories
🔄 Sequential Epic Expansion
Large projects decompose sequentially for reliability. ProjectSplitterAgent first identifies major epics, then EpicAgent expands each one individually. This prevents context overflow and allows for quality checks between expansions.
Scenario E: Recurring Habit RECURRING
Long-term habits with daily/weekly recurrence. Creates initial stories and tracks streaks over time. Stories regenerate on schedule.
Learn Spanish daily
👤
User Input
"Learn Spanish daily"
→
🧠
TaskScopeClassifier
Detects recurrence
→
🔄
RecurrenceHandler
Create habit
🇪🇸
Classification Result
Recurrence pattern detected
RECURRING
Type
learning
Domain
30m/day
Session
0.89
Confidence
🔄
Weekly Schedule
MON
7:00 AM
TUE
7:00 AM
WED
7:00 AM
THU
7:00 AM
FRI
7:00 AM
SAT
9:00 AM
SUN
9:00 AM
📚
This Week's Stories (Auto-Generated)
Habit Learn Spanish Daily, 30m
Mon Lesson 1: Basic Greetings 30m
Tue Lesson 2: Common Phrases 30m
Wed Lesson 3: Numbers 1-20 30m
Thu Lesson 4: Introducing Yourself 30m
Fri Review: Week 1 Quiz 30m
7
Days/Week
12
Current Streak
45
Longest Streak
3.5h
Weekly Total
🔥 Streak Tracking
Recurring habits track completion streaks via WorkItemRecurrence metadata. New stories are generated each week. Missing a day doesn't reset the habit - it just affects the streak counter. The system learns preferred times from completion patterns. 
interface WorkItemRecurrence {
  pattern: 'daily' | 'weekly' | 'monthly';
  daysOfWeek?: number[];           // 0=Sun, 6=Sat
  preferredTime?: string;         // "07:00"
  sessionDurationMinutes: number;  // 30
  streaks: {
    current: number;               // 12
    longest: number;               // 45
    missedLast7Days: number;       // 0
  };
}