How It Works

This Neovim AI client enhances prompt engineering by allowing users to reference project-specific "rules" (via #) and files (via @) within prompts. These references are automatically resolved and injected into the AI's context for more relevant assistance. The agent supports multiple AI CLI backends, including OpenCode, Claude, and Cursor-agent, offering flexible AI model selection.

Quick Start & Requirements

• Installation: Integrate the provided Lua configuration snippet into your Neovim setup, typically via Lazy.nvim.
• Prerequisites: A supported AI CLI tool (opencode, claude, or cursor-agent) must be installed. The nvim-cmp plugin is required for autocompletion of rule and file references.
• Dependencies: Neovim, nvim-cmp, and a compatible AI CLI.
• Configuration: Setup involves specifying AI providers, logger settings, and completion sources in _99.setup().

Highlighted Details

• Contextual Prompting: Dynamically injects context by referencing custom SKILL.md files (rules) and project files via Neovim's completion engine.
• Multi-Provider Support: Integrates with OpenCodeProvider (default), ClaudeCodeProvider, and CursorAgentProvider.
• Visual Selection Input: A keymap (<leader>9v) enables using Neovim's visual selections as input for AI requests.
• Automatic Context Discovery: Automatically includes AGENT.md files found in parent directories relative to the current working directory.

Maintenance & Community

This project is explicitly ALPHA, with ongoing API changes and potential for severe issues. Bug reporting is handled via debug logs; feature requests are discouraged and will be closed. Public discussions for features are planned on Twitch.

Licensing & Compatibility

The provided README snippet does not specify a software license. Compatibility is primarily for Neovim. No explicit notes on commercial use or linking restrictions are present.

Limitations & Caveats

As an ALPHA product, expect instability and breaking changes. Known usability issues include incorrect virtual text placement for long function definitions, potential for visual selections to send entire files, and occasional "jacked up" code replacements. Prompting may require careful crafting to avoid partial replacements or unintended outputs.