Sandbox
@boringmarketer/kimi-first

Claude Code skill for Kimi Code CLI work orders

This skill routes frozen implementation tasks from Claude Code to Kimi Code CLI. Claude writes the spec and checks the result, while Kimi does the mechanical coding, migrations, test fills, and CI fixes. It uses a temp prompt file, runs `kimi -p` headless, reviews the diff, and retries with `kimi --continue` before Claude takes over directly.

46 stars2 forksUpdated 13d ago
Who it's for

Builders who use Claude Code and want a separate execution lane for repetitive implementation work.

What it delivers

You can keep Claude on judgment work while Kimi handles the bulk coding and Claude verifies the result.

What it does

Work-order routing

Sends only frozen, spec-ready tasks to Kimi Code CLI instead of delegating design or tiny edits.

Full-context prompt writing

Writes the goal, repo paths, constraints, non-goals, proof, and output shape into a temp file before calling Kimi.

Headless Kimi execution

Runs `kimi -p` in the target repo with tool calls auto-approved and answer output on stdout.

Claude-led review

Claude reviews the diff like a PR reviewer and runs tests itself before accepting the change.

Retry and fallback flow

Uses `kimi --continue` for follow-up rounds, then Claude takes over after two failed attempts.

How to get it

  1. 11. Kimi Code CLI (skip if you have it — check with kimi --version)
    curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash
    kimi login   # device-code flow, needs a Kimi membership
  2. 22. The skill, into Claude Code
    mkdir -p ~/.claude/skills/kimi-first
    curl -fsSL https://raw.githubusercontent.com/boringmarketer/kimi-first/main/skills/kimi-first/SKILL.md \
      -o ~/.claude/skills/kimi-first/SKILL.md
  3. 3Or clone and symlink
    git clone https://github.com/boringmarketer/kimi-first.git
    ln -s "$(pwd)/kimi-first/skills/kimi-first" ~/.claude/skills/kimi-first

README

kimi-first

A Claude Code skill that routes implementation work-orders to Kimi Code CLI (Kimi k3) lanes — Kimi types, Claude thinks and verifies.

Claude keeps the judgment work: specs, architecture, orchestration, review, and the merge button. Kimi gets the hands-on volume: frozen-spec builds, mechanical migrations, test fills, bulk sweeps, CI fixes. Kimi's flat-rate membership makes it a cheap, high-throughput implementation lane while Claude spends its tokens on the parts that need judgment.

This is a port of @steipete's excellent codex-first skill (from agent-scripts) — same delegation doctrine, swapped from the Codex CLI to the Kimi Code CLI.

How it works

When a task reads as a work order (the spec is frozen, no design decisions left), Claude:

  1. Writes a full-context prompt to a temp file — goal, repo paths, constraints, non-goals, expected proof, output shape. Kimi starts with zero session context, so spec quality decides success.
  2. Runs kimi -p headless in the target repo (tool calls auto-approved, thinking to stderr, answer to stdout).
  3. Reviews the full diff like a hostile PR reviewer, runs the tests itself — Kimi's claims are advisory.
  4. Iterates via kimi --continue; after 2 failed rounds, takes over and does it directly.
  5. Ships through normal repo protocol — Claude's hands on the merge button, never Kimi's.

Design work, tiny edits, prod operations, and review are never delegated.

Install

1. Kimi Code CLI (skip if you have it — check with kimi --version):

curl -fsSL https://code.kimi.com/kimi-code/install.sh | bash
kimi login   # device-code flow, needs a Kimi membership

Docs: moonshotai.github.io/kimi-code

2. The skill, into Claude Code:

mkdir -p ~/.claude/skills/kimi-first
curl -fsSL https://raw.githubusercontent.com/boringmarketer/kimi-first/main/skills/kimi-first/SKILL.md \
  -o ~/.claude/skills/kimi-first/SKILL.md

Or clone and symlink:

git clone https://github.com/boringmarketer/kimi-first.git
ln -s "$(pwd)/kimi-first/skills/kimi-first" ~/.claude/skills/kimi-first

Claude Code picks it up automatically; invoke it explicitly with /kimi-first or just hand Claude a work-order-shaped task.

Verified behavior

The CLI mechanics in the skill were live-tested against Kimi Code CLI v0.26.0:

  • kimi -p runs headless with tool calls auto-approved — answer on stdout (-prefixed), thinking on stderr
  • --yolo refuses to combine with -p (and isn't needed)
  • kimi --continue -p resumes the latest session for that directory with full context
  • --continue is cwd-filtered — never run two concurrent sessions in the same directory
  • Headless mode auto-approves, so only delegate from repos without prod secrets in reach

Credits

License

MIT

Files in the repo

Repository payload4 top-level entries
  • agents
  • skills
  • LICENSE
  • README.md

Discussion (0)

Ask about usage, or say what you built with it

Sign in to join the discussion.

No comments yet. Be the first to say what this is good for.

More skills

obra/
superpowers

An agentic skills framework & software development methodology that works.

285k
1 add

Turn any codebase, with its docs, SQL schemas, configs, and PDFs, into a queryable knowledge graph. A /graphify skill for Claude Code, Cursor, Codex, and Gemini CLI: local deterministic AST parsing, every edge explained, no vector store.

117k
1 add
Vincentwei1021/
anything2explainer

Topic in, narrated explainer video out. A Claude Code / Codex skill that turns any topic into a black-canvas motion-graphics explainer video with TTS voiceover, subtitles and a chapter progress bar. Chinese or English; every frame drawn in code with Remotion.

666

Open-source AI job search: scan job portals, evaluate listings into a structured A-H report with a global 1-5 score, tailor your CV, track applications — runs locally in your AI coding CLI (Claude Code, Codex, OpenCode, Antigravity…)

71k