aiflow

Stop rewriting AI context. Every terminal window, always in sync.

aiflow keeps your AI context in sync across every terminal window. Drop a progress.md in your project — any window, any moment, one command to know where you are and paste the right context into your AI.

No daemon. No server. No config. Just three shell scripts and a markdown file.

Install

curl -fsSL https://raw.githubusercontent.com/warp-context/rightStage/main/install.sh | bash

git clone https://github.com/warp-context/rightStage
cd aiflow && bash install.sh

The problem

You open a new terminal window. You forget where you left off. You spend two minutes re-reading your notes before you can write a single prompt. You do this ten times a day.

30-second quickstart

Step 1 — create your task list (once per project):

mkdir -p .ai && cat > .ai/progress.md << 'EOF'
[1] Login UI        done
[2] API integration  doing
[3] Error handling   todo
[4] Unit tests       todo
EOF

Step 2 — inject context into your AI:

ai_inject .                          # print context
ai_inject -c .                       # print + copy to clipboard
ai_inject -c . "help me with retry"  # include your prompt too

Step 3 — commit with a task number:

git commit -m "[2] API integration complete"
# Next time you run ai_inject, [2] is automatically marked done

That's it. Any window. Any time. Always in sync.

Commands

ai_inject [options] [project_dir] ["your prompt"]

Generates the minimal context block to paste into your AI.

Options:
  -c, --copy      Copy output to clipboard
  -n, --no-update Skip auto-sync from git
  -h, --help      Show help

Output (auto-assembled, < 100 lines):

[CURRENT]    what you're working on right now
[NEXT]       the task after this one
[GOAL]       project goal (first 15 lines)
[API]        API spec (first 10 lines, optional)
[CONSTRAINT] tech constraints (first 10 lines, optional)
[ISSUES]     last 3 known failure notes (optional)
[INPUT]      your prompt

ai_progress [project_dir]

Reads your git log and updates progress.md automatically.

• Commit message contains [2] or matches a task name → marked done
• First unfinished task → doing
• Rest → todo

Runs automatically inside ai_inject (skip with -n).

ai_done [project_dir]

Marks the current doing task as done, promotes the next todo to doing.

File structure

your-project/
└── .ai/
    ├── progress.md        ← the only file you write
    └── ctx/               ← auto-generated, do not edit
        ├── 00_goal.md     ← project goal (optional, you write once)
        ├── 02_current.md  ← auto-generated
        ├── 03_next.md     ← auto-generated
        ├── 04_constraint.md  ← tech constraints (optional)
        ├── 05_api.md         ← API spec (optional)
        └── 07_issue.md       ← known issues log (optional)

Global fallback: ~/.ai/ctx/ — shared defaults across all projects.

Works with any AI

Native integration in Warp

A Rust port of aiflow's core logic has been proposed as a built-in feature for Warp Terminal — see warpdotdev/warp#10344 (tracking issue: #10342).

If merged, Warp will read .ai/progress.md directly and inject the current task + minimal context into every AI Agent / Warp AI session opened in the project — same behaviour as ai_inject, no shell scripts required.

Supported platforms

• macOS (zsh / bash)
• Linux (bash)
• Windows Git Bash

Design principles

1. Context is runtime state, not documentation — written for the model, not for humans
2. Current task first — CURRENT > NEXT > everything else
3. Must self-update — anything requiring manual maintenance will rot
4. Token-minimal — < 100 lines keeps the model sharp

Uninstall

rm ~/.local/bin/ai_inject ~/.local/bin/ai_progress ~/.local/bin/ai_done
rm -rf ~/.ai   # optional: removes global context dir

Contributing

See CONTRIBUTING.md. PRs welcome — especially for new shell environments or AI tool integrations.

If aiflow saves you time, consider starring the repo. It helps others find it.

License

MIT — see LICENSE

curl -fsSL https://raw.githubusercontent.com/warp-context/rightStage/main/install.sh | bash

[1] 登录UI     done
[2] API接入    doing
[3] 错误处理   todo
[4] 单元测试   todo

ai_inject .           # 查看当前状态
ai_inject -c .        # 复制到剪贴板
ai_inject -c . "帮我实现错误处理的 retry 逻辑"

git commit -m "[2] 完成API接入，支持token刷新"