Skip to content
Learn Kiro.

Reference · Chapter 61 of 63

Glossary

Every Kiro term explained in one or two sentences, in alphabetical order, each linked to the chapter that covers it in full.

All levels 12 min read last reviewed 2026-09-04

◎ Learning objective

Look up any Kiro term quickly and know which chapter explains it properly.

Kiro has a lot of vocabulary, and most of it is only confusing until someone gives you one clear sentence. This page gives you that sentence for every term, in alphabetical order, with a link to the chapter that explains it properly.

Use it as a lookup table. If you want the ideas in a teaching order instead, read Core Concepts.

Jump to: A · B · C · D · E · F · G · H · I · K · M · O · P · Q · R · S · T · V · W

A

ACP (Agent Client Protocol)

The open protocol Kiro’s surfaces use to talk to the agent. Run kiro-cli acp and another editor can drive Kiro over JSON-RPC 2.0 on stdin and stdout, which is how JetBrains and Zed use it.

Agent harness

The single agent implementation shared by every Kiro surface. Because the IDE, CLI, Web, and Mobile all use it, an improvement to the harness reaches all of them at once. See Core Concepts.

Agent Plugin format

The open, vendor-neutral packaging standard powers have used since August 2026. It lets you install a power from a local folder or a GitHub URL and makes powers portable across compatible tools.

Agent Skills standard

The open standard Kiro skills follow: a SKILL.md file with YAML front matter naming and describing the skill, plus optional scripts and templates.

Agentic development

Working with an AI that acts rather than only suggests: it plans steps, edits files, runs commands, checks results, and adjusts. Your job shifts from typing every line to directing and reviewing. See Agentic development.

AGENTS.md

A cross-tool convention file that Kiro reads as steering. Kiro finds AGENTS.md anywhere in the workspace tree, not only at the root, and these files are always included with no inclusion modes. Recipe: AGENTS.md as steering.

App Store (Kiro Crew)

Crew’s built-in catalogue of apps, which are packages that add agents, skills, MCP servers, cron jobs, dashboard pages, or backend processes. Install clones the repository and Enable registers it; third-party apps are disabled by default. See Kiro Crew.

Auto

Kiro’s model router, not a model. It combines frontier models with specialised ones plus intent detection and caching, routes to Sonnet-class or better on Free and Opus-class or better on paid tiers, never routes to experimental models, and is defined as the 1.0x credit baseline. See Choosing a Model.

Automations (Kiro Web)

Scheduled runs in Kiro Web on Pro and above. Choose hourly, daily, or a CRON expression, up to five schedules per automation in UTC, aim it at one or more repositories, and the output is a pull request. Recipe: schedule a Web Automation.

Autonomous mode (Kiro Web)

A toggle in the Kiro Web chat input, off by default. The agent runs clarification, then planning, then execution by specialised sub-agents in an isolated sandbox, and opens a pull request. It picks the model itself and can enter a “needs attention” state. See Web.

Autopilot mode

The IDE chat mode where the agent works end to end, creating files, modifying code, and running commands, with everything viewable and revertible afterwards. Its opposite is Supervised mode. Set it under Settings, Agent, Agent Autonomy.

B

Bugfix spec

A spec variant for fixing bugs, where the root-cause analysis lands in bugfix.md instead of requirements.md, followed by design and fix tasks you approve. Recipe: write a bugfix spec.

C

Capability

The unit Kiro permissions are written against: fs_read, fs_write, shell, web_fetch, web_search, mcp, subagent, skill, power, context, diagnostics, and sandbox_network, plus the meta-capabilities all, builtin, and filesystem. Recipe: permissions.yaml.

Checkpoint

An automatic snapshot created on every prompt, covering the files the agent changed with its built-in file tools. Restoring one puts the files and the agent’s context back and discards the later chat. Manual edits, formatter runs, MCP tool changes, and bash edits are not tracked. Recipe: checkpoints and rewind.

Cloud session

A Kiro session where the agent runs in a managed AWS sandbox rather than on your machine, belonging to your account rather than to one app. Start it in one surface, close the laptop, and reattach from another. See Cloud sessions.

Code OSS

The open-source base of VS Code, which the Kiro IDE is built on. That is why the layout, keybindings, and most extensions feel familiar. See IDE.

Compaction

What Kiro does when the context window fills: it summarises older history, keeping task status, file paths, decisions, and next steps while summarising tool output, old code snippets, and resolved errors. It is one-way, so history before the compaction point is gone. Recipe: compact a long session.

Configuration scope

Where a piece of Kiro configuration lives, and therefore which one wins. Global is ~/.kiro/, project is .kiro/, and some settings are set per agent; the closer scope wins. Steering, skills, and hooks merge across scopes, while permissions use deny-overrides.

Context provider

A # reference that attaches precise context to a chat message: #file, #folder, #codebase, #git diff, #terminal, #problems, #url, #code, #repository, #current, #steering, #docs, #spec, and #mcp. See the Cheat Sheet.

Context window

The maximum amount of text a model can consider at once, measured in tokens. 1M is roughly a large codebase; 200K is a substantial slice of one. See Choosing a Model.

Credit

Kiro’s unit of billing: one unit of agent work performed in response to a prompt, billed fractionally in 0.01 increments. Prompts, spec refinement, task execution, and hook execution all consume credits. See Credits and Pricing.

Credit multiplier

A per-model factor applied to credit consumption, expressed relative to Auto at 1.0x. A 2.2x model spends 2.2 credits where Auto spends 1, and a 0.1x model spends a tenth. See Choosing a Model.

Custom agent

A scoped version of the agent with its own instructions, allowed tools, and context. The CLI defines them as JSON in .kiro/agents/ or ~/.kiro/agents/; the IDE uses Markdown in the same folders. Recipe: create a custom agent.

D

Deny rule

A permission rule with the deny effect, which always wins regardless of scope and regardless of Autopilot. Deny beats ask, and ask beats allow. Recipe: permissions.yaml.

E

EARS

Easy Approach to Requirements Syntax: requirement sentences shaped like WHEN <condition> THE SYSTEM SHALL <behavior>, with IF … THEN for error cases and WHILE for ongoing states. Every line can become a test. See Specs.

F

fileMatch

A steering inclusion mode that loads a file only when matching files are involved, set with inclusion: fileMatch and a fileMatchPattern glob or list of globs in the YAML front matter.

Foundational steering files

The three steering files Kiro includes by default: product.md (what the product is and who it serves), tech.md (frameworks, libraries, constraints), and structure.md (file organisation and naming). See Steering.

G

glibc

The GNU C Library, the low-level system library most Linux desktop software links against. Its version is set by your distribution, and the Kiro IDE needs 2.39 or newer. See Troubleshooting.

Governance ceiling

Kiro Crew’s rule that the tightest applicable policy wins, so a permissive local setting cannot loosen an administrator’s restriction. See Kiro Crew.

H

Headless

Running a tool non-interactively, with no human present, typically inside a script or CI job. In Kiro that is kiro-cli chat --no-interactive "prompt" with KIRO_API_KEY for authentication. Recipe: headless CLI in CI.

Hook

An automation that runs an agent prompt or a shell command when something happens: a file save, create, or delete, a prompt submission, an agent turn completing, a tool invocation, a spec task starting or finishing, or a manual button. See Hooks.

Hunk

A small block of changed lines in a diff: the unit you accept or reject in Supervised mode.

I

IAM Identity Center

An AWS sign-in option for Kiro, alongside GitHub, Google, and AWS Builder ID. Organisations on Identity Center may need an administrator to enable features such as cloud sessions, and Identity Center sign-in for Kiro Web runs in us-east-1 only.

Inclusion mode

The YAML front matter setting that decides when a steering file loads: always (the default), fileMatch with a pattern, manual (referenced with #steering-file-name), or auto with a name and description Kiro matches against your request. AGENTS.md files do not support inclusion modes.

introspect

A built-in Kiro tool and subagent that explains Kiro’s own features and your configuration back to you. Useful when you cannot remember which steering file is loaded or why a tool was blocked.

K

Kiro CLI

kiro-cli, the agent in your terminal, with the full primitive set: specs, steering, hooks, custom agents, skills, and MCP. It is the surface that works over SSH, in containers, and in CI. See the CLI Guide and CLI.

Kiro Crew

Kiro’s fifth surface: an open-source personal AI agent, Apache 2.0 since 4 August 2026, that runs locally or remotely on your own hardware and is persistent, self-learning, and self-evolving. See Kiro Crew.

Kiro IDE

The desktop editor built on Code OSS with the agent woven in: chat with Autopilot and Supervised modes, the specs experience, steering and hook panels, MCP management, and visual diff review. See IDE.

Kiro Web

The browser agent at app.kiro.dev, generally available since 1 September 2026 on Pro and above. It works against GitHub and GitLab and returns pull or merge requests. See Web.

.kiroignore

A gitignore-syntax file that keeps paths out of the agent’s reach. Workspace files live at the repository root or in subdirectories, the global one at ~/.kiro/settings/kiroignore. You cannot re-include a file inside an excluded directory. Recipe: .kiroignore for secrets.

KIRO_API_KEY

The environment variable that authenticates a headless CLI run. Keep it in your CI system’s secret store, never in the repository. Recipe: headless CLI in CI.

KIRO_HOME

The environment variable that relocates Kiro’s global configuration directory away from ~/.kiro/, which is how you keep separate profiles on one machine.

M

MCP

Model Context Protocol, an open standard that lets AI tools call external services and data sources through small add-on servers. Configured in .kiro/settings/mcp.json for a project or ~/.kiro/settings/mcp.json for you. See MCP and the recipe for adding a server.

Memory

In Kiro Web, what the agent learns from the task creator’s pull request feedback: per user, across repositories, viewable and deletable but not editable under Settings, Memory. Kiro Crew has its own memory covering preferences, history, lessons, and semantic facts.

Mobile

Kiro’s iOS app, in preview through TestFlight, which steers cloud sessions and syncs with Kiro Web. Access is requested at kiro.dev/mobile.

O

Open-weight model

A model whose weights are published, offered in Kiro alongside the Anthropic and OpenAI lineups at low credit multipliers. They are the models available on the Free tier alongside Claude Sonnet 4.5. See Choosing a Model.

P

Permissions

Kiro’s capability-based rules for what the agent may do. Each rule has an effect (deny, ask, or allow), a match glob, and an optional exclude; deny always wins. Rules live in ~/.kiro/settings/permissions.yaml for you and in a workspace file kept outside the repository, across six scopes. Recipe: permissions.yaml.

Power

A bundle that makes the agent good at one technology: a POWER.md steering file telling it which MCP tools it has and when to use them, an MCP server configuration, and optional steering or hooks. Powers load on demand when keywords match. See Powers.

Property-based testing

Testing that checks a general rule holds across many generated inputs instead of a few hand-picked examples. In Kiro it verifies that an implementation matches its spec.

Q

Q Developer

Amazon Q Developer, Kiro’s predecessor. Kiro CLI succeeded Q Developer CLI on 17 November 2025, new signups closed on 15 May 2026, and plugins and paid subscriptions reach end of support on 30 April 2027. See Migrate from Amazon Q Developer.

R

Reasoning effort

How hard the model thinks, set to low, medium, high, xhigh, or max. Lower effort is faster and cheaper. Set it in the IDE from the Effort panel, or in the CLI with /effort or --effort; session beats workspace beats user beats built-in. Recipe: set reasoning effort.

Requirements analysis

An optional Kiro pass over requirements.md that flags ambiguity, contradictions, and gaps before you approve it. Running it costs a minute and catches the failures that would otherwise become code.

Rewind

Forking a CLI conversation at an earlier turn into a new session, leaving your files untouched. It is CLI-only, and it is not the same as restoring a checkpoint, which does change files. Recipe: checkpoints and rewind.

S

Sandbox mode

Kiro Crew’s operating-system sandbox setting. Auto hides .gnupg, .gcloud, .azure, and .docker; Strict also hides .aws, .ssh, and .kube; Off is available on Linux and macOS only and removes the protection. See Kiro Crew.

Skill

A portable instruction pack for a procedure: a SKILL.md with a name and description in YAML front matter, plus optional scripts and templates. Kiro discovers skills at session start and loads one when its description matches or you invoke /skill-name. Recipe: write a skill.

Spec

A written plan Kiro builds with you before code exists, producing requirements.md, design.md, and tasks.md in your repository. Feature specs offer Requirements-First, Design-First, and Quick Plan workflows. See Specs.

Steering

Markdown files that give the agent persistent project knowledge so you stop re-explaining your stack and conventions. They live in .kiro/steering/ or ~/.kiro/steering/, with the workspace winning on conflict. See Steering.

Subagent

A separate agent the main agent invokes for a scoped job, through the built-in invoke_subagent tool and the subagent permission capability. Kiro Crew runs subagents in parallel in the background.

Supervised mode

The IDE chat mode where the agent pauses after each turn that edits files so you accept or reject changes hunk by hunk. It is the opposite of Autopilot, and it is unavailable in cloud sessions.

T

Tangent

A CLI side-conversation that inherits the full history of the session you branched from, started with /tangent and browsed with /tangent ls. Useful for a quick detour you do not want polluting the main thread.

Task wave

How Kiro executes tasks.md: independent tasks run concurrently in waves, and dependent tasks wait for theirs. Small, decoupled tasks parallelise; one giant intertwined task serialises everything.

Tools

The actions the agent can take. Cross-surface built-ins include read_file, list_directory, file_search, grep_search, fs_write, fs_append, str_replace, delete_file, execute_bash, web_search, web_fetch, invoke_subagent, disclose_context, introspect, and todo_list, with extras specific to the IDE and CLI. Custom agents restrict access with the tools field: read, write, shell, web, @builtin.

V

Vibe coding

Freeform conversation where the agent codes immediately, with no written plan. It is the right choice for exploration, quick prototypes, and throwaway scripts, and the wrong one for anything a teammate must understand later. See Use Cases.

W

Workflows (Kiro Crew)

Crew’s multi-agent pipelines and fan-out patterns, defined as Python scripts. See Kiro Crew.

Frequently asked questions

What does Kiro mean by a spec?

A spec is a written plan Kiro creates with you before code exists. It produces three Markdown files in your repository: requirements.md for what the system should do, design.md for how it will be built, and tasks.md for the ordered steps. Bugfix specs use bugfix.md in place of requirements.md.

What is the difference between steering, skills, and powers?

Steering is always-on knowledge about your project, such as your conventions and stack. A skill is an on-demand procedure for a task, packaged as a SKILL.md file. A power is a technology bundle: POWER.md steering plus an MCP server configuration, loaded when keywords match.

What is a Kiro credit?

A credit is Kiro's unit of billing, consumed by prompts, spec refinement, task execution, and hook execution. Billing is fractional in 0.01-credit increments, and each model has a multiplier expressed relative to Auto at 1.0x.

What is EARS notation?

EARS stands for Easy Approach to Requirements Syntax. It writes requirements as WHEN condition THE SYSTEM SHALL behavior, with IF and THEN for error cases and WHILE for ongoing states, so every line is testable.

☰ Chapter summary

  • Sixty-plus Kiro terms in alphabetical order, each in one or two sentences with a link to the full explanation.
  • Covers the building blocks: specs, steering, hooks, MCP, custom agents, skills, powers, and Crew.
  • Covers the safety vocabulary: permissions, capabilities, sandbox modes, .kiroignore, checkpoints, and rewind.
  • Covers the money vocabulary: credits, multipliers, reasoning effort, and the Auto router.
  • Use it as a lookup table, not a chapter to read end to end.

All chapter summaries are collected on the revision page.

Was this chapter helpful?