Asking OpenAI to generate RSpec tests for your Rails application can be a powerful way to tackle a large amount of tech debt programmatically, or generate specs — depending on how much you trust your AI’s responses.

I found that by default, OpenAI’s GPT-3.5 and GPT-4 model would generate RSpec request specs using get ":create" or post ":update" instead of proper routes such as api_v1_users_path or api_v1_user_path(user.id). It was slightly tricky to generate a prompt with the exact wording so that OpenAI would only use proper routes. To implement this, I was able to use the Neovim plugin ChatGPT.nvim and create a custom actions file in the default folder where ChatGPT.nvim looks for custom actions: /Users/aaron/.config/nvim/lua/config/gpt/actions/rails.json. I couldn’t find another example of a gpt/actions/*.json file on Github, so I thought I actually might be the first one to do this, at least publicly on GitHub.

I hope you find this helpful!

“RSpec” prompt

Note that the backticks are there to cause AI to finish a “started” markdown code block, and the curly braces are template inputs for the Neovim command.

For the following instructions, do not use \"get :create\" etc and
instead use full routes like \"api_v1_birds_path\". Implement RSpec tests for
the following code.\n\nCode:\n```{{filetype}}\n{{input}}\n```\n\nTests:\n```{{filetype}}

“Missing method” prompt

You are a Ruby on Rails programmer.
Implement the {{method}} method(s) for the following code.\n\nExisting
code:\n```{{filetype}}\n{{input}}\n```\n\n```{{filetype}}

actions.json

{
  "WriteRailsMethod": {
    "type": "chat",
    "opts": {
      "template": "You are a Ruby on Rails programmer. Implement the {{method}} method(s) for the following code.\n\nExisting code:\n```{{filetype}}\n{{input}}\n```\n\n```{{filetype}}",
      "title": "Write Rails code",
      "strategy": "replace",
      "params": {
        "model": "gpt-3.5-turbo",
        "stop": ["```"]
      }
    },
    "args": {
      "method": {
        "type": "string",
        "optional": "true"
      }
    }
  },
  "WriteRSpecTests": {
    "type": "chat",
    "opts": {
      "title": "Write RSpec tests",
      "template": "For the folllowing instructions, do not use \"get :create\" etc and instead use full routes like \"api_v1_birds_path\". Implement RSpec tests for the following code.\n\nCode:\n```{{filetype}}\n{{input}}\n```\n\nTests:\n```{{filetype}}",
      "strategy": "append",
      "params": {
        "model": "gpt-3.5-turbo",
        "stop": ["```"],
        "max_tokens": 2048
      }
    }
  },
  "WriteRSpecShouldaFactoryBotTests": {
    "type": "chat",
    "opts": {
      "title": "Write RSpec tests",
      "template": "For the following command, use shoulda_matchers and assume FactoryBot factories exist for all models. Implement RSpec tests for the following code.\n\nCode:\n```{{filetype}}\n{{input}}\n```\n\nTests:\n```{{filetype}}",
      "strategy": "append",
      "params": {
        "model": "gpt-3.5-turbo",
        "stop": ["```"]
      }
    }
  }
}

ChatGPT.nvim command configuration

For lazyvim, configure the keys section of ChatGPT.nvim like so:

  {
    "jackMort/ChatGPT.nvim",
    dependencies = {
      "MunifTanjim/nui.nvim",
      "nvim-lua/plenary.nvim",
      "nvim-telescope/telescope.nvim",
    },
    cond = function()
      local api_key = os.getenv("OPENAI_API_KEY")
      return api_key and api_key ~= "" and true or false
    end,
    opts = {
      actions_paths = {
        -- Add the path to the actions.json file you've created, mine was called rails.json
        actions_config_path .. "/rails.json",
      },
      api_key_cmd = 'op item get "Neovim ChatGPT" --fields credential',
      openai_params = {
        model = "gpt-3.5-turbo-16k", -- "gpt-4"
        frequency_penalty = 0,
        presence_penalty = 0,
        max_tokens = 300,
        temperature = 0.3,
        top_p = 0.3,
        n = 1,
        -- Lower temperature and top_p can help with repetitive responses and make the AI more creative
        -- temperature = 0,
        -- top_p = 1,
      },
      openai_edit_params = {
        model = "gpt-3.5-turbo-16k",
        -- model = "gpt-4",
        frequency_penalty = 0,
        presence_penalty = 0,
        temperature = 0,
        top_p = 1,
        n = 1,
      },
      show_quickfixes_cmd = "Trouble quickfix",
    },
    cmd = {
      "ChatGPT",
      "ChatGPTActAs",
      "ChatGPTCompleteCode",
      "ChatGPTEditWithInstructions",
      "ChatGPTRun",
    },
    -- stylua: ignore
    keys = {
      -- ... default keys above
      { "<leader>Ra", "<cmd>ChatGPTRun WriteRailsMethod<cr>", desc = "Write Rails code" },
      { "<leader>Rt", "<cmd>ChatGPTRun WriteRSpecTests<cr>", desc = "Write RSpec tests", mode = "v" },
    }
  }

WriteRailsMethod video

WriteRSpecTests video

Back to top