AWS Kiro Course: Spec Driven Agentic AI Coding - 2026

Section overview: Kiro IDE installation and plan configuration, first contact with Kiro chat, other AI features, plus optional lectures on VS Code basics and agentic AI fundamentals.

Downloading Kiro IDE from kiro.dev/downloads on Windows, macOS and Linux, installing Kiro CLI via a curl command, verifying with kiro --version, and signing in with AWS Builder ID, GitHub, Google or AWS IAM Identity Center. Covers the free plan versus the paid Kiro Pro plan.

Kiro IDE interface walkthrough: chat view, Explorer sidebar, terminal panel, model selector with per-model credit costs, context window usage, attaching current and open files as context, the agent selector with default, bug fix and plan agents, and the autopilot toggle for auto-approving file edits.

Kiro AI features outside the chat view: enabling autocomplete ghost text, inline error fixing, right-click Ask Kiro on selected code, AI-generated Git commit messages in the source control tab, and the credit usage dashboard.

VS Code features inherited by Kiro: the extension marketplace and publisher trust, the command palette via Ctrl/Cmd+Shift+P or F1, user versus workspace settings and the .vscode folder, and theme and font customization.

Agentic AI fundamentals: large language models, tokens and token-based pricing, LLM use cases from reasoning and summarization to code generation, and tools as callable functions — Kiro's built-in file read, file write and web search.

Spec-driven development as the opposite of vibe coding, and its adoption beyond Kiro in GitHub Spec Kit. Notes the removed vibe/spec mode choice from early Kiro versions still seen in older tutorials.

How spec-driven development works in Kiro: requirements.md with user stories and EARS acceptance criteria in when/then/system shall form, design.md with technical architecture, and tasks.md with a dependency graph that groups tasks into parallel execution waves.

Creating the sample project used throughout the course: scaffolding a Vite, React and TypeScript app with ESLint using npm create vite@latest, and denying Kiro's suggested vanilla template command in favor of the correct one.

Hands-on spec workflow: creating .kiro/settings/permissions.yaml to allow the fsWrite and string replace actions, running the spec agent with Claude Sonnet 4.5 to generate requirements, design and tasks documents, reviewing and editing them, then executing all tasks in parallel dependency waves.

Kiro plan mode compared with spec mode: plan mode chats about implementation options such as React Router versus TanStack Router without producing artifacts, while spec mode generates the requirements, design and tasks files for complex features.

Kiro configuration scopes: the project-level .kiro folder holding agents, hooks, settings, skills and specs, versus the global .kiro folder in the home directory. Workspace settings override user settings and can be committed for team use.

Kiro steering files in .kiro/steering: reducing prompt repetition and enforcing code standards across a team. Covers generating project steering documentation — product.md, structure.md and tech.md — and the default always-included inclusion mode.

Custom steering files with conditional inclusion: fileMatch patterns targeting TSX files in a components folder to enforce upper snake case constants, and manual inclusion for a JSDoc comment standard added to context on demand.

AI skills as reusable instruction packages: SKILL.md markdown files with optional scripts in JavaScript or Python, progressively loaded into context when the model decides they are relevant. Originated with Claude and adopted by GitHub Copilot and ChatGPT.

Building a Kiro skill from scratch in .kiro/skills following the Open Agent Skills standard: a SKILL.md file whose name and description drive invocation, plus a referenced template.md resource that shapes the skill response.

Skill with executable code: a count-components skill combining SKILL.md, a Node.js script in a scripts folder, and template.md. Covers JSON.stringify console output for predictable parsing, saving tokens over asking Kiro directly, and slash-command invocation.

Section overview: Model Context Protocol as a cross-assistant standard connecting Kiro to external servers and APIs, and Kiro Powers as a Kiro-only wrapper combining MCP servers, skills and configuration.

Model Context Protocol defined: an open standard for connecting AI applications to external systems, replacing per-client tool implementations for ChatGPT, Claude Code and GitHub Copilot with one standardized set — the USB-C port analogy for AI.

Installing MCP servers in Kiro: editing mcp.json at workspace or user scope, npx-based TypeScript servers versus uvx-based Python servers and their runtime requirements, and configuring the Airbnb MCP server to run its search and listing details tools.

Installing the official AWS MCP server in Kiro: adding a remote MCP server by URL instead of a command in mcp.json, authenticating through a signed-in AWS console session, and querying up-to-date official AWS documentation from chat.

Kiro Powers as a fix for MCP context bloat, loading only the tools relevant to a prompt. Covers the Powers tab, the official power library including Postman API testing, installing the AWS CDK and CloudFormation power, and keyword-based activation.

Building a custom Kiro power: the powers folder structure with plugin.json, using the Build a Power power to scaffold a code review helper, importing a power from a folder, and fixing an unsupported or missing plugin.json schema.

Section overview: custom agents and hooks in Kiro, two mechanisms for extending and constraining agent capabilities — scoped tool access and model selection for agents, event-triggered automation for hooks.

Kiro custom agents in .kiro/agents: markdown definitions with name, description, model, tools and permission fields for pre-approving tools and limiting tool access. Notes the JSON agent format required by Kiro CLI and the known IDE versus CLI format mismatch.

Creating Kiro agent hooks with the guided form: file, tool, prompt and lifecycle triggers. Builds a preToolUse hook that blocks airbnb_search on a keyword, and a promptSubmit hook running a Node.js script that appends every prompt to a JSON file.

AWS Kiro hands-on project overview: a pizza web store REST API implemented in .NET, Java, JavaScript/TypeScript and Python, a Node.js HTML UI, and prerequisites for SQL database integration with spec-driven development, MCP and agent skills.

Kiro codebase analysis and steering file generation in the .kiro/steering folder (product, structure, tech stack). Covers AI model selection in Kiro: Claude Sonnet 4.5 for low-cost prompts, Claude Opus with high effort for project steering rules, and large-repository context handling.

Vibe coding in Kiro without spec-driven development: adding a DELETE order endpoint to a TypeScript Node.js REST API, OpenAPI specification update, TypeScript compilation checks and automatic agent verification with curl commands.

MySQL database setup for replacing an in-memory database: MySQL installation check, MySQL CLI commands (SHOW DATABASES, CREATE DATABASE, port 3306), and SQLTools MySQL extension connection in Kiro and VS Code.

Kiro spec file optimization: adding spec generation rules to structure.md steering files to keep requirements.md, design.md and tasks.md minimal, limit acceptance criteria, skip diagrams and test plans, and reduce token usage and generation time.

Kiro spec agent workflow for a MySQL database migration: generating requirements.md, design.md and tasks.md, mysql2 driver and connection pool design, schema creation tasks, Run all tasks execution, and credit and time costs.

Debugging a Kiro-generated MySQL integration: TypeScript 7 tsconfig.json compiler options, @types/node, database credentials and environment variables, SQLTools table inspection, and end-to-end testing through the Node.js UI with persisted orders.

MySQL MCP server setup in Kiro: evaluating MCP servers on GitHub, npx and Node.js requirements, workspace mcp.json configuration, .env file credentials (MYSQL_PASS), read-only mysql query tool, and natural language database queries.

Kiro agent skills: generating a Repo Coherence Audit skill that checks code, documentation, steering files, specs and configuration for drift, invoking skills with slash commands, cross-model code review, and AI-suggested project skills.

Section overview: Kiro CLI as a terminal alternative to the Kiro IDE, covering developer preference, running long or complex agent tasks with broader system permissions, and the security considerations involved.

Installing Kiro CLI from the kiro.dev documentation: a curl install command on macOS and Linux, a downloadable installer on Windows, launching with kiro-cli rather than kiro, and inherited authentication from an already configured Kiro IDE.

Kiro CLI in practice: slash commands including /context, /model and /mcp, selecting models such as Claude Haiku 4.5, invoking the Airbnb MCP server from the terminal, JSON-format custom agents alongside kiro-default, kiro-guide and kiro-planner, and running parallel CLI instances.