AIGC · AI Programming

CC Switch Adds a GUI for Managing AI Coding Skills, Prompts, and Sessions

By 小溪彼岸 · Jun 27, 2026

Read original on juejin.cn ↗ Google Translate ↗ Alt translation

As the number of AI coding agents multiplies, keeping Skills and prompts synchronized across Claude, Codex, Cursor, and others becomes a maintenance tax. A GUI that centralizes this with symlinks cuts that tax and makes the setup accessible to developers who won't touch a terminal-based manager.

Summary

CC Switch v3.16.1 replaces terminal-based Skill wrangling with a GUI that handles installation, configuration, and updates across multiple AI coding tools. Skills can be restored from backups, imported from ZIP archives, pulled from existing Claude or Codex directories, or discovered from GitHub repositories and skills.sh. Once installed, enabling a Skill for a given tool simply creates a symlink in that tool's skills directory — no file duplication.

A new prompt manager lets users define reusable system prompts with a name, description, and body, then toggle them on to inject the content directly into agent configuration files like `~/.codex/AGENTS.md`. Session management aggregates conversation history from all supported AI tools into a single searchable, filterable list, with the ability to inspect full prompt chains, tool calls, and outputs, then restore a session into a terminal or delete it in bulk.

The symlink architecture is the key design choice: one canonical copy of each Skill lives under `~/.cc-switch/skills`, and every tool that needs it gets a lightweight pointer. This avoids the drift and disk waste of maintaining separate copies per tool, while the GUI lowers the barrier for developers who find TUI-based workflows friction-heavy.

Takeaways

— CC Switch v3.16.1 provides a GUI for managing Skills, prompts, and sessions across multiple AI coding tools.

— Skills can be installed from local backups, ZIP files, existing agent directories, GitHub repos, or skills.sh.

— Enabling a Skill for a tool creates a symlink in that tool's skills directory; the source file stays in `~/.cc-switch/skills`.

— A one-click "Check for Updates" scans all installed Skills for newer versions.

— Custom prompts are saved with a name, description, and body, and toggling them on writes the content into agent config files like `~/.codex/AGENTS.md`.

— Session management shows conversation history from all supported tools, filterable by tool, with full prompt and tool-call visibility.

— Sessions can be restored into a terminal (configurable: Terminal.app, Warp, etc.) or deleted individually or in bulk.

— Symlinks are the underlying mechanism: `ln -s` creates a pointer, `ls -l` inspects it, and `unlink` removes it.

Conclusions

Symlink-based Skill sharing is a pragmatic alternative to per-tool duplication — it keeps a single source of truth without requiring each agent to support a shared plugin directory natively.

The prompt manager effectively turns `AGENTS.md` and similar config files into GUI-managed, toggleable modules, which could reduce the friction of maintaining project-specific system prompts.

Session restore into a terminal is a small feature with outsized utility: it lets a developer pick up an AI-assisted debugging session exactly where they left off, across tools.

The limitation to global Skills and MCP means this release targets personal workstation setup rather than project-level or team-shared configurations, which is where the next complexity layer lives.

GUI-based Skill discovery from GitHub and skills.sh lowers the threshold for less terminal-comfortable developers to adopt community Skills, potentially widening the audience for AI coding assistants beyond CLI power users.

Concepts & terms

Symlink (symbolic link)

A special filesystem entry that points to another file or directory by path, rather than storing its own data. Creating a symlink with `ln -s` makes the target appear at the link location without duplicating content; deleting the symlink with `unlink` removes only the pointer, not the target.

Skill (in AI coding tools)

A packaged set of instructions, context, or tool definitions that extends an AI coding agent's capabilities. Skills are typically stored as directories containing markdown or configuration files and are loaded by agents like Claude Code or Codex at runtime.

MCP (Model Context Protocol)

An open protocol for connecting AI models to external tools and data sources. MCP servers expose resources, prompts, and tools that AI agents can discover and call during a session.

AGENTS.md

A configuration file used by OpenAI's Codex CLI (and compatible tools) to inject system-level instructions or prompts that apply across all sessions. It functions as a persistent, file-based system prompt.

Source: juejin.cn ↗ Google Translate ↗ Backup ↗