Install CodeBuddy CLI and run Craft-style terminal coding
At a glance
A Tencent Cloud community deep-dive shows installing CodeBuddy CLI via npm, logging in (WeChat or GitHub/Google), running codebuddy /init for a project map, then using natural-language Craft workflows aimed at pro/DevOps users. This workflow is listed under sponsor, coding-agent, devtools, cli.
- Agents / tools
- CodeBuddy Code, CodeBuddy CLI, Craft Agent, Node.js, npm
Building a tool for these workflows? See weekly rates.
What they did
Install Node.js LTS, then npm install -g @tencent-ai/codebuddy-code (or install CLI from CodeBuddy IDE Craft). Run codebuddy to complete browser login—WeChat for the default DeepSeek path, or GitHub/Google when you need third-party models. In a project directory run codebuddy /init so the agent scans the repo and writes codebuddy.md (modules and deps). From there ask Craft/agent to inspect structure, propose a change plan, and execute small end-to-end tasks (scripts, refactors, deploy helpers). The article also walks a modular Next.js image-tools demo (resize, format convert, SVG→image/PPT, logo remove) published at cnb.cool/ai4meng/web_tools.
How it went
Author reports verifying CLI at codebuddy -version 1.7.0 after install, and that the generated online-tools demo ran basic features without further edits. Article cites Tencent-internal figures (AI-generated code share >50%) and claims ~40%+ average efficiency lift plus ~10MB footprint / ~3× faster startup vs IDE plugin—treat as vendor/community claims, not independent benchmarks.
Worth knowing
Windows may need admin PowerShell plus Set-ExecutionPolicy RemoteSigned for npm, and the default npm global path must be on PATH or codebuddy is not found. If the browser does not open for login, paste the terminal URL manually; stuck auth can be cleared with Ctrl+C and re-running codebuddy. Prefer /init early so later prompts stay grounded in codebuddy.md.
