VibesPlug Pro: A 176-Skill Multi-Agent Development Suite

Who It's For

• For developers who want to use multiple IDEs (Antigravity, Claude Code, Cursor, Windsurf, OpenAI Codex) in the same project while getting unified agent rules and a skill library; the official positioning is a universal 6-IDE and MCP connector, with a built-in MCP v1.x server supporting both stdio and SSE modes, 42 tools, and an 8-stage autonomous development workflow.
• Not suitable for environments with strict supply-chain security requirements: the officially recommended primary installation method is a one-line curl ... | bash command that executes a remote script from the repository's main branch — installing means running code that has not been reviewed locally. The repository does not state supported operating systems, Node/runtime version requirements, or whether an existing API key or credentials are needed before installation — all of these are unknown and must be verified yourself before use.

First Try in 3 Minutes

The only primary installation path officially listed is a one-line command. Run it in your project directory in a terminal (the installation location is not officially stated, and which paths the script writes to is also unknown):

curl -fsSL https://raw.githubusercontent.com/roedyrustam/vibes-plug/main/install.sh | bash

Side effects and checkpoints: this downloads and immediately executes install.sh from the main branch. Since the repository evidence does not include the script's contents, it cannot be confirmed which files it writes locally, whether it makes network callbacks, or whether it modifies existing IDE configurations. The safe approach is to first open the same URL in a browser and review the script, or run it in an isolated environment (container/VM).

For your first task after installation, choose a low-risk analysis-type task: ask any supported IDE agent for a codebase diagnostic or project analysis, rather than directly launching the official "8-stage zero-to-prod" autonomous workflow — the latter advances development stage by stage and may generate and modify files and execute commands. The success signal is whether the IDE agent loads the skills and tool list provided by VibesPlug; the exact presentation is not itemized in official documentation, but if the agent can list and invoke one of its 42 MCP tools, the connection is confirmed.

Note: this skill is unrelated to text-type use cases; it targets software development workflows, not writing. Don't expect dedicated flows for novel review or document polishing.

Use Cases

1. Unified rules across IDEs: open the same project in both Cursor and Claude Code and let the agents work with the agent rules and skill files generated by VibesPlug. The official docs mention that the vibes evolve command and the multi-platform Codex generator write/regenerate these rules and skill files inside your project — this is an explicit file-writing side effect; check the target directory contents before running to avoid overwriting your own agent configuration.
2. Tool access via MCP: in an MCP-capable client, configure it as an MCP server over stdio or SSE, and call one read-only tool (e.g., a query/analysis type) as a smoke test. It can be confirmed that "the tool list and connection method come from official descriptions," but each tool's specific name, parameters, and side effects are not listed in the available evidence. Before the first call, verify the descriptions in the client's tool list to avoid accidentally triggering a tool that modifies the repository or accesses external resources.

Not recommended as a first try: directly launching the "autonomous 8-stage zero-to-prod" workflow, which autonomously drives the full lifecycle from zero to production and has the widest side-effect surface.

Usage Boundaries

• File writes: both the install script and vibes evolve write files — these are the two side effects explicitly documented; whether the skill itself also modifies your source code when executed is not itemized in official docs and should be treated as unknown.
• Command execution and network: both the pipe-to-shell installation and the "autonomous development workflow" imply command execution and network access, but the boundaries are not delineated in official documentation, so no conclusion like "zero risk" or "won't touch your system" can be given.
• Platforms: the official docs name Antigravity, Claude Code, Cursor, Windsurf, and OpenAI Codex (the title says 5 IDEs, while the feature description says a 6-IDE connector — the count is inconsistent; defer to the repository's original text). This does not mean it is entirely unusable in other agents, only that it is not officially confirmed.
• Unverified key information: uninstall method, usage restrictions beyond the license, the complete list of 42 tools, what exactly profile masking changes, and the MCP server's startup command and port. Before use, consult the repository README and the install.sh source to fill in these gaps.