🪴 Brain

CLI-Anything

3 min read

CLI-Anything

🚀 Tomorrow’s software users will be agents, not humans. CLI-Anything bridges that gap — feed it a codebase or a GitHub repo and it auto-generates a full, structured CLI harness that LLM agents can drive directly. No UI automation, no brittle RPA, no dumbed-down API wrappers.

Distributed as a Claude Code plugin (also works with Pi, OpenClaw, OpenCode, Codex, Qodercli, GitHub Copilot CLI). Core idea: CLI is the universal interface for both humans and AI agents — structured, composable, self-describing, deterministic.

🗒️ Description

  • CLI-Hub: pip install cli-anything-hub then cli-hub install <name> to browse/install community CLIs.
  • Main command: /cli-anything <software-path-or-repo> — runs the full 7-phase pipeline.
  • Output: installable Python package with Click CLI, stateful REPL, JSON output mode, SKILL.md for agent discovery.
  • 30+ production harnesses: GIMP, Blender, Inkscape, LibreOffice, Zotero, Obsidian, Draw.io, Godot, n8n, Ollama, ComfyUI, Exa, QGIS, Mermaid, Shotcut, Kdenlive, etc. 2,152 passing tests across all.

Description

Download or use

  • Claude Code: /plugin marketplace add HKUDS/CLI-Anything → /plugin install cli-anything
  • Then: /cli-anything ./<path-or-repo> or refine with /cli-anything:refine

🧩 Key Features

  • 7-phase pipeline: Analyze → Design → Implement → Plan tests → Write tests → Document → Publish.
  • Authentic integration — generates valid project files (ODF, MLT XML, SVG) and delegates rendering to the real backend (Blender bpy, LibreOffice headless, FFmpeg).
  • Dual interaction model — stateful REPL + subcommand mode, both with --json for agent consumption.
  • SKILL.md per CLI — auto-generated in Phase 6.5, installed inside the package, discoverable by agents.
  • Refinement loop — /cli-anything:refine does gap analysis vs. full software capability and adds missing commands/tests incrementally.

🏗️ Architecture

Core design principles:

  1. Use the real software — no replacements, no toy renderers.
  2. Dual mode — REPL for agents + subcommand for pipelines.
  3. Consistent UX via shared repl_skin.py.
  4. --json on every command.
  5. Zero-compromise dependencies — backends are hard requirements; tests fail (not skip) when missing.

🎬 Demos

  • Draw.io HTTPS handshake diagram — agent produces .drawio + .png in ~4 min.
  • Slay the Spire II gameplay automation.
  • VideoCaptioner auto-subtitling.

Reasoning for

Very interesting for Agentic Systems and Qamera AI — every GUI-based tool that today would need UI automation (Playwright, screenshots) can be turned into a structured CLI with --json output. Especially for workflows with Blender / LibreOffice / Draw.io / n8n. Pairs with Agent Skills and Harness Engineering — the generated SKILL.md drops straight into .claude/skills/.

📖 Further reading

Template: tool

Graph View