GAAI workflow
A .gaai/ folder that splits agent work into discovery and delivery: a story with criteria, then isolated implementation and QA.
- 163 ★ on the repository
- Last commit
Broad access
Why this level
- The background daemon launches several agent sessions on its own, and they write code and merge it into the `staging` branch with no human in the loop.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/gaaiDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The background daemon launches several agent sessions on its own, and they write code and merge it into the `staging` branch with no human in the loop.
This is third-party code. Review the repository files before installing.
What it does
GAAI drops a `.gaai/` folder of markdown, YAML and bash scripts into a project with a ready structure for agent work. The `/gaai-discover` command starts a discovery agent that clarifies the task and writes a story with acceptance criteria into the backlog without executing anything. `/gaai-deliver` runs one story in the current session, and `/gaai-daemon` starts a background daemon. The daemon picks refined stories from the backlog and runs each in its own `claude -p` session inside tmux, three in parallel by default. Separate subagents plan, implement and verify, and the result passes a QA gate and merges into the `staging` branch. Decisions and patterns persist in project memory across sessions. Adapters exist for Claude Code, Codex CLI, OpenCode, Cursor and Windsurf.
Who it is for. Solo founders and senior engineers who know what to build and want an agent to ship it without improvising or forgetting decisions.
Good fit when
- An agent touches code it should not and a boundary of authority is needed.
- Decisions from earlier sessions get forgotten and need to live in project memory.
- Several backlog stories must run in parallel, each in its own session.
- Every implementation has to be checked against acceptance criteria before merge.
Not a fit when
- The project is small and a single `AGENTS.md` is enough.
- Ideas are still rough, while the framework suits those with product clarity.
- There is no `staging` git branch or tmux cannot be installed for the daemon.
Example request
/gaai-discover Add rate limiting: 100 requests per minute per user, a 429 response when exceeded.Limitations
The daemon needs a git repository with a `staging` branch, the Claude Code CLI or Codex CLI, `python3`, `node` and tmux. It was tested on macOS with Apple Silicon, while Linux and WSL are expected to work but are not confirmed. The Elastic License 2.0 is only partly open: it restricts, for example, offering the framework as a service. Launch scripts pin the environment strictly, and a shell with stray `GIT_*` or `PYTHONPATH` variables will not start the daemon.
How to disable. Stop the daemon with `/gaai-daemon --stop` and delete the `.gaai/` folder together with adapter files such as `CLAUDE.md` or `.cursor/rules/`.
FAQ
How do I install it?
Copy `.gaai/` into the project. The README gives two paths: a prompt for your agent, or cloning into a temp folder and running `install.sh --wizard`, which is then deleted.
How do discovery and delivery differ?
Discovery reasons and writes a story with criteria but never executes. Delivery executes but never decides scope, and runs in a separate process.
What does /gaai-daemon do?
It polls the backlog, takes stories with `refined` status, launches tmux sessions, retries failures and archives completed work. The `--status`, `--monitor` and `--stop` modes control it.
How is it different from BMAD and Spec Kit?
BMAD simulates an agile team, while GAAI is lighter on discovery and stricter on delivery. Spec Kit runs a spec-first pipeline, and GAAI adds QA gates, memory and a daemon.
Which agents work?
Claude Code is supported deeply with `CLAUDE.md` and commands in `.claude/commands/`. OpenCode and Codex CLI work through `AGENTS.md`, Cursor and Windsurf through their own rules files.
Discussion
No questions or reviews yet.
Related
WorkflowsA 13-skill delivery workflow for coding agents: from idea to plan, review, security audit, tests and a shipped pull request
A CLI that keeps RFCs, ADRs and work items as TOML in your repo and stops an agent from closing work until checks pass
Agent, skill, rule and template sets for 19 AI tools, from the software lifecycle to incident analysis
One desktop window for Claude Code, Codex, Gemini CLI and OpenCode: terminals, pull requests, diffs and approvals
Garry Tan's skill suite that turns Claude Code into a virtual team: product, architecture, design, review, QA and shipping
A CLI and skill that record an agent's goals, decisions and rejected paths in Git so the next agent does not repeat them
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Repository stars | Main difference |
|---|---|---|---|
| GAAI (this card) | no | 163 | |
| BMad Method | no | 55,033 | simulates an agile team of roles; GAAI is lighter on discovery, stricter on delivery and runs it with a background daemon |
| Spec Kit | no | 147,764 | builds a spec-to-tasks pipeline; GAAI adds QA gates, cross-session memory and parallel delivery |
README in short
The README shows a discovery and delivery dialogue, a comparison with AGENTS.md, BMAD, LangGraph and Spec Kit, the `.gaai/` layout and daemon setup.