SkillFoxx

Blade Code CLI

A coding agent for the terminal, browser and CI across 38 model providers, with cross-session memory, MCP, skills and hooks.

  • 181 ★ on the repository
  • Last commit
CLI

Broad access

Why this level

  • The agent runs shell commands and edits and deletes files, and `yolo` mode removes confirmations.
  • Project code and prompts go to the chosen model provider, and keys sit on disk.
All reasons and checks

echovic/blade-codeAuthor:echoVic

Install

In your terminal, with SkillFoxx CLI

npx skillfoxx add cli/blade-code

Detects the agents on your machine, shows the access level and pins the version.

Other ways to install

Access and checks

How we review

Broad accessThe tool can change data, run commands or handle secrets.

  • The agent runs shell commands and edits and deletes files, and `yolo` mode removes confirmations.
  • Project code and prompts go to the chosen model provider, and keys sit on disk.

This is third-party code. Review the repository files before installing.

What it does

Blade Code is a standalone coding agent with three modes: an interactive terminal CLI, a web interface via `blade web` and a headless `blade --headless` mode with JSONL output for CI and sandboxes. Underneath is one runtime on pi-ai that connects OpenAI, Anthropic, DeepSeek, Google, Bedrock and dozens more providers, with model metadata fetched automatically. It has more than twenty built-in tools: file editing, code search, shell, Git, page fetching and a browser on an isolated Chromium with ARIA snapshots. The agent remembers a project across sessions, delegates subtasks and runs Goal mode and specs. You extend it through MCP, plugins, skills and hooks. There are four permission levels, `default`, `autoEdit`, `plan` and `yolo`, plus tool allow and deny lists. Tokens, cache and a cost estimate for every turn persist in the session and `/cost` shows them.

Who it is for. Developers who want their own coding agent on any model with a web interface and CI runs, with no vendor lock-in.

Good fit when

  • A coding agent on DeepSeek or another non-mainstream model is wanted, with a single interface.
  • A task must run in CI and the result is parsed line by line from JSONL.
  • Several repositories make up the project and tasks are best handed out from a shared board.
  • The agent must open a site in an isolated browser and walk a page through an ARIA snapshot.

Not a fit when

  • Claude Code or Codex is enough and a second agent is redundant.
  • Your machine runs a Node.js older than 22.19.0.
  • You need Russian-language support: the README and interface are in Chinese and English.

Example request

Start blade in plan mode and draft a refactoring plan for the payment module, then show the changes without writing to files.

Limitations

Node.js 22.19.0 or newer is required. Provider keys are entered on first launch and stored in `~/.blade/auth.json` with `0600` permissions. The browser tool needs the pinned Chromium: `blade browser install`. The `yolo` mode removes confirmations, so the agent runs commands without asking. Documentation is mostly in Chinese and the English README is shorter. The latest commit is from October 2026, so the project is active, but release pace on young projects can be uneven.

How to disable. Uninstall the package with `npm uninstall -g blade-code`, remove the `~/.blade/` directory with settings and keys by hand, and delete project `.blade/` folders from repositories.

FAQ

How do I try Blade Code quickly?

Run `npx blade-code` with no install. For regular use install `npm install -g blade-code` and launch `blade`.

How do I run it in CI?

Use `blade --headless --output-format jsonl "task"`. Output arrives as JSONL events including a `token_usage` field per turn.

How do I choose a model?

In the first-run wizard: provider, model and key. Later `/model add` adds a model and `/model switch` changes it. Settings live in `~/.blade/config.json`.

Which permission modes exist?

`default`, `autoEdit`, `plan` and `yolo`. Tool allow and deny lists can be added on top.

How do I see token spend?

Use `/cost` in a session: it shows tokens, cache reads and writes and a cost estimate. The data persists after a restart.

Discussion

No questions or reviews yet.

CLI tools

A Python terminal coding agent in the manner of Claude Code: reads the repository, edits files, runs commands, with memory, sub-agents and MCP

CLIBroad access1K

A terminal coding agent tuned for the deepseek-v4 model: deep thinking, reasoning effort control, Agent Skills and MCP

CLIBroad access2.2K

Kode CLI is a terminal coding agent that reads AGENTS.md and CLAUDE.md, connects MCP servers and works with many models.

CLIBroad access5.2K

A terminal coding agent in Go: GLM, DeepSeek and local Ollama models, 59 built-in tools and MCP

CLIBroad access110

Over twenty skills for coding agents: MCP and hooks, plan mode, a bug-fix protocol, prompting and a bash safety hook

SkillBroad access178

README in short

The README covers three launch modes, a model setup wizard and a command summary. It describes the `packages/cli` architecture, permission levels and extensions through MCP, plugins, skills and hooks. Ready-made `/commit`, `/review` and `/test` commands and `.blade/skills` sit alongside.

README badge

Report a problem