Star CLI
A TypeScript terminal agent: streaming REPL, OpenAI-compatible and Anthropic models, plan mode and edit rollback.
- 110 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The agent runs shell commands and edits files, with safety resting on the permission gate.
- Code and prompts go to the chosen model provider.
Install
In your terminal, with SkillFoxx CLI
npx skillfoxx add cli/star-cliDetects 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 agent runs shell commands and edits files, with safety resting on the permission gate.
- Code and prompts go to the chosen model provider.
This is third-party code. Review the repository files before installing.
What it does
Star CLI is a standalone coding agent for the terminal. The `/connect` wizard sets up a provider (OpenAI, Anthropic or OpenAI-compatible), after which the agent reads and edits files, runs commands and searches the web. Every risky action passes a permission gate, and file writes and edits show a diff before approval. It includes a read-only plan mode, subagents for narrow subtasks, `PreToolUse`, `PostToolUse` and `Stop` hooks, file snapshots with `/undo` and `/rewind` rollback, history compaction with `/compact`, session resume and `--json` output for scripts. Prompt caching is on for Anthropic, and re-reading an unchanged file returns a short note.
Who it is for. Developers who want a light open-source terminal agent with a choice of model and strict permission checks instead of a closed client.
Good fit when
- A task goes to a model from a third-party OpenAI-compatible provider.
- Before writing a file you want to see the diff and approve the edit.
- A complex task is first researched in plan mode and only then is code changed.
- A script consumes the agent output, so `--json` NDJSON is switched on.
Not a fit when
- You need deep editor integration: the tool runs only in the terminal.
- You need proven maturity: the project is young and the extension ecosystem is small.
- The environment has no Node.js 20 or newer.
Example request
Run `star`, connect a provider via `/connect`, then in plan mode find the cause of the failing test and propose a fix.Limitations
It needs Node.js 20 or newer and your own provider API key; Star CLI does not use other services' subscriptions. The agent runs shell commands and edits files, so permissions and diffs should be read rather than approved in bulk. The user pays for tokens, and the in-app cost estimate is approximate. The project is small, and configuration format and commands may change between versions. Some features such as prompt caching depend on the provider and do not work everywhere.
How to disable. Run `npm uninstall -g @cryer/star-cli`. Delete session history and settings manually from the configuration directory.
FAQ
Which models are supported?
OpenAI, Anthropic and any provider with an OpenAI-compatible interface. Setup goes through the interactive `/connect` wizard.
How do I undo a file edit?
With `/undo`, which rolls back a write from its snapshot. `/rewind` returns to a checkpoint.
Can it run without the interface?
Yes. `star -p "prompt"` prints the answer and exits, and `--json` emits events line by line for scripts.
What is plan mode?
The agent only reads and researches the project, then shows a plan. Edits begin after you approve it.
How do I cut token use?
Re-reading an unchanged file answers in one line, and auxiliary calls (titles, summaries) can go to a cheaper `smallModel`.
Discussion
No questions or reviews yet.
Related
CLI toolsLauncher and dashboard for long Codex, Claude Code and OpenCode sessions: tmux, a web panel and Tailscale links.
A teaching terminal coding agent in Python, built as a "nanoGPT for agents": read it, take it apart and fork it
An open coding agent for the terminal, browser and desktop that works with models from OpenAI and Claude to Ollama
Educational Swift CLI agent in the spirit of Claude Code: tool loop, subagents, skills, context compaction and tasks
mi is agentic coding in 30 lines: a loop, two tools and a model; works with any OpenAI-compatible API and runs in Docker.
An AI-agent workspace with parallel worktrees, cross-provider orchestration, and bots across chat channels.
README in short
Describes terminal agent features, requirements, installing from npm and from source, provider setup, slash commands and hooks.