agents

name	agents
description	Agent routing system — automatically selects the best specialist agent based on user request domain. Handles keyword matching, Vietnamese and English triggers, compound-keyword routing, vague-request fallback, and multi-agent coordination. This skill is always loaded silently.
encoding	UTF-8
metadata	{"category":"system","auto-load":true}

Purpose

This is the master agent routing skill. It analyzes every user request and silently applies the most relevant specialist knowledge.

Memory Protocol

START: Read .ai-memory.md from project root — check which agents were used previously and what domain knowledge has been accumulated. END: Update .ai-memory.md using Memory Compaction Rules with which agent was applied and why.

Agent Selection Matrix

Domain	Primary Agent	Agent File	Trigger Keywords	Skill Files to Load
Bug/Error analysis	@debugger	.kiro/skills/agents/agents/debugger.md	debug, lỗi, error, bug, crash, 500, exception, fail, fix, sửa, hỏng, lỗi rồi	.kiro/skills/debug/SKILL.md, .kiro/skills/test/SKILL.md
API/Backend	@backend-specialist	.kiro/skills/agents/agents/backend-specialist.md	api, server, endpoint, route, middleware, auth, jwt, database query, tạo API, backend	.kiro/skills/create/SKILL.md, .kiro/skills/deploy/SKILL.md
UI/Frontend	@frontend-specialist	.kiro/skills/agents/agents/frontend-specialist.md	ui, ux, component, page, css, layout, responsive, animation, tạo trang, giao diện, trang	.kiro/skills/create/SKILL.md, .kiro/skills/ui-ux-pro-max/SKILL.md
Database	@database-architect	.kiro/skills/agents/agents/database-architect.md	database, schema, migration, query, sql, prisma, orm, model, cơ sở dữ liệu	.kiro/skills/create/SKILL.md, .kiro/skills/deploy/SKILL.md
Security	@security-auditor	.kiro/skills/agents/agents/security-auditor.md	security, vulnerability, xss, csrf, injection, auth, encrypt, ssl, bảo mật	.kiro/skills/debug/SKILL.md, .kiro/skills/test/SKILL.md
DevOps/Deploy	@devops-engineer	.kiro/skills/agents/agents/devops-engineer.md	deploy, ci/cd, docker, kubernetes, aws, vercel, pipeline, build, triển khai, phát hành	.kiro/skills/deploy/SKILL.md
Testing	@test-engineer	.kiro/skills/agents/agents/test-engineer.md	test, coverage, jest, vitest, playwright, e2e, unit test, spec, kiểm thử, viết test	.kiro/skills/test/SKILL.md
Performance	@performance-optimizer	.kiro/skills/agents/agents/performance-optimizer.md	performance, slow, optimize, cache, bundle, lighthouse, speed, chậm, nhanh hơn, tối ưu	.kiro/skills/enhance/SKILL.md
Documentation	@documentation-writer	.kiro/skills/agents/agents/documentation-writer.md	docs, readme, api docs, changelog, comment, jsdoc, tài liệu	.kiro/skills/create/SKILL.md
Code Explanation	@documentation-writer	.kiro/skills/agents/agents/documentation-writer.md	explain, giải thích, walkthrough, how does this work, what does this do, hiểu code	.kiro/skills/explain/SKILL.md
Planning/Multi-domain	@orchestrator	.kiro/skills/agents/agents/orchestrator.md	orchestrate, coordinate, full-stack, multi-step, complex, tổng hợp	.kiro/skills/orchestrate/SKILL.md, .kiro/skills/plan/SKILL.md
Project planning	@project-planner	.kiro/skills/agents/agents/project-planner.md	plan, milestone, breakdown, estimate, roadmap, sprint, kế hoạch, lên kế hoạch	.kiro/skills/plan/SKILL.md, .kiro/skills/brainstorm/SKILL.md

Compound Keyword Rules (checked BEFORE single keywords)

Pattern	Route
test + fail/error/broken/lỗi	@debugger → debug test failures (.kiro/skills/debug/SKILL.md)
create + test / viết test	@test-engineer → write new tests (.kiro/skills/test/SKILL.md)
deploy + fail/error/lỗi	@debugger + @devops-engineer → debug deploy issues (.kiro/skills/debug/SKILL.md)
review + code	@security-auditor + @performance-optimizer + @documentation-writer → holistic review (.kiro/skills/enhance/SKILL.md)

Routing Protocol

1. Analyze the user's request for domain keywords from the matrix above.
2. Select the primary agent from the matching row.
3. Read the agent's persona file at the path listed in "Agent File" column.
4. Read the skill file(s) listed in "Skill Files to Load" column.
5. If request spans multiple domains, read .kiro/skills/agents/agents/orchestrator.md to coordinate.
6. Announce: "🤖 Applying @{agent} knowledge..."

Vague / General Request Fallback

When the user's request does NOT match any domain keywords:

User says	Action
"Help me", "What should I do?"	Read .kiro/skills/status/SKILL.md → then .kiro/skills/plan/SKILL.md
"Review my code", "review code cho tôi"	Read .kiro/skills/enhance/SKILL.md with @security-auditor + @performance-optimizer + @documentation-writer (holistic review)
"I'm stuck", "Don't know where to start"	Read .kiro/skills/status/SKILL.md → then .kiro/skills/brainstorm/SKILL.md
"Explain this", "giải thích code", "how does this work?"	Read .kiro/skills/explain/SKILL.md with @documentation-writer
Complaints about code ("too long", "messy", "dài quá", "rối quá")	Read .kiro/skills/enhance/SKILL.md or .kiro/skills/clean/SKILL.md — ask Socratic Gate to clarify intent
Any multi-part or unclear request	Read .kiro/skills/agents/agents/orchestrator.md → triage into sub-tasks
Greeting ("hi", "hello", "xin chào")	Respond warmly, list skills, ask how to help
Empty message	Respond with skill menu, ask what to work on
Completely unclear	Ask ONE clarifying question, then route to the best matching skill

Never skip routing for general requests. General requests need the MOST help, not the least.

Multi-Agent Scenarios

For complex requests that span domains:

1. Read .kiro/skills/agents/agents/orchestrator.md as coordinator
2. Orchestrator decomposes task into workstreams
3. For each workstream, read the specialist agent file from the matrix
4. Execute sequentially, sharing state via .ai-memory.md
5. Merge results with consistency checks

• Clean code chuẩn (Standard clean code applied)
• Cập nhật đầy đủ tất cả các file liên quan (All related files fully updated)

Quality Gate

After completing any routed task, read .kiro/skills/_scripts/checklist.md for cross-cutting quality checks.

Rules

• Agent selection is SILENT — do not ask the user which agent to use
• Always announce which agent is being applied
• Never skip routing for general/vague requests — use the fallback table above
• Never override an explicit user request with agent logic