/kimi-coworker

Invoke Kimi K2.5 CLI as a coworker for implementation, brainstorming, and parallel task execution.

When to Delegate to Kimi

MANDATORY for:

•Design exploration — Generating visual proposal catalogs
•UI implementation — React/Tailwind components, CSS
•Design themes — Token systems, globals.css
•Frontend work — Any visual coding tasks

Prefer Kimi when:

•Budget matters (Kimi is ~80% cheaper than Codex)
•Frontend/visual coding (Kimi K2.5's strength)
•Parallel subtasks (Agent Swarm handles 100+ agents)
•Multimodal input needed (screenshots, diagrams)
•Complex reasoning (use --thinking mode)

Keep for Codex when:

•Git-heavy workflows (Codex has better git integration)
•You already have deep Codex context loaded
•Backend/API work without visual components
•Mission-critical code needing GPT-5 capabilities

Design Workflow Integration

Kimi is the designated agent for all design-related skills:

Skill	Kimi's Role
`/design-exploration`	Generate 6-12 visual proposals in parallel
`/design-catalog`	Build 5-8 design options
`/design-theme`	Implement token system in CSS
`/aesthetic-system`	Execute visual implementation
`/ui-skills`	Build UI components

Pattern: Research (Gemini) → Build (Kimi) → Review (Claude)

Invocation Patterns

Single Agent

code

mcp__kimi__spawn_agent({
  "prompt": "Create a React hook for form validation. Follow pattern in src/hooks/useAuth.ts.",
  "thinking": false
})

With Extended Reasoning

code

mcp__kimi__spawn_agent({
  "prompt": "Debug why the checkout flow fails intermittently. Trace through payment.ts and cart.ts.",
  "thinking": true
})

Parallel Agents (Agent Swarm)

code

mcp__kimi__spawn_agents_parallel({
  "agents": [
    {"prompt": "Write unit tests for UserProfile component"},
    {"prompt": "Write unit tests for UserSettings component"},
    {"prompt": "Write integration tests for user flow", "thinking": true}
  ]
})

Pre-Delegation Checklist

•Existing tests? → Warn: "Don't break tests in [test file]"
•Add or replace? → Be explicit: "ADD to this file" vs "REPLACE this file"
•Quality gates? → Include: "Run pnpm typecheck && pnpm lint after"
•Patterns to follow? → Include: "Follow pattern in [reference file]"

Post-Delegation Validation

bash

git diff --stat        # See what changed
pnpm typecheck         # Type check
pnpm lint             # Lint
pnpm test             # Run tests

Kimi K2.5 Modes

Mode	When to Use
Default	Standard tasks
`--thinking`	Complex debugging, architecture decisions
Agent Swarm	Parallel independent subtasks

Cost Comparison

Provider	Input	Output
Kimi K2.5	$0.15/M	$2.50/M
Codex (GPT-5)	~$2.50/M	~$10/M
Claude Opus	$15/M	$75/M

Kimi is ~80% cheaper than Codex for equivalent tasks.

Examples

Design Catalog Generation (Parallel)

code

mcp__kimi__spawn_agents_parallel({
  "agents": [
    {"prompt": "Generate proposal: Midnight Editorial. DNA: editorial, dark, display-heavy, orchestrated, spacious, gradient. Build .design-catalog/proposals/01-midnight-editorial/ with preview.html, styles.css", "thinking": true},
    {"prompt": "Generate proposal: Swiss Brutalist. DNA: grid-breaking, monochrome, minimal, none, compact, solid. Build .design-catalog/proposals/02-swiss-brutalist/", "thinking": true},
    {"prompt": "Generate proposal: Warm Workshop. DNA: asymmetric, brand-tinted, text-forward, subtle, mixed, textured. Build .design-catalog/proposals/03-warm-workshop/", "thinking": true}
  ]
})

UI Component Implementation

code

mcp__kimi__spawn_agent({
  "prompt": "Create a Modal component in src/components/Modal.tsx. Use Tailwind CSS. Follow the pattern in Button.tsx. Include: open/close state, backdrop click to close, escape key handler, focus trap. Apply ui-skills constraints. Run pnpm typecheck after."
})

Theme Token Implementation

code

mcp__kimi__spawn_agent({
  "prompt": "Implement design theme in app/globals.css using Tailwind 4 @theme directive. Colors (OKLCH): primary oklch(0.7 0.15 250), background oklch(0.98 0 0). Typography: heading font JetBrains Mono, body Inter. Modular scale. Then update components to use tokens.",
  "thinking": true
})

Bug Investigation

code

mcp__kimi__spawn_agent({
  "prompt": "Investigate why API calls in src/api/client.ts sometimes return stale data. Check caching logic, request deduplication, and race conditions. Report findings without making changes.",
  "thinking": true
})

Parallel Test Writing

code

mcp__kimi__spawn_agents_parallel({
  "agents": [
    {"prompt": "Write tests for src/utils/format.ts covering all exported functions"},
    {"prompt": "Write tests for src/utils/validate.ts covering edge cases"},
    {"prompt": "Write tests for src/utils/transform.ts with snapshot tests"}
  ]
})