by mitaku
A one-person software factory: AI coding agents take tasks from a ledger through review to merge, and call the human only for decisions only they can make.
Available after the catalog security scan passes.
⚠️ Third-Party Software Notice
This skill is third-party open-source software developed and hosted independently on GitHub. SkillsLLM lists this project for discovery and does not control or maintain the underlying repository.
Any security checks, ratings, or warnings displayed by SkillsLLM are automated and limited in scope. They do not constitute a security certification or guarantee that the software is safe, error-free, or free from malicious code, vulnerabilities, compromised dependencies, or prompt-injection risks.
Review the source code, permissions, dependencies, and configuration before installing or running any third-party skill. Use is at your own risk. To the maximum extent permitted by applicable law, SkillsLLM is not liable for losses arising from third-party software.
The deep catalog scan for this skill is still queued. Run an instant dependency check now instead.
# Add to your Claude Code skills
git clone https://github.com/mitaku/andonGuides for using ai agents skills like andon.
See how andon compares with popular alternatives.
Andon — a one-person software factory. AI coding agents take issues from a ledger through implementation, review and merge, and the human is called only for the decisions that only they can make.
一人で使うソフトウェア工場。エージェントが kanban の課題を実装・レビュー・マージまで流し、オーナーは自分にしかできない決断のときだけ呼ばれる。kanban は、プロジェクトごとの課題・問い・ルーチン・履歴を置くファイル群で、kanban コマンドで読み書きする。
What it is. Andon — a one-person software factory — is the package, the contracts and the documents for running a software "line" with AI agents. One long-lived container (a cell) is set up per trust boundary (one client, one hobby, ...). Inside it, a coordinating agent (the Chief of Staff) reads a file-based issue ledger, starts one worker agent per issue in its own git worktree, asks a reviewer agent of a different model family to review the pull request, and lets a gate decide from rules what can merge without a person. The human is called (the andon light turns on) only for five kinds of decisions: what to build, how users should experience it, promises to the outside, irreversible risks, and changes to the line itself. Which of these a person must answer can be delegated per project, except what to build.
The design calls this a selective dark factory. A "dark factory" is a factory that runs with the lights off because no person is on the floor. Andon is selective: machines stop what machines can decide (tests, gates), language models handle what needs judgment, and a person sees only what is left. The name comes from the andon cord and light of the Toyota Production System. Andon is not a manufacturing-floor andon product.
Why a cell. Andon is a one-person factory seen from two sides: the cell is the place (a boundary and a space, set up one per trust boundary), and Andon is the mechanism (the discipline kept inside that boundary). There is not one root for the whole of Andon but one per cell. The root directory of each cell is itself a git repository (the cell repo, /workspace in the container). Everything the cell's agents need hangs from that root: projects with their ledger, project instructions and skills (projects/<slug>/), the cell's wiki (wiki/), the cell's own instructions (AGENTS.local.md), and the client repositories it works on (.ghq/). So instructions, skills and the wiki are versioned, can be moved to another machine, and the cell can be rebuilt from a clone; the ledger stays outside git as a log that lives with the cell. The container image draws the boundary and carries the tools (Claude Code, Codex, herdr and hc, Playwright with a pre-installed browser, and Andon itself under /opt/andon), so every cell has the same tools at the same versions, the host needs only Docker and bash, and an Andon upgrade is one pull request that bumps FROM. Client repositories are only cloned into the cell; workers change them through pull requests from per-issue worktrees, and the ledger and the cell settings never go into them. Details (Japanese): The cell, ADR-0005, ADR-0016.
Who it is for. One developer who already runs several coding agents in terminals and wants them to keep working while the developer is away, without giving up control of merges and irreversible operations. It is built for its author's own use first.
Status. Experimental. It is used by one person, and the interfaces change without notice. Most documents are written in Japanese; English is added only where needed. There are no releases with compatibility promises yet; what changed in each tagged version, and what to do when upgrading, is in CHANGELOG.md. The host must be Linux or WSL with Docker and bash. The cell image bundles agent CLIs (Claude Code, Codex) and the terminal workspace herdr; you bring your own accounts for them.
Getting started. The step-by-step procedure (Japanese) is docs/operate.md: pull the image, create a cell repo with andon init <slug>, run bin/bootstrap, then start the Chief of Staff and run one issue. The commands are also in Set up a cell below.
Map.
| Path | What is there |
|---|---|
| docs/design.md | What Andon is, its vocabulary, the repo layout and tool choices (the main design document) |
| docs/vision.md | Vision, mission and values on one page, with a figure of the line (stages, roles, mechanisms, and what Andon does not have yet) |
| docs/contract.md, schema/ | Machine-readable contracts (ledger keys, findings, reports, questions, notifications, gate, delegation). schema/ is the source of truth |
| docs/operate.md | How to set up a cell and run one issue |
| docs/dispatcher.md, docs/ledger.md, docs/image.md | The dispatcher (resident notifier), the ledger CLI, the container image |
| docs/adr/, docs/decisions.md, docs/decisions/ | Architecture decision records and the decision log (one file per decision from 2026-11-07) |
| process/, docs/process/ | Definitions of the Chief of Staff's workstreams and of the state machines (YAML), and the pages rendered from them (ADR-0017) |
cmd/andon, cmd/kanban, internal/ |
The andon CLI and the ledger CLI (kanban), in Go |
package/ |
What is shipped into the cell at /opt/andon: skills, agent profiles, gate rules, managed settings, hooks, compose file, host templates |
image/, init/ |
The base image, and the template that andon init copies into a new cell repo |
tests/ |
Shell tests run on every pull request, including tests/public.sh (no secrets or personal values) |
Contributing and security. This is a one-person project and does not take pull requests for now. To report a vulnerability, see SECURITY.md.
License. Apache-2.0.
設計の言葉では「選択的 Dark Factory」と呼んでいる。Dark Factory は、床に人間がいないので灯りを消して回る工場のこと。Andon はそのうち、止められるものは機械(テスト・gate)、判断が要るものは LLM、決断が要るものだけオーナーが見る。名前は TPS のアンドン(監督者を呼ぶ灯り)から。製造業のアンドン製品ではない。設計は docs/ に PR で入る。public の repo(2026-11-09 から。秘密と、案件や個人に固有の値は入れない。docs/design.md §2.6)。
文書と commit メッセージにある Y-<数>(Y-123 など)は、前身の仮設の工場の kanban の番号で、この repo の課題の id(A-0001 の形)とは別物。履歴は書き換えない。
今の状態: 実験的で、1人で使っている。約束(コマンド・形式)は断りなく変わる。文書は日本語が中心。1人用の repo なので、今は PR を受け付けていない。脆弱性の知らせ方は SECURITY.md。
schema/(JSON Schema。tests/schema.sh が例と突き合わせる)kanban の約束(英語)process/ の YAML から andon process render が作る(ADR-0017)projects/<slug>/ の構成: 決定andon news ack): 決定<news> を混ぜない: 決定andon is an open-source ai agents skill for AI coding assistants such as Claude Code, Codex CLI, and ChatGPT, built by mitaku. A one-person software factory: AI coding agents take tasks from a ledger through review to merge, and call the human only for decisions only they can make. It has 117 GitHub stars.
andon's catalog security scan is still queued. You can run an instant dependency and prompt-injection check now with the "Scan for vulnerabilities" button above.
Clone the repository with "git clone https://github.com/mitaku/andon" and add it to your Claude Code skills directory (see the Installation section above).
andon is primarily written in Go. It is open-source under mitaku on GitHub, so you can review or fork the full source.
Yes. SkillsLLM lists many other AI Agents skills you can browse and compare side by side. Open the AI Agents category from the badge at the top of this page, or use the Related Skills and comparison links further down to weigh andon against similar tools.
No comments yet. Be the first to share your thoughts!