Task Decomposition

Overview

Task decomposition breaks complex tasks into manageable, well-defined subtasks with clear dependencies, effort estimates, and parallelization opportunities. It covers hierarchical work breakdown structures (WBS), dependency graph construction, critical path analysis, task sizing, and identification of concurrent execution opportunities. Essential for planning multi-step implementations, project estimation, and autonomous loop task selection.

Announce at start: "I'm using the task-decomposition skill to break this work into a structured hierarchy with dependencies and estimates."

Trigger Conditions

• Complex task needs to be broken into subtasks
• Dependency mapping is needed between work items
• Effort estimation is required for planning
• Parallelization opportunities need to be identified
• /decompose command invoked
• Transition from planning skill for complex plans
• Autonomous loop needs task selection guidance

---

Phase 1: Scope Definition

Goal: Define clear boundaries for the decomposition.

1. Define the overall deliverable and acceptance criteria
2. Identify the boundaries (what is in scope, what is not)
3. Determine the decomposition granularity level
4. Identify stakeholders and their requirements
5. Establish constraints (time, resources, dependencies)

Granularity Decision Table

Context	Target Level	Typical Duration	Rationale
Autonomous loop (Ralph)	L3-L4 (Task/Subtask)	15 min - 4 hours	ONE task per loop iteration
Sprint planning	L2-L3 (Story/Task)	0.5-2 days	Sprint-sized work items
Roadmap planning	L0-L1 (Epic/Feature)	2-8 weeks	High-level milestone tracking
Bug fix	L3-L4 (Task/Subtask)	15 min - 2 hours	Focused, specific fixes

Granularity Levels

Level	Name	Typical Duration	Example
L0	Epic	2-8 weeks	"User authentication system"
L1	Feature	2-5 days	"OAuth2 login flow"
L2	Story	0.5-2 days	"Google OAuth provider integration"
L3	Task	1-4 hours	"Implement Google callback handler"
L4	Subtask	15-60 minutes	"Parse OAuth token response"

STOP — Do NOT proceed to Phase 2 until:

• Deliverable and acceptance criteria are defined
• Scope boundaries are clear (in/out)
• Target granularity level is chosen
• Constraints are identified

---

Phase 2: Hierarchical Breakdown

Goal: Decompose the work into a tree structure meeting the INVEST criteria.

1. Identify top-level work streams (epics or major components)
2. Break each work stream into features or milestones
3. Decompose features into implementable tasks
4. Apply the "2-hour rule" — no task should exceed 2 hours of focused work
5. Ensure each task has a clear definition of done
6. Verify MECE (Mutually Exclusive, Collectively Exhaustive) coverage

The INVEST Criteria for Tasks

Criterion	Question	Bad Example	Good Example
Independent	Can this be done without waiting for others?	"Implement auth after DB is ready"	"Implement auth with mock DB"
Negotiable	Is the approach flexible?	"Use Redis for caching"	"Add caching layer for user sessions"
Valuable	Does completing this deliver value?	"Set up folder structure"	"Create user registration endpoint"
Estimable	Can you estimate the effort?	"Improve performance"	"Add database index for user lookup query"
Small	Can one person finish it in < 2 hours?	"Build the dashboard"	"Create dashboard chart component for revenue data"
Testable	Can you verify it is done?	"Make it better"	"Response time < 200ms for /api/users"

MECE Verification

Check	Question
Mutually Exclusive	Does any task overlap with another? (Should not)
Collectively Exhaustive	Do all tasks together cover the full deliverable? (Should)
No orphans	Does every task contribute to the deliverable?
No gaps	Is there any work needed that has no task?

STOP — Do NOT proceed to Phase 3 until:

• All work streams are identified
• Tasks meet INVEST criteria
• MECE coverage is verified
• No task exceeds 2 hours

---

Phase 3: Dependency Mapping

Goal: Build a directed acyclic graph (DAG) of task dependencies.

1. Identify input/output dependencies between tasks
2. Classify dependency types
3. Build a directed acyclic graph (DAG)
4. Identify the critical path (longest dependency chain)
5. Flag circular dependencies as errors to resolve
6. Mark external dependencies (API access, approvals, third-party)

Dependency Types

Type	Symbol	Meaning	Example
Finish-to-Start (FS)	A -> B	B cannot start until A finishes	"Deploy" after "Build passes"
Start-to-Start (SS)	A => B	B can start when A starts	"Write docs" when "Write code" starts
Finish-to-Finish (FF)	A =>> B	B cannot finish until A finishes	"Testing" finishes after "Development"
Start-to-Finish (SF)	A ~> B	B cannot finish until A starts	Rare — shift handoff scenarios

Dependency Notation Format

Task 1: Set up database schema
Task 2: Create data access layer         [depends: 1]
Task 3: Implement API endpoints          [depends: 2]
Task 4: Write unit tests for DAL         [depends: 2]
Task 5: Write API integration tests      [depends: 3, 4]
Task 6: Create frontend components       [depends: none]
Task 7: Connect frontend to API          [depends: 3, 6]
Task 8: End-to-end testing               [depends: 5, 7]

Parallel tracks:
  Track A: 1 -> 2 -> 3 -> 5 -> 8
  Track B: 1 -> 2 -> 4 -> 5 -> 8
  Track C: 6 -> 7 -> 8
  Critical path: 1 -> 2 -> 3 -> 7 -> 8

Circular Dependency Resolution

Detection	Resolution
A depends on B, B depends on A	Break into smaller tasks that remove the cycle
A depends on B's interface, B depends on A's interface	Define interfaces first as a separate task
Tight coupling between components	Introduce an abstraction layer task

STOP — Do NOT proceed to Phase 4 until:

• All dependencies are mapped
• No circular dependencies exist
• Critical path is identified
• External dependencies are flagged

---

Phase 4: Parallelization Planning

Goal: Identify independent task clusters that can run concurrently.

1. Identify independent task clusters (no dependencies between them)
2. Group tasks by resource type (read, write, build, test)
3. Determine maximum parallelism based on resource constraints
4. Sequence tasks within each parallel track
5. Plan synchronization points (merge gates)

Resource-Based Parallelism Limits

Resource Type	Max Parallel	Rationale
Code reading / analysis	Unlimited	No side effects
File creation / editing	3-5	Avoid merge conflicts
Build / compile	1	Resource contention
Test execution	1-2	Shared state, ports
Database migrations	1	Sequential by nature
Documentation	Unlimited	Independent files

Parallelization Pattern Decision Table

Pattern	When to Use	Example
Independent Clusters	Work streams with no shared state	Backend, Frontend, Infra
By Layer	Layers touch different files	API, Service, Data
By Feature Area	Independent vertical slices	Auth, Profile, Billing
By Task Type	Code, tests, docs touch different files	Implement, Test, Document

Synchronization Points

  +------+     +------+     +------+
  |Task A|     |Task B|     |Task C|
  +--+---+     +--+---+     +--+---+
     |            |            |
     v            v            v
  ======================================
     SYNC GATE: All Complete
     Verify: no conflicts, tests pass
  ======================================
                 |
                 v
          +----------+
          |Next Phase|
          +----------+

STOP — Do NOT proceed to Phase 5 until:

• Independent clusters are identified
• Resource constraints are considered
• Synchronization points are defined
• Maximum parallelism is determined

---

Phase 5: Estimation and Prioritization

Goal: Estimate effort for each task and create an execution timeline.

T-Shirt Sizing to Hours

Size	Hours	Confidence	Example
XS	0.5-1h	High	Rename a variable, fix a typo
S	1-2h	High	Add a simple endpoint, write a test
M	2-4h	Medium	Implement a feature with known pattern
L	4-8h	Low	New feature with research needed
XL	8h+	Very Low	Must be decomposed further

Estimation Heuristics

Scenario	Multiplier	Rationale
Known pattern	1.2x base estimate	20% buffer for unknowns
Unknown pattern	2x base estimate	Add research spike task first
Integration work	1.5x sum of components	Integration is harder than parts
First-time technology	3x "if I knew how" estimate	Learning curve
Bug fixes	Time-box 2h investigation	Then re-estimate

Three-Point Estimation

Expected = (Optimistic + 4 * Most Likely + Pessimistic) / 6

Example:
  Optimistic:  2 hours (everything goes smoothly)
  Most Likely: 4 hours (normal development pace)
  Pessimistic: 10 hours (major unexpected issues)
  Expected:    (2 + 16 + 10) / 6 = 4.7 hours

Prioritization Decision Table

Factor	Weight	How to Evaluate
On critical path	Highest	Delays here delay everything
Blocks other tasks	High	Unblocking multiplies throughput
Business value	High	User-facing impact
Risk reduction	Medium	De-risks unknowns early
Quick win	Medium	Low effort, high morale
Nice to have	Low	Only after core work is done

---

Work Breakdown Structure Template

# WBS: [Project Name]

## 1. [Work Stream A]
### 1.1 [Feature]
- [ ] 1.1.1 [Task] — Est: 2h — Deps: none — Priority: P0
- [ ] 1.1.2 [Task] — Est: 1h — Deps: 1.1.1 — Priority: P0
- [ ] 1.1.3 [Task] — Est: 3h — Deps: 1.1.1 — Priority: P1

### 1.2 [Feature]
- [ ] 1.2.1 [Task] — Est: 1h — Deps: none — Priority: P0
- [ ] 1.2.2 [Task] — Est: 2h — Deps: 1.2.1, 1.1.2 — Priority: P1

## 2. [Work Stream B]
### 2.1 [Feature]
- [ ] 2.1.1 [Task] — Est: 1h — Deps: none — Priority: P0
- [ ] 2.1.2 [Task] — Est: 4h — Deps: 2.1.1 — Priority: P0

## Summary
- Total tasks: N
- Estimated total effort: Xh
- Critical path duration: Yh
- Max parallelism: Z tracks
- External dependencies: [list]

---

Critical Path Analysis

How to Find the Critical Path

1. List all tasks with durations and dependencies
2. Forward pass: calculate earliest start (ES) and earliest finish (EF)
3. Backward pass: calculate latest start (LS) and latest finish (LF)
4. Float = LS - ES (tasks with zero float are on the critical path)
5. The critical path is the longest chain through the dependency graph

Optimization Strategies

Strategy	When	Effect	Risk
Parallelize	Independent tasks on critical path	Reduces calendar time	Low
Fast-track	Overlap sequential tasks	Reduces duration	Medium — may cause rework
Crash	Add resources to critical tasks	Reduces duration	Medium — coordination cost
Scope reduction	Remove non-essential tasks	Reduces total work	Low — if non-essential is correct
Spike first	Unknown tasks blocking the path	De-risks estimates	Low

---

Anti-Patterns / Common Mistakes

Anti-Pattern	Why It Fails	Correct Approach
Tasks too large to estimate	"Build the backend" is not a task	Decompose until estimable (< 2h)
Missing dependencies	Surface during implementation, cause rework	Map ALL dependencies upfront
Circular dependencies	Indicate unclear architecture	Break the cycle with interface tasks
All tasks sequential	No parallelism possible	Identify independent clusters
Estimation without decomposition	Guessing at L0 level, always wrong	Estimate at leaf level, sum up
Ignoring external dependencies	Block progress unexpectedly	Flag and plan for them
Over-decomposition	Noise, not signal (50 subtasks for a form)	Stop at meaningful, testable units
Ignoring critical path	Priorities work on non-critical tasks	Always prioritize critical path
Not re-estimating	Estimates drift as you learn	Re-estimate after each phase
Tasks without acceptance criteria	Cannot verify completion	Every task has definition of done

---

Anti-Rationalization Guards

If you catch yourself thinking:

• "I can estimate the whole thing without breaking it down..." — No. Decompose first.
• "Dependencies are obvious, I don't need to map them..." — Map them. Hidden dependencies cause failures.
• "This task is fine at 8 hours..." — Decompose it. XL tasks must be broken down.
• "The critical path doesn't matter for small projects..." — It does. It tells you what to prioritize.

---

Subagent Dispatch Opportunities

Task Pattern	Dispatch To	When
Parallelizable leaf tasks identified during decomposition	Parallel subagents via Agent tool	When tasks have no shared dependencies
Architecture analysis of task boundaries	planner agent	When decomposition reveals cross-cutting concerns
Validation of decomposition completeness	spec-reviewer agent	When task tree is complete but unverified

Mark each decomposed task with a parallelizable: yes/no flag in the output table. Follow the dispatching-parallel-agents skill protocol when dispatching.

---

Integration Points

Skill	Relationship	When
planning	Upstream — provides the plan to decompose	Complex plans need WBS
task-management	Downstream — receives decomposed tasks	For execution tracking
dispatching-parallel-agents	Downstream — receives parallelizable clusters	For concurrent execution
autonomous-loop	Downstream — task selection from WBS	Ralph task selection
executing-plans	Downstream — batch creation from WBS	Plan execution
subagent-driven-development	Downstream — independent tasks for subagents	Delegated implementation
spec-writing	Complementary — specs inform decomposition	Understanding requirements

---

Concrete Examples

Example: Decomposition of "Add User Authentication"

# WBS: User Authentication

## 1. Core Auth
### 1.1 Token Management
- [ ] 1.1.1 Implement JWT generation — Est: 1h — Deps: none — P0
- [ ] 1.1.2 Implement JWT validation — Est: 1h — Deps: 1.1.1 — P0
- [ ] 1.1.3 Implement refresh token rotation — Est: 2h — Deps: 1.1.2 — P1

### 1.2 Auth Middleware
- [ ] 1.2.1 Create auth middleware — Est: 1h — Deps: 1.1.2 — P0
- [ ] 1.2.2 Add role-based access control — Est: 2h — Deps: 1.2.1 — P1

## 2. Auth Endpoints
### 2.1 Registration
- [ ] 2.1.1 POST /auth/register endpoint — Est: 1h — Deps: 1.1.1 — P0
- [ ] 2.1.2 Email validation — Est: 30m — Deps: none — P0

### 2.2 Login
- [ ] 2.2.1 POST /auth/login endpoint — Est: 1h — Deps: 1.1.1, 1.2.1 — P0
- [ ] 2.2.2 POST /auth/refresh endpoint — Est: 1h — Deps: 1.1.3 — P1

## Summary
- Total tasks: 8
- Estimated total effort: 10.5h
- Critical path: 1.1.1 -> 1.1.2 -> 1.2.1 -> 2.2.1 (4h)
- Max parallelism: 3 tracks (Token, Middleware, Endpoints)
- External dependencies: none

---

Skill Type

RIGID — Follow the decomposition phases in order. Every task must meet the INVEST criteria and have explicit dependencies, estimates, and acceptance criteria. The dependency graph and critical path analysis are mandatory for multi-day work.