Kody Engine Lite

Comment @kody on a GitHub issue. Get back a tested, reviewed PR. Free and open source.

Kody wraps Claude Code with a 7-stage autonomous pipeline — classify, plan, build, verify, review, fix, ship — with quality gates between every stage. If verify catches a bug, it gets fixed before review ever sees it. No blind retries, no context drift, no babysitting.

• Repo-aware prompts — bootstrap analyzes your codebase and generates customized instructions for every stage, not generic "write clean code" prompts
• Quality gates — runs your repo's typecheck, tests, and lint between stages + AI code review in a fresh session
• AI failure diagnosis — classifies errors as fixable/infrastructure/pre-existing before retrying
• Self-improving — learns conventions, remembers architectural decisions, discovers existing patterns
• Runs anywhere — locally from your terminal or via GitHub Actions
• Anthropic-compatible models — Anthropic natively, or other providers (MiniMax, Gemini, etc.) via LiteLLM proxy

Why Kody? → · Full comparison →

@kody on issue
    │
    ▼
① TASKIFY ─── classify, scope, detect complexity, ask questions
    │
    ▼
② PLAN ────── TDD implementation plan (deep reasoning)
    │          🛑 HIGH risk? Pause for human approval
    ▼
③ BUILD ───── implement via Claude Code tools
    │
    ▼
④ VERIFY ──── run your quality commands (typecheck, tests, lint)
    │          ✗ fail → AI diagnosis → autofix → retry
    ▼
⑤ REVIEW ──── AI code review (fresh session, no build bias)
    │
    ▼
⑥ REVIEW-FIX ─ fix Critical and Major findings
    │
    ▼
⑦ SHIP ────── push branch, create PR with Closes #N

Quick Start

Prerequisites: A GitHub repo + an Anthropic API key (or compatible provider key).

For local CLI usage, you also need: Node.js >= 22, Claude Code CLI, GitHub CLI, git.

1. Set up GitHub

Add your API key as a secret — via GitHub web UI or CLI:

gh secret set ANTHROPIC_API_KEY --repo owner/repo

Then in GitHub: Settings → Actions → General → "Allow GitHub Actions to create and approve pull requests"

2. Initialize

Copy the workflow template to .github/workflows/kody.yml and add a kody.config.json to your repo root. Or use the CLI to auto-generate both:

npm install -g @kody-ade/kody-engine-lite
cd your-project
kody-engine-lite init

3. Bootstrap

Create a new GitHub issue (e.g., "Setup Kody") and comment:

@kody bootstrap

This analyzes your codebase with an LLM and generates:

• Project memory (.kody/memory/ — architecture and conventions)
• Customized step files (.kody/steps/ — repo-aware prompts for every stage)
• GitHub labels for lifecycle tracking (14 labels)

4. Use

Comment on any GitHub issue:

@kody

Kody picks up the issue and works through the pipeline autonomously. You'll see:

• Labels updating in real-time: kody:planning → kody:building → kody:review → kody:done
• Progress comments on the issue at each stage
• A PR with a rich description, passing quality checks, and Closes #N

If the task is HIGH-risk, Kody pauses after planning and asks for approval before writing code.

Switch to a different model (optional)

// kody.config.json
{ "agent": { "provider": "minimax" } }

# .env
ANTHROPIC_COMPATIBLE_API_KEY=your-key-here

Kody auto-starts the LiteLLM proxy. Full LiteLLM guide →

Commands

| Command | What it does | |---------|-------------| | @kody | Run full pipeline on an issue | | @kody review | Review any PR — structured findings + GitHub approve/request-changes (falls back to comment if self-review blocked) | | @kody fix | Re-run from build with human PR feedback + Kody's review as context | | @kody fix-ci | Fix failing CI checks (auto-triggered with loop guard) | | @kody resolve | Merge default branch into PR, AI-resolve conflicts, verify, push | | @kody rerun | Resume from failed or paused stage | | @kody rerun --from <stage> | Resume from a specific stage | | @kody approve | Resume after questions or risk gate | | @kody bootstrap | Regenerate project memory and step files |

kody-engine-lite init [--force]          # Setup repo: workflow + config
kody-engine-lite bootstrap [--force]     # Generate memory + step files + labels
kody-engine-lite run --issue-number 42 --local --cwd ./project
kody-engine-lite run --task "Add retry utility" --local
kody-engine-lite review --pr-number 42   # Standalone PR review
kody-engine-lite fix --issue-number 42 --feedback "Use middleware pattern"
kody-engine-lite fix-ci --pr-number 42
kody-engine-lite resolve --pr-number 42   # Merge + resolve conflicts
kody-engine-lite rerun --issue-number 42 --from verify

Full CLI reference with all flags and options →

Key Features

• Repo-Aware Step Files — auto-generated prompts with your repo's patterns, gaps, and acceptance criteria (details)
• Standalone PR Review — @kody review on any PR for structured code review with GitHub approve/request-changes (details)
• Shared Sessions — stages share Claude Code sessions, no cold-start re-exploration (details)
• Risk Gate — HIGH-risk tasks pause for human approval before building (details)
• AI Failure Diagnosis — classifies errors as fixable/infrastructure/pre-existing/abort before retry (details)
• Question Gates — asks product/architecture questions when the task is unclear (details)
• Auto Fix-CI — CI fails on a PR? Kody fetches logs, diagnoses, and pushes a fix (details)
• Pattern Discovery — searches for existing patterns before proposing new ones (details)
• Decision Memory — architectural decisions extracted from reviews persist across tasks (details)
• Auto-Learning — extracts coding conventions from each successful run (details)
• Retrospective — analyzes each run, identifies patterns, suggests improvements (details)
• Anthropic-Compatible Models — route through LiteLLM to use other providers like MiniMax, Gemini, etc. (setup guide)

Documentation

Understand Kody: About · Features · Pipeline · Comparison

Set up & use: CLI · Configuration · Bootstrap · LiteLLM

Reference: Architecture · FAQ

License

MIT