Skip to content
Learn Kiro.

Reference · Chapter 10 of 16

Cheat Sheet

Kiro on one page: concepts, files, commands, workflows, and the mistakes that bite, in scannable card form.

All levels last reviewed 2026-06-11

◎ Learning objective

Find any core Kiro fact (concept, file path, command, or workflow) in under ten seconds.

Bookmark this page. Everything links back to the full chapter when you need depth.

The mental model

Specswhat gets built (plan first)
Steeringwhat the agent knows
Hookswhen it acts on its own
MCP / Powerswhat it can reach
Agents / Skillswho it is / what it can do
IDE / CLI / Webwhere it all happens

Three surfaces

IDEdesktop app, VS Code-style, from kiro.dev
CLIkiro-cli in any terminal
Webapp.kiro.dev → agents make PRs (preview, paid)

Spec files

requirements.mdwhat: user stories, EARS criteria
design.mdhow: architecture, sequences
tasks.mdsteps: runs in concurrent waves
bugfix.mdbug analysis (bugfix specs)

EARS patterns

  • WHEN <event> THE SYSTEM SHALL <behavior> for normal paths
  • IF <error> THEN THE SYSTEM SHALL <handling> for error paths
  • WHILE <state> THE SYSTEM SHALL <constraint> for ongoing states
  • No IF/THEN lines = requirements not finished

Files & folders

.kiro/steering/project rules (product/tech/structure.md)
.kiro/settings/mcp.jsonworkspace MCP servers
.kiro/agents/custom agent JSON
.kiro/skills/team skills (SKILL.md)
~/.kiro/…same, but global; workspace wins

CLI essentials

install (macOS/Linux)curl -fsSL https://cli.kiro.dev/install | bash
install (Windows 11)irm 'https://cli.kiro.dev/install.ps1' | iex
startkiro-cli
as personakiro-cli --agent reviewer
resumekiro-cli chat --resume / --resume-picker
direct shell!command (no AI involved)
headless (CI)kiro-cli chat --no-interactive "…" + KIRO_API_KEY

Slash commands

/context showwhat’s loaded + token cost
/context add "glob"include files
/chat save|load <path>export / import session
/chat new [prompt]fresh session
/agent create|swapmake / switch personas
/editor /replycompose in $EDITOR

Core workflows

  • Feature: spec → review requirements → review design → execute tasks → verify
  • Bug: bugfix spec → regression test → fix → verify
  • Refactor: steering current? → Supervised mode → small diffs → tests green
  • Chore: make it a hook, scoped narrowly

Steering rules of thumb

  • Three foundations: product.md, tech.md, structure.md
  • One domain per file, with the why
  • Modes: always · fileMatch · manual · auto
  • Never paste secrets into steering
  • Commit workspace steering: it’s team law

MCP config keys

  • Local: command, args, env (${VAR} expands)
  • Remote: url, headers, oauth
  • Control: disabled, disabledTools
  • autoApprove: [] default; "*" = sudo, beware
  • Debug via “Kiro - MCP Logs” channel

Modes & safety

  • Supervised: approve every hunk (learning, risky work)
  • Autopilot: end-to-end autonomy (trusted, bounded work)
  • Checkpoints: roll the session back anytime
  • Golden rule: nothing ships unreviewed

Top mistakes

  • Vague prompts → confident wrong code
  • Accepting diffs unread
  • autoApprove: "*" on untrusted servers
  • Hooks that trigger themselves
  • Steering files full of secrets
  • Trusting 2025 blog posts (check history)

Official links

☰ Chapter summary

  • All Kiro state lives under .kiro/ (workspace) and ~/.kiro/ (global): steering, settings, agents, skills.
  • Spec = requirements.md + design.md + tasks.md; bugfix specs use bugfix.md.
  • EARS: WHEN for normal behavior, IF/THEN for errors, WHILE for ongoing states.
  • CLI: kiro-cli to start; /context, /chat, /agent are the three command families that matter.
  • Safety dial: Supervised for risky work, Autopilot for trusted work, checkpoints either way.

All chapter summaries are collected on the revision page.