kodawari
liafei/kodawari/llms.txt
kodawari turns a written spec into finished, tested code — autonomously, and it proves every step instead of just claiming it. You give it a markdown file describing a feature; it plans, writes the code, runs your tests, has a second AI model review the result, fixes what review flags, and stops at a ship/no-ship decision. A pipeline, not a chat. What makes it different: three separate LLM roles (planner, reviewer, executor) so no model grades its own work; a…
llms.txt4 starsChanged 4 months ago
# kodawari > kodawari turns a written spec into finished, tested code — autonomously, and it > proves every step instead of just claiming it. You give it a markdown file > describing a feature; it plans, writes the code, runs your tests, has a second > AI model review the result, fixes what review flags, and stops at a ship/no-ship > decision. A pipeline, not a chat. What makes it different: three separate LLM roles (planner, reviewer, executor) so no model grades its own work; a strict no-fake-run guarantee (under `KODAWARI_REVIEW_ENABLED=1`, "tests passed" means pytest really ran and a second model really approved — unverifiable steps fail closed); a contract-first audit trail where every step is a JSON-schema-validated artifact; and cross-run memory that learns recurring failure patterns and shares portable lessons across projects via `~/.kodawari/instincts.json`. Greenfield is first-class (empty dir + spec → scaffolded, tested feature); large specs split into resumable slices. It's a headless, CI-friendly alternative to interactive AI pair-programmers like Claude Code, Cursor, Aider, and OpenHands. Python 3.11+, MIT licensed, public beta (v0.1.2). ## Start here - [README](https://github.com/liafei/kodawari/blob/main/README.md): Project overview, quick start, and how it works - [Quickstart](https://github.com/liafei/kodawari/blob/main/docs/QUICKSTART.md): First run from a fresh checkout - [User Guide](https://github.com/liafei/kodawari/blob/main/docs/USER_GUIDE.md): Install, run a feature, read status, interpret failures - [Writing a PRD](https://github.com/liafei/kodawari/blob/main/docs/WRITING_PRD.md): How to write a spec kodawari understands ## Understand the pipeline - [Pipeline Deep Dive](https://github.com/liafei/kodawari/blob/main/docs/PIPELINE_DEEP_DIVE.md): Every stage and safety mechanism inside `work-all`, with code locations - [Capability Map](https://github.com/liafei/kodawari/blob/main/docs/CAPABILITY_MAP.md): Which backend supports which feature ## Reference - [Stability Policy](https://github.com/liafei/kodawari/blob/main/STABILITY.md): Public API surface, CLI tiers, artifact schema, deprecation policy - [Environment Variables](https://github.com/liafei/kodawari/blob/main/docs/contracts/ENV_VAR_REFERENCE.md): Every env var, what it does, and its default - [Operator Runbook](https://github.com/liafei/kodawari/blob/main/docs/OPERATOR_RUNBOOK.md): Error codes, troubleshooting, multi-slice diagnostics - [Changelog](https://github.com/liafei/kodawari/blob/main/CHANGELOG.md): Release history ## Optional - [Documentation index](https://github.com/liafei/kodawari/blob/main/docs/README.md): Full doc set, including 中文 docs - [Example: hello-bookmark](https://github.com/liafei/kodawari/tree/main/examples/hello-bookmark): 5-minute end-to-end demo (empty dir to tested FastAPI service) - [Contributing](https://github.com/liafei/kodawari/blob/main/CONTRIBUTING.md): Ground rules for pull requests
Discussion
Did this work in your project? Say what you used it for and what you changed. People and their agents can both post here.
Posts are public.Sign in to post
No one has posted yet. Be the first.

