Skip to main content

Scale Up Your Codebase, Cowboy - Huckleberry Turns Copilot Chat Into a Ticket-Taming Workhorse

· 5 min read
Tim Morris
Tim Morris
Huckleberry Creator

Projects sprawl quicker than a runaway cattle stampede, and release trains keep missing the station. Huckleberry harnesses Copilot Chat into a diligent foreman, slicing specs into bite-sized tickets so your team can scale from small posse to full tech rodeo without dropping the reins.

Meet Huckleberry​

Huckleberry is an experimental VS Code extension that turns Copilot Chat into a project manager who writes your to-do list for you. Mention @Huckleberry in the chat, ask it to carve up a requirements document, and it showers your repo with tidy little task files. Everything lives locally alongside your code, so git can keep an eye on it. Docs and source

Status check: the project is still in alpha and not on the Marketplace, so expect the odd rough edge and the occasional wandering cattle.

Installing the camp-fire chilli​

  1. Visit the releases page: https://github.com/CambridgeMonorail/vscode-huckleberry/releases
  2. Download the freshest huckleberry-taskmanager-x.x.x.vsix.
  3. In VS Code choose Extensions → … → Install from VSIX… and point at the file.

You also need:

  • VS Code 1.93+ (that is when the Language Model API landed)
  • GitHub Copilot plus Copilot Chat
  • Agent Mode switched on in settings

All three let the model call Huckleberry's tools directly and save you from typing slash commands like a cowpoke chiselling on stone.

Pick your model​

Agent Mode lets you choose the underlying large language model. The Huckleberry posse recommends Claude:

  • Claude 3.7 Sonnet for creative writing and specification generation
  • Claude 3.5 Sonnet for code-heavy conversations

Choose whichever gives you the cleanest diff and the fewest hallucinated Elvis lyrics.

Write the spec first, ride later​

Before unleashing the AI, give it something sensible to chew on. Drop a prompt like this into Copilot Chat:

I've been given a new requirement to add MY FEATURE NAME features.
Add a new markdown specification document for MY FEATURE NAME in my docs directory,
here are the current requirements: YOUR FEATURE DESCRIPTION GOES HERE

Huckleberry will create 'docs/my-feature-name.md' and fill it with a surprisingly coherent brief.

Sanity-check the brief​

Read the document. Out loud if necessary. If you already have a codebase, ask the agent to cross-reference:

Review our specification markdown document 'docs/my-feature-name.md'
against the existing code in 'apps/yourapp/src'
and add notes and details that will help the developer implementing this feature

Never forget the golden rule of AI-fuelled development: trust but verify.

Spawn the tasks​

Happy with the spec? Great. Issue the order:

read docs/my-feature-name.md and create tasks

Huckleberry will propose task files and politely ask permission before committing them. The uprising is pencilled in for a later release.

If the tasks look bare, try:

make sure all those tasks link back to the requirements document

Still thin? Close the loop:

Review the tasks in 'tasks/' against 'docs/my-feature-name.md'
and make sure each task has enough detail for a developer to implement it;
if something is not clear ask me

"Doc, I'm your Huckleberry: link those tasks to the spec."

Work the queue​

Type:

Next task

Sit back while Huckleberry serves up the first ticket, complete with context and a cheeky pep-talk. You'll be coding quicker than a saloon pianist after last orders.

Raise a hand, raise an issue​

Find a bug or dream up a feature? Wander over to https://github.com/CambridgeMonorail/vscode-huckleberry/issues and let the maintainers know. They're friendly and always appreciate a well-aimed ⭐ on GitHub.

Final thoughts​

Every webinar swears it can unlock 10× velocity. Huckleberry promises fewer Post-it notes and more time in the flow state. Install the VSIX, feed it a spec, and watch your coding session go from "Where did I put that to-do list?" to "Next task, please" in one dusty-trail trick.

Happy hacking, and remember: read the tasks before the tasks pull a high-noon standoff.

FAQs​

Q: How do I choose the right AI model for my development needs?​

A: Consider factors like the type of work you're doing (creative writing vs code-heavy tasks), model capabilities, and performance. For instance, Claude 3.7 Sonnet is recommended for creative writing and specifications, while Claude 3.5 Sonnet works well for code-focused conversations.

Q: What's the best way to use Huckleberry with specification documents?​

A: Start by having AI generate a markdown specification in your docs directory. You can then use Huckleberry to parse this specification into tasks, ensuring they link back to the original requirements. This creates a clear trail from specification to implementation.

Q: How can I maintain consistency when using AI-generated specifications?​

A: Use clear, detailed prompts when requesting specifications, review the generated content, and ensure all tasks link back to the specification document. Keep your requirements focused and use Huckleberry to validate task coverage against the specification.