siga:给 Claude Code 会话内进度条,构建/测试不再「假死」
开源插件 siga 为 Claude Code 提供可在历史消息位置持续刷新的进度条,自动包裹 npm/cargo/pytest/docker/make 及 tqdm、indicatif 等常见进度库,避免长构建时终端「看起来卡住」。
让 Claude Code 跑长构建或大测试套件时,对话区一旦滚过当前行,传统「改写同一行」的进度条就失效了——你只能干瞪眼猜它是否还活着,甚至误以为会话已挂起。开源插件 siga 专治这件事:把可观测性留在对话历史里,而不是藏进滚动条外的终端行。
据 @cutetoxicguy(约 2026-10-29)介绍,siga 能在已经打印过的历史消息位置持续更新动画进度条,而无需另开仪表盘或反复刷「still building…」。自动检测并包裹的常见命令包括:npm/bun 安装、cargo build/test、Go 编译、pytest、docker build、make,以及 curl/wget/git clone/rsync、ffmpeg/yt-dlp 等;并对 tqdm/Rich/Click/alive-progress(Python)、cli-progress(Node)、indicatif(Rust)、mpb/pb(Go)等进度库做了对接。仓库还附带约 30 个已测命令示例。
对编码代理工作流的意义是:长任务进度可见,减少误杀进程与重复重跑,也减少「看起来卡住就重开会话」造成的上下文丢失。原帖未写安装命令与支持的 Claude Code 最低版本;接入前应阅读仓库 README,确认进度包装不会改写 CI 退出码语义,并在代理权限边界内只对白名单命令启用包装,避免把生产部署脚本一并劫持。
siga: Live Progress Bars Inside Claude Code
🚀 Claude Code now has live progress bars that keep updating even after the terminal has already printed them. And it works with builds, tests, downloads, and almost any command.
Ever asked Claude Code to build a project, install dependencies, or run a large test suite, only to stare at the terminal wondering whether it's still working?
A new open-source plugin called siga solves that problem with persistent, animated progress indicators displayed directly inside Claude Code's conversation.
No separate dashboard. No endless "still building..." messages. Just one progress bar that updates in place.
And the technical implementation is surprisingly clever.
🎯 What makes siga different?
Traditional terminal progress bars update by repeatedly rewriting the current line.
But Claude Code's conversation history works differently. Once a message has been printed, the terminal generally doesn't redraw it.
siga can update a progress bar inside a line that was printed several messages ago.
For example, Claude starts compiling a Rust application.
The original message might show:
Building project... [████████░░░░] 40%
Claude can continue displaying other messages, and the same bar keeps updating in its original location.
When compilation finishes, the bar reaches 100%.
⚙️ Which commands does it support?
The plugin automatically detects and wraps common development commands.
▪️ JavaScript: npm, bun, package installation
▪️ Rust: cargo build, cargo test
▪️ Go: Compilation and build commands
▪️ Python: pytest and supported progress libraries
▪️ Containers: docker build
▪️ Build systems: make and custom build scripts
▪️ Downloads: curl, wget, git clone, rsync
▪️ Media processing: ffmpeg, yt-dlp
It also supports progress reporting from popular libraries:
▪️ Python: tqdm, Rich, Click, alive-progress
▪️ Node.js: cli-progress, progress
▪️ Rust: indicatif
▪️ Go: mpb, pb, progressbar
The repository includes 30 tested command examples covering different tools and progress-reporting mechanisms.
And importantly, siga preserves command output and exit codes, so a failed build still reports a failure.
🧠 What if a command doesn't report progress?
This is one of the most useful features.
Many build commands don't provide percentages or a reliable estimate of remaining work.
siga handles this in two ways.
- When real progress is available
It reads information such as:
▪️ Percentage completed: 45%
▪️ Completed steps: 12/40
▪️ ffmpeg processing timestamps
▪️ Docker build steps
▪️ Git transfer progress
▪️ Terminal progress escape sequences
It prioritizes the most reliable available progress signal.
- When no progress information exists
On an initial run, siga can show an activity animation.
After previous executions, it estimates progress using the median duration of the last five runs.
For example:
If your project usually takes 60 seconds to compile, siga can estimate how far along the next build should be based on elapsed time.
The indicator stops short of 100% until the command actually finishes.
It's an estimate, not a measurement of completed work, but it's much more informative than staring at a silent terminal.
🖥️ How does it update text that has already been printed?
Here's the interesting engineering trick.
siga uses different rendering techniques depending on the terminal.
👻 Ghostty
The plugin installs a specialized font and a small GPU shader.
▪️ The progress bar consists of specially encoded glyphs.
▪️ The shader recognizes those glyphs on screen.
▪️ siga sends updates through OSC 4 terminal palette escape sequences.
▪️ The shader redraws the progress indicator without requiring the original text to be printed again.
The terminal text stays where it is, while its visual representation changes.
🐈 Kitty
Kitty uses a completely different approach.
▪️ siga renders the progress bar as an image.
▪️ It displays that image using Kitty's graphics protocol and Unicode placeholders.
▪️ New image data is uploaded under the same image ID as progress changes.
▪️ The existing placeholders display the updated image.
Both implementations allow progress indicators to remain attached to their original messages.
And neither requires Claude to generate a new text response for every percentage update.
🔧 It also works outside Claude Code
Although the plugin is designed for Claude Code, siga has a standalone command-line interface.
You can manually wrap commands:
siga run -- make
For programs that render their progress only when connected to a terminal:
siga run --pty -- cargo build
For progress libraries that need direct integration:
siga run --shim -- python https://t.co/zwgHUXdpbg
You can even report progress from your own scripts using:
siga set 3/10
Then mark the operation as completed or failed with siga done or siga fail.
There are also customizable TOML configuration files.
▪️ Global configuration: ~/.config/siga/config.toml
▪️ Project-specific configuration: siga.toml
These let you define additional commands that should be monitored automatically.
📦 How to install it in Claude Code
The official installation requires Claude Code 2.1.275 or newer.
Run these commands inside Claude Code, one at a time:
/plugin marketplace add souvlakeee/siga
/plugin install siga@siga
/reload-plugins
/siga:setup
The setup process detects your terminal.
For Ghostty:
▪️ Shows the planned configuration changes
▪️ Requests confirmation
▪️ Installs the custom font and shader
▪️ Creates configuration backups
For Kitty:
No additional font or shader installation is required.
The plugin uses Python 3.11+ and relies on the standard library rather than external Python packages.
🛡️ Compatibility and limitations
The current implementation supports:
▪️ Ghostty 1.3+
▪️ Kitty 0.49+
▪️ herdr 0.9.5+, when used with a supported terminal
There are some restrictions worth knowing.
▪️ macOS: Tested by the developer
▪️ Linux: Expected to work, but not formally tested
▪️ Windows: Support remains unconfirmed
▪️ Other terminals: May display a static progress bar instead of live animation
▪️ Multiple jobs: Up to four independently tracked progress bars per Ghostty window before older slots are reused
▪️ Kitty: Progress bars share an image, so earlier bars may be replaced by newer ones
You can temporarily disable the integration with siga off, re-enable it with siga on, or remove its configuration using siga setup --remove.
💰 Pricing and availability
siga is completely open source under the MIT license.
▪️ No subscription
▪️ No API key
▪️ No external AI model required
▪️ No additional inference costs for the progress indicator itself
It monitors the commands your coding agent already runs.
💡 Why this matters
AI coding agents are becoming increasingly autonomous.
They can install dependencies, compile applications, run tests, build containers, and perform multi-step development tasks without constant supervision.
But their interfaces still struggle with something surprisingly basic:
Showing developers what's happening during long-running operations.
siga doesn't make your builds faster or your AI model smarter.
It makes the execution process more observable.
And because it uses real command progress when available, developers can distinguish between actual work being completed and a simple elapsed-time estimate.
It's a small quality-of-life improvement, but one that could make terminal-based AI development considerably more pleasant.
Sometimes the most useful AI coding tool isn't another model. It's a progress bar that actually works.
🔗 GitHub: https://t.co/5W1CfyZCea
#ClaudeCode #OpenSource #AICoding #DevTools #Terminal