Spec-Driven Development with AI Coding Agents: Spec Kit, Kiro, and OpenSpec

Class Duration

14 hours of live training delivered over 2-3 days to accommodate your scheduling needs.

Student Prerequisites

• Professional software development experience
• Working use of at least one AI coding agent (GitHub Copilot, Claude Code, Cursor, Codex, Kiro, or Google Antigravity)
• Comfort with the command line, Git, and pull requests
• Experience writing or reviewing requirements is helpful

Target Audience

Software engineers, tech leads, architects, and the product owners and analysts who write requirements. Platform teams that want a standard way to plan and review AI-generated code will also benefit. Teams adopting the full AWS method can continue with AI-Driven Development Life Cycle (AI-DLC), and Kiro teams can go deeper in Spec-Driven Agentic Coding with Kiro.

Training Materials

Comprehensive courseware is distributed online at the start of class. All students receive a downloadable MP4 recording of the training.

Software Requirements

Git, Python 3.11 or later with uv and the Spec Kit specify CLI, Node.js 22.12 or later with the OpenSpec CLI, a GitHub account with a sample repository, and an active subscription or API key for at least one supported coding agent (GitHub Copilot, Claude Code, Cursor, Codex CLI, Kiro, or Antigravity).

Spec-Driven Development Foundations

• Where prompt-first workflows break down
• Requirements lost in chat history
• Specs as versioned contracts
• Spec-first, spec-anchored, and spec-as-source
• Relation to TDD, BDD, and ADRs
• Matching ceremony to the change

Writing Specs and Acceptance Criteria

• Separating what from how
• User stories and functional requirements
• Measurable success criteria
• Constraints and out-of-scope boundaries
• Explicit open questions
• EARS requirement patterns
• Event-driven and state-driven requirements
• Unwanted behavior and optional features
• WHEN/THEN scenarios
• Mapping criteria to tests
• Vague and oversized specs

The Project Constitution

• Non-negotiable project principles
• Quality, testing, and security rules
• Creating a constitution with /speckit.constitution
• Versioning and amendments
• The Sync Impact Report
• Constitution Check gates
• Grounding a constitution in an existing repository
• Constitutions vs. AGENTS.md and steering

GitHub Spec Kit Setup and Project Structure

• Installing the specify CLI
• specify init --integration
• Adding Spec Kit with --here
• The .specify/ directory
• Feature directories under specs/
• Supported agent integrations
• Agent-specific command spellings
• Switching agents with specify integration switch
• Extensions, presets, and template overrides
• The opt-in git extension

The Spec Kit Workflow: Specify Through Analyze

• Writing the spec with /speckit.specify
• [NEEDS CLARIFICATION] markers
• The built-in requirements checklist
• Resolving gaps with /speckit.clarify
• Planning with /speckit.plan
• Research, data model, and contracts
• Custom checklists with /speckit.checklist
• Phased task lists with /speckit.tasks
• Parallel task markers
• Consistency checks with /speckit.analyze

Implementation and Convergence

• Running tasks with /speckit.implement
• Correcting an agent that drifts
• Gap analysis with /speckit.converge
• The implement-converge loop
• Splitting large features into specs
• Workflows with specify workflow run
• The bug and github extensions

Keeping Specs Current

• Changing an existing feature
• New or revised specs
• Living specs and other persistence models
• How specs and code drift
• Catching drift in review
• Preserving decision records
• What stays in version control

Kiro Specs Compared

• Feature specs in .kiro/specs/
• Requirements in EARS notation
• Requirements-first vs. design-first
• Quick Spec and Plan sessions
• Bugfix specs
• Property-based correctness checks
• Running tasks in waves
• Steering vs. a Spec Kit constitution
• Spec Kit in Kiro CLI

OpenSpec for Changing Existing Code

• A brownfield-first model
• Current specs in openspec/specs/
• Change folders in openspec/changes/
• Proposals, designs, and tasks
• Delta specs
• Added, modified, removed, and renamed requirements
• SHALL requirements and WHEN/THEN scenarios
• Propose, apply, and archive
• The /openspec-propose skill
• Merging deltas with openspec archive
• openspec validate --strict

BMAD Planning and Spec-as-Source with Tessl

• BMAD Method agent personas
• Analyst, PM, and architect agents
• From PRD to architecture
• Epics and stories
• Sizing the planning track
• Feeding BMAD output to implementation
• Tessl and spec-as-source
• Tessl's skills registry
• Limits of spec-as-source

Choosing an Approach and Integrating Agents

• Comparing the four approaches
• Greenfield vs. brownfield fit
• Tool lock-in
• Choosing per type of change
• Spec Kit in Copilot and Claude Code
• Other agent integrations
• Markdown specs as the portable layer
• Specs within AWS AI-DLC

Review Discipline and CI Integration

• Review gates at every phase
• Who reviews specs, plans, and code
• Tracing changes to requirement IDs
• Spec diffs in pull requests
• Branch conventions for spec work
• Non-interactive runs in CI
• Structural spec validation
• Flagging behavior changes without spec updates
• Rework and convergence metrics