Structure a large AI assisted project into discovery, design, implementation, and verification waves.
Enforce a hard readiness gate so implementation cannot start before requirements are solid.
Route small versus large tasks through different amounts of process automatically.
Have independent reviewer roles check work instead of the same role approving its own output.
| taxideftis/bathos | akitaonrails/ratatui-bubbletea | deepdiy/pdf2md | |
|---|---|---|---|
| Stars | 20 | 20 | 20 |
| Language | Rust | Rust | Rust |
| Setup difficulty | — | easy | easy |
| Complexity | — | 2/5 | 2/5 |
| Audience | developer | developer | developer |
Figures from each repo's GitHub metadata at analysis time.
BATHOS is a method and toolkit for turning a single Claude Code session into something closer to a structured product team, rather than one long open ended conversation with an AI. The name comes from a Greek word meaning depth, chosen because the project aims for thorough, disciplined work instead of quick, surface level answers. The core problem it addresses is that a normal AI chat tends to lose context over a long project: the plan and the actual implementation drift apart, quality checks get skipped, and the same AI that writes the code is also the one judging whether it is good. BATHOS answers this by organizing the work into 17 specialist roles moving through a 7 wave delivery pipeline, with a router that adjusts how much process is applied depending on how big the task is, from a tiny fix up to a large project. Before implementation can begin, a hard readiness check has to pass, and separate reviewer roles check the work instead of the same role marking its own output. BATHOS runs on two layers. The first is orchestration: markdown based slash commands, role definitions, and hooks that live inside a project's Claude Code configuration, and these are what you interact with directly, such as typing a command to start the discovery wave. The second is a small Rust program called bathos that acts as the deterministic engine underneath, tracking state, enforcing the quality gates, and handling wave transitions and routing, so that these rules are consistent rather than left to chance. To use it you need Claude Code version 2.1.32 or later with an experimental Agent Teams feature turned on, a Rust toolchain to build the engine, and on macOS or Linux the jq command line tool plus a shell for the hooks, while Windows uses PowerShell instead. The project is released under the MIT license and documentation is available in English, Korean, Spanish, German, and Japanese.
A method and Rust powered engine that organizes a Claude Code session into 17 specialist roles and a 7-wave pipeline with enforced quality gates.
Mainly Rust. The stack also includes Rust, Claude Code, Markdown.
Mainly developer.
This repo across BitVibe Labs
double-check against the repo, no cap.