Engineering Agent Configuration

A reusable Kimi Code configuration that turns the CLI into a disciplined full-stack product-engineering agent.

What this repo contains

• AGENTS.md — delivery contract and project rules.
• .agents/agents/product-engineer.md — custom main agent profile.
• .agents/skills/* — reusable skills for project bootstrap, prompt engineering, product discovery, architecture, UI, API, migrations, browser verification, security review, and release engineering.
• .kimi-code/mcp.json — Playwright MCP server for browser verification.
• SYSTEM.md — template for a user-level system prompt that applies prompt-engineering discipline to every session.
• kimi.plugin.json — plugin manifest so the whole configuration can be installed via /plugins install.
• hooks/ — lifecycle hooks, including UserPromptSubmit prompt engineering.
• docs/ — product brief, architecture, acceptance criteria, progress log, and runbook.

How to use it

1. Clone or copy this repo into your project root.
2. Start a new Kimi Code session with the product-engineer agent:

   cd /path/to/your-project
   kimi --agent product-engineer

3. Kimi discovers the project skills automatically from .agents/skills/.
4. Use /skill:<name> to invoke a specific skill, e.g.:

   /skill:bootstrap
   /skill:browser-verification
   

Bootstrapping a new project

Start Kimi in an empty or existing project folder and invoke the bootstrap skill:

cd /path/to/your-project
kimi --agent product-engineer

Then ask:

/skill:bootstrap

The agent will orient to the directory, classify its state, and create or update AGENTS.md and .kimi-code/PROJECT.md so future sessions start informed.

Automated prompt engineering

The product-engineer agent automatically reframes every request into a structured software-engineering prompt (goal, context, acceptance criteria, scope, constraints, verification plan, risks) before acting.

To rewrite a prompt explicitly without acting on it:

/skill:prompt-engineer fix the login bug

For global automation across every Kimi session — not just sessions started with --agent product-engineer — copy SYSTEM.md to your user-level Kimi home:

cp SYSTEM.md ~/.kimi-code/SYSTEM.md

This replaces the default main-agent system prompt with the prompt-engineering discipline. Changes take effect in new sessions.

Install as a Kimi Code plugin

For the most automatic setup, install the repo as a plugin. This enables the prompt-engineering UserPromptSubmit hook and injects the discipline into every session, regardless of which agent you select.

/plugins install https://github.com/Quattro-Commas/engineering-agent

Then reload or start a new session:

/reload

The plugin ships:

• All engineering-agent skills under /skill:<name>
• The product-engineer agent available as a sub-agent
• A UserPromptSubmit hook that scaffolds vague requests
• Playwright MCP for browser verification

Validation

Run the local validation script before committing changes to agent or skill files:

python3 scripts/validate-config.py

CI runs the same script on every push and pull request.

Requirements

• Kimi Code CLI ≥ 2.2.0
• Node.js and npm (for Playwright MCP)
• A project that has real acceptance criteria and a runnable dev environment

Status

Initial configuration committed on 2026-10-18. Ready for vertical-slice testing.