Create Task Specification Skill

Purpose

Create hyper-detailed task specifications that embed ALL context needed for implementation, eliminating the need for developers to search for information during coding.

Core Capabilities:

• Requirement gathering and clarification
• Architecture context extraction with source references
• Component analysis (data models, APIs, UI)
• Sequential task breakdown with validation checkpoints
• Self-contained specifications

BMAD Pattern (Key Innovation):

• SM reads PRD + Architecture + Previous task notes
• Creates detailed task spec with embedded context
• Dev reads ONLY task spec (no architecture lookup)
• Result: Context never lost, implementation never blocked

Prerequisites

• Configuration file (.claude/config.yaml) with task settings
• Task template (.claude/templates/task-spec.md)
• Architecture documentation (optional but recommended)
• Project coding standards

---

Workflow

Step 0: Load Configuration and Validate Prerequisites

Action: Load configuration and verify prerequisites.

Execute:

python .claude/skills/bmad-commands/scripts/read_file.py \
  --path .claude/config.yaml \
  --output json

Extract: documentation.architecture, development.alwaysLoadFiles, development.taskLocation, templates.taskSpec

Verify: Task location exists, template exists, determine next task ID

Halt if: Config missing, template missing, cannot determine task ID

See: references/templates.md for configuration format and references/requirements-gathering-guide.md for requirement clarification

---

Step 1: Gather Requirements from User

Action: Clarify requirement, acceptance criteria, and priority.

Ask user:

1. What needs to be implemented?

   • Feature description
   • Problem it solves
   • Primary user/beneficiary

2. What defines "done"?

   • Testable outcomes
   • Acceptance criteria (2-5 specific criteria)
   • Edge cases to consider

3. Priority and complexity?

   • Priority: P0 (Critical) | P1 (High) | P2 (Medium) | P3 (Low)
   • Complexity: Simple | Medium | Complex
   • Related epic/feature

Confirm understanding:

• Restate as user story: As a [role], I want [action], so that [benefit]
• Verify acceptance criteria
• Get approval to proceed

Halt if:

• User cannot articulate requirement clearly
• Acceptance criteria are ambiguous
• User does not approve proceeding

See: references/requirements-gathering-guide.md for detailed gathering techniques

---

Step 2: Load Architecture Context

Action: Load architecture documentation and previous task learnings.

Load files:

1. Always-required files (coding standards):

   python .claude/skills/bmad-commands/scripts/read_file.py \
     --path {always_load_file} \
     --output json
   

2. Architecture documentation:

   • Read documentation.architecture files
   • Extract: System design, patterns, conventions

3. Previous task insights:

   • Read most recent task file from taskLocation
   • Extract: Completion notes, lessons learned, patterns

Context categories: Tech stack, project structure, coding standards, testing strategy, data models, API specs, component specs, patterns

CRITICAL RULES: ONLY use information from source documents, NEVER invent details, ALWAYS cite sources, note when info missing

See: references/context-extraction-guide.md for detailed extraction strategies and references/templates.md for context format examples

---

Step 3: Analyze Relevant Components

Action: Identify affected components and extract specific technical details.

Identify components:

• What data models are involved?
• What API endpoints need creation/modification?
• What UI components are affected?
• What external services are used?

Extract details: Data models (schemas, validation, relationships), API specs (endpoints, request/response, auth, errors), component specs (location, props, state, styling), file locations (exact paths), constraints (performance, security, reliability, testing)

All extractions must cite: [Source: filename#section]

See: references/context-extraction-guide.md for analysis patterns and references/templates.md for extraction format examples

---

Step 4: Create Sequential Task Breakdown

Action: Break requirement into 3-15 sequential tasks with subtasks.

Task breakdown:

1. Break into 3-15 tasks:

   • Each task implements specific functionality
   • Tasks build on each other logically
   • Each task maps to one or more acceptance criteria

2. Create subtasks for each task:

   • Specific implementation steps
   • Test writing requirements
   • Validation checkpoints

3. Structure in implementation order:

   • Data layer first (models, migrations)
   • Business logic second (services, utilities)
   • API layer third (routes, controllers)
   • UI layer fourth (components, pages)
   • Integration last (E2E tests, documentation)

4. Link to acceptance criteria: Example: "Task 1: Create user model (AC: 1, 2)"

Halt if: Cannot break down coherently, >15 tasks (too complex), <3 tasks (too simple)

See: references/task-breakdown-guide.md for detailed strategies and references/templates.md for task structure examples

---

Step 5: Populate Task Specification Template

Action: Generate task file from template with all context embedded.

Load template:

python .claude/skills/bmad-commands/scripts/read_file.py \
  --path .claude/templates/task-spec.md \
  --output json

Replace placeholders: task_id, title, date, priority, status, user_story, acceptance_criteria, context, tasks, validation

Context must include: Previous insights, data models, API specs, component specs, file locations, testing requirements, constraints - ALL with [Source: filename#section] references

See: references/templates.md for placeholder examples and formats

---

Step 6: Validate Task Specification Completeness

Action: Verify all required elements are present and properly formatted.

Validation checklist:

Required Sections:

• Status (Draft)
• Objective (clear user story)
• Acceptance Criteria (2-5 specific criteria)
• Context section with embedded details
• Tasks / Subtasks (3-15 tasks)
• Implementation Record (empty placeholder)

Context Completeness:

• Previous insights referenced
• Data models included (if applicable)
• API specs included (if applicable)
• Component specs included (if applicable)
• File locations specified
• Testing requirements documented
• All context has [Source: ...] references

Task Quality:

• Tasks in logical implementation order
• Each task linked to acceptance criteria
• Subtasks include test writing steps
• Validation checkpoints included
• File paths are specific, not generic

Halt if:

• Critical sections missing
• Context lacks source references
• Tasks are ambiguous or generic
• No validation checkpoints

See: references/validation-approval-guide.md for validation details

---

Step 7: Save Task Specification and Get User Approval

Action: Save task file and present summary for approval.

Save file:

python .claude/skills/bmad-commands/scripts/write_file.py \
  --path {taskLocation}/{task_id}.md \
  --content "{task_spec}" \
  --output json

Present summary: Show task ID, objective, acceptance criteria, task breakdown, embedded context sources, and request approval.

Update status if approved: Change from "Draft" to "Approved"

See: references/templates.md for complete summary format and references/validation-approval-guide.md for approval workflow

---

Output

Return structured output with task file path, task ID, task count, status, and telemetry metrics.

See: references/templates.md for complete output format

---

Best Practices

1. Context Embedding is Critical - Implementation should never need architecture docs
2. Be Specific, Not Generic - Exact file paths, schemas, API specs
3. Source Everything - Cite [Source: filename#section] for all technical details
4. Task Granularity - 3-15 tasks (>15 too complex, <3 too simple)

See: references/task-breakdown-guide.md for detailed practices

---

Routing Guidance

Use this skill when:

• Breaking features into implementable tasks
• Creating detailed specifications for development
• Embedding architectural context for implementation
• Need clear, unambiguous work items

Always use before:

• execute-task skill (implementation)
• Task must be "Approved" status

Do not use for:

• Quick bug fixes (too much overhead)
• Urgent hotfixes (too structured)
• Exploratory coding (too constrained)

---

Reference Files

• references/requirements-gathering-guide.md - Clarifying requirements, acceptance criteria, user stories
• references/context-extraction-guide.md - Loading architecture, extracting technical details, sourcing
• references/task-breakdown-guide.md - Creating sequential tasks, granularity, validation checkpoints
• references/validation-approval-guide.md - Completeness validation, saving, approval workflow
• references/templates.md - Task spec template, output format, placeholder examples, configuration format

---

Using This Skill

Invoked with requirement input (feature description) and optional priority (P0-P3)

---

Part of BMAD Enhanced Planning Suite