| name | g-skl-platform-kiro-cli |
| description | Authoritative reference for Kiro CLI (Amazon's terminal agent, the Q Developer CLI rebrand) customization in gald3r projects. Covers .kiro/steering/ + AGENTS.md, Agent Skills (SKILL.md), JSON custom agents + subagents, lifecycle hooks, slash commands, MCP, and gald3r install verification. Distinct from Kiro IDE (g-skl-platform-kiro). |
| crawl_max_age_days | 7 |
| vault_doc_path | research/platforms/kiro-cli/ |
| vault_docs_url | https://kiro.dev/docs/cli |
| docs_url | https://kiro.dev/docs/cli |
| docs_url_secondary | ["https://kiro.dev/docs/cli/skills/","https://kiro.dev/docs/cli/steering/","https://kiro.dev/docs/cli/custom-agents/configuration-reference/","https://kiro.dev/docs/cli/hooks/","https://kiro.dev/docs/cli/mcp/","https://kiro.dev/docs/cli/reference/slash-commands/"] |
| last_doc_scan | "2026-06-02T00:00:00.000Z" |
| capability_status | {"hooks":"✅ native lifecycle hooks in agent JSON config (agentSpawn/userPromptSubmit/preToolUse/postToolUse/stop; STDIN-JSON; exit 2 blocks PreToolUse)","rules":"✅ steering files .kiro/steering/*.md (product/tech/structure auto-loaded) + reads AGENTS.md","skills":"✅ Agent Skills (SKILL.md) auto-loaded from .kiro/skills/ + ~/.kiro/skills/; also /skill-name","commands":"✅ Skills-as-slash-commands (/skill-name) + /prompts create; no standalone command-file format","agents":"✅ native JSON custom agents (filename=name) + subagents (isolated context, up to 4)","mcp":"✅ native — mcpServers JSON at .kiro/settings/mcp.json + ~/.kiro/settings/mcp.json"} |
| token_budget | low |
| subsystem_memberships | ["PLATFORM_INTEGRATION"] |
g-skl-platform-kiro-cli
Activate for: setting up gald3r with Kiro CLI (terminal variant), authoring steering/skills/agents/hooks, understanding differences from Kiro IDE, or verifying the Kiro CLI gald3r install.
Full 9-section breakdown + evidence URLs in PLATFORM_SPEC.md (this folder). Status: ✅ full
parity — Kiro CLI natively supports commands, rules, agents, skills, hooks, and MCP, reads
AGENTS.md (not CLAUDE.md), and auto-discovers SKILL.md under .kiro/skills/, so gald3r's
AGENTS.md + SKILL.md artifacts are drop-in reusable. (Verified 2026-06-02 against
https://kiro.dev/docs/cli.) Caveats are format not capability: agents are JSON-only, commands
route through skills/prompts, and hooks wire per-agent-JSON via STDIN.
1. Platform Overview
Kiro CLI is Amazon's terminal agentic coding assistant — the rebrand/successor of the
Amazon Q Developer CLI (q / q chat entry points preserved). It is spec-driven and
agent-centric, with the full extensibility surface under .kiro/. It shares the .kiro/
directory and steering context mechanism with the Kiro IDE, but has a richer
agent/hook/command/skills surface than the IDE (see g-skl-platform-kiro, T1472).
- Instruction file:
.kiro/steering/*.md (foundation files auto-loaded) + AGENTS.md
standard — no CLAUDE.md/GEMINI.md
- Skills: ✅ Agent Skills
SKILL.md auto-loaded from .kiro/skills/ + ~/.kiro/skills/; each
also exposed as a /skill-name slash command (v2.1, 2026-04-24)
- Custom agents: ✅ JSON configs (filename without
.json = agent name; /agent create) +
subagents (isolated context, up to 4 at once) — the IDE lacks this
- Lifecycle hooks: ✅ in agent config —
agentSpawn, userPromptSubmit, preToolUse,
postToolUse, stop (Cursor-like taxonomy; the IDE uses fileEdited hooks instead)
- Commands: ✅ Skills-as-slash-commands +
/prompts; built-ins (/agent, /context, /model,
/guide, /settings)
- MCP: ✅
mcpServers JSON at .kiro/settings/mcp.json (migrated from ~/.aws/amazonq/mcp.json)
- AWS integration: Amazon Q / Bedrock model access via AWS credentials
Full capability assessment: see PLATFORM_SPEC.md (9 sections, doc-verified 2026-06-02).
2. Config Layout
<project-root>/
├── AGENTS.md ← read by Kiro CLI (AGENTS.md standard; NOT CLAUDE.md)
└── .kiro/
├── steering/ *.md ← always-on context (product/tech/structure auto-loaded)
├── skills/ <name>/SKILL.md ← Agent Skills (YAML: name, description)
├── settings/
│ └── mcp.json ← mcpServers object
└── (custom-agent JSON configs — filename without .json = agent name)
Global mirror under ~/.kiro/ (steering/, skills/, settings/mcp.json). Workspace skills
override identically-named global skills. Same .kiro/ directory as Kiro IDE — steering + MCP
are shared, but agent/hook/command/skills config is CLI-specific.
3. gald3r Integration
Cheapest high-parity install: ship gald3r's AGENTS.md (+ g-skl-*/SKILL.md under
.kiro/skills/) — Kiro CLI reads AGENTS.md and auto-discovers SKILL.md (each also becomes
/skill-name). Add gald3r rules as .kiro/steering/*.md, a gald3r custom-agent JSON, and
.kiro/settings/mcp.json for MCP servers.
Verify
Test-Path AGENTS.md # instruction file Kiro CLI reads
Test-Path .kiro/steering # always-on context
Test-Path .kiro/skills # Agent Skills (SKILL.md)
Test-Path .kiro/settings/mcp.json # MCP config
kiro-cli --version # or: q --version (preserved Q Developer CLI entry point)
❓ The exact --version/headless flags for the current Kiro CLI are unverified — confirm against
https://kiro.dev/docs/cli/reference/slash-commands/ before scripting CI.
4. Common Pitfalls
- Reads
AGENTS.md, NOT CLAUDE.md — gald3r's instruction content must live in AGENTS.md
(or .kiro/steering/), not a CLAUDE.md root file.
- No standalone command-file format — custom commands are expressed as Skills (
/skill-name)
or local prompts (/prompts create); a gald3r command-per-file model maps onto skills.
- Custom agents are JSON, not markdown —
g-agnt-*.md need markdown→JSON translation (pin
steering via resources), not a file drop.
- Hooks wire per-agent JSON, STDIN payload — hooks live in each agent's
hooks field (not a
central hooks.json); cross-agent automation is replicated per agent; gald3r .ps1 hooks must
read $input/stdin (not $env:*); exit code 2 blocks (PreToolUse only).
- No per-rule glob scoping — steering has no
alwaysApply/globs; foundation files are always
on, others scope via an agent resources glob.
- Shared
.kiro/ with Kiro IDE — installing steering/MCP for one benefits the other, but the
CLI's agent/hook/command/skills config is not the IDE's (fileEdited hooks, no custom agents).
5. Capability Summary
| Feature | Status | Notes |
|---|
Hooks (g-hk-*.ps1) | ✅ | agent JSON hooks; agentSpawn/userPromptSubmit/preToolUse/postToolUse/stop; STDIN-JSON; exit 2 blocks (PreToolUse); per-agent wiring |
Skills (g-skl-*/SKILL.md) | ✅ | Agent Skills auto-loaded from .kiro/skills/ + ~/.kiro/skills/; also /skill-name (v2.1) |
Agents (g-agnt-*.md) | ✅ | native JSON custom agents (filename=name) + subagents (isolated, up to 4); markdown→JSON translation |
Commands (@g-*) | ✅ | Skills-as-slash-commands (/skill-name) + /prompts create; no standalone command-file format |
Rules (g-rl-*) | ✅ | .kiro/steering/*.md (product/tech/structure auto-loaded) + reads AGENTS.md; no per-rule glob scoping |
| MCP | ✅ | mcpServers JSON at .kiro/settings/mcp.json + ~/.kiro/settings/mcp.json |
Differences from Kiro IDE (g-skl-platform-kiro, T1472) — do not conflate the two:
| Capability | Kiro IDE | Kiro CLI |
|---|
| Custom agents | ❌ none | ✅ JSON configs (/agent create) + subagents |
| Hooks | ⚠️ fileEdited JSON files in .kiro/hooks/ | ✅ lifecycle hooks in agent config (agentSpawn/userPromptSubmit/preToolUse/postToolUse/stop) |
| Slash commands | ❌ none (spec workflow) | ✅ Skills-as-slash-commands + /prompts + built-ins |
| Skills | (IDE: see kiro spec) | ✅ Agent Skills SKILL.md auto-loaded |
| Steering / MCP | ✅ shared | ✅ shared |
Full assessment + evidence in PLATFORM_SPEC.md. Re-verify on the next @g-platform-scan-docs kiro-cli (crawl_max_age_days: 7).