dotagent.nvim

dotagent.nvim adds Claude Code and Codex-style /command, /skill, and
/prompt completion to your Ctrl+G prompt editor in Neovim, configured to your
local agent directories.

Requirements

• Neovim 0.10+
• saghen/blink.cmp

Release Status

The current recommended public tag is v0.4.0.

The plugin uses git tags for releases. There is no separate version file in the
repo.

Features

• Scans native agent roots for commands/, skills/, and optional prompts/
• Supports explicit command_dirs, skill_dirs, and prompt_dirs overrides
• Provides a Blink source for /command, /skill, and /prompt completion
• Uses distinct Dotagent menu icons by default: ⚡ for commands, 󰧑 for
  skills, 󰘧 for prompts
• Auto-attaches only when the editor session is launched with
  DOTAGENT_EDITOR_PROMPT=1
• Supports manual :DotagentAttach and :DotagentDetach
• Includes :DotagentRefresh, :DotagentBrowse, and :DotagentHealth

Install

Lazy example:

{
  "0xble/dotagent.nvim",
  config = function()
    require("dotagent").setup()
  end,
}

By default dotagent.nvim looks for native Codex and Claude roots:

• ~/.agent/shared
• ~/.agent/runtimes/codex
• ~/.agent/runtimes/claude
• ~/.claude

For each root it reads:

• commands/*.md
• skills/*/SKILL.md
• prompts/*.md, only when the prompts/ folder exists

Blink setup:

{
  "saghen/blink.cmp",
  opts = function(_, opts)
    opts.sources = opts.sources or {}
    opts.sources.default = function()
      local sources = { "lsp", "path", "snippets" }
      if require("dotagent").is_buffer_enabled(0) then
        table.insert(sources, 1, "dotagent")
      elseif vim.bo.filetype ~= ""
        and vim.bo.filetype ~= "markdown"
        and vim.bo.filetype ~= "text"
        and vim.bo.filetype ~= "gitcommit"
      then
        table.insert(sources, "buffer")
      end
      return sources
    end

    opts.sources.providers = opts.sources.providers or {}
    opts.sources.providers.dotagent =
      require("dotagent.completion.blink").provider()
  end,
}

Recommended Integration

The best integration is an external-editor flow where Claude Code or a Codex
launcher opens Neovim for prompt composition.

By default the plugin uses /, which is the more common command prefix in
editor and agent UIs.

Use one shared launcher for those flows:

#!/bin/sh
export DOTAGENT_EDITOR_PROMPT=1
exec nvim "$@"

Then point your agent surface at that launcher:

• Claude Code: configure its external editor, then use Ctrl+G
• Codex or an editor-backed wrapper like ai: set EDITOR to the same launcher

When Neovim starts with DOTAGENT_EDITOR_PROMPT=1, dotagent.nvim attaches
only to the initial prompt buffer. Additional buffers opened later in that
session stay unaffected.

Default Behavior

Without DOTAGENT_EDITOR_PROMPT=1, the plugin does not auto-attach in regular
buffers.

The fallback path is manual:

• :DotagentAttach
• :DotagentDetach

That keeps slash completion out of normal code editing by default.

Commands

• :DotagentAttach [bufnr]
• :DotagentDetach [bufnr]
• :DotagentRefresh
• :DotagentBrowse
• :DotagentHealth

Configuration

require("dotagent").setup({
  icons = {
    command = "⚡",
    skill = "󰧑",
    prompt = "󰘧",
  },
  activation = {
    mode = "contextual",
    env_var = "DOTAGENT_EDITOR_PROMPT",
  },
  agent_dirs = {
    vim.fn.expand("~/.claude"),
  },
  sources = {
    {
      type = "items",
      items = {
        {
          name = "ship",
          kind = "command",
          description = "Example Lua-defined item",
        },
      },
    },
  },
})

If your agent root has a custom prompt folder, dotagent.nvim reads it
automatically:

~/.claude/
  commands/
  skills/
  prompts/

Use per-type overrides when one kind lives elsewhere:

require("dotagent").setup({
  agent_dirs = {
    vim.fn.expand("~/.claude"),
  },
  prompt_dirs = {
    vim.fn.expand("~/custom-prompts"),
  },
})

Legacy sources path definitions still work. If you provide agent_dirs,
command_dirs, skill_dirs, or prompt_dirs, those directory lists become the
authoritative path configuration for their type, while items sources still
merge normally.

icons is optional. Override either kind if you want a different
source-specific menu icon, or set a value to "" to fall back to Blink's normal
kind icon for
that item kind.

If you want a different prefix, override it explicitly.

My setup keeps $:

require("dotagent").setup({
  prefixes = { "$" },
  activation = {
    mode = "contextual",
    env_var = "DOTAGENT_EDITOR_PROMPT",
  },
  command_dirs = {
    vim.fn.expand("~/.claude/commands"),
  },
  skill_dirs = {
    vim.fn.expand("~/.claude/skills"),
  },
})

activation.mode values:

• "contextual": attach only when launched with the editor prompt env marker
• "manual": never auto-attach
• "global": enable in every non-terminal buffer