Proposal: Improving LLM Context by Decoupling User Stories into Separate Files

[Katekko]

I'd like to propose a change to how we structure user stories for tasks. Instead of consolidating all stories into a single file, I suggest we create a tasks/ folder with each user story separated into its own file.

This aligns with the best practice of implementing story-by-story. It provides better context for LLMs by allowing us to feed them smaller, more focused chunks of information, preventing them from 'going wild' with a massive context window.

Does this cleaner, decoupled approach sound good to everyone?

[airtonix]

I've come to realise that speckit is a attempt by github to capture more github customers that vibe code.

As such, they're more than likely not going to:

• get rid of git behaviour in spec kit
• do anything that creates more confusion for the market segment that are using vibe coding as a "my first app" experience.

You might find that https://github.com/MrLesk/Backlog.md is a better fit.

I doubt the github team are going to listen to anything that requires them to deviate from the customer acquisition strategy they're using this project for.

[thlandgraf]

This is exactly the approach I landed on. I've been building a VS Code extension (SPECLAN) that stores each entity — goal, feature, requirement, scenario, acceptance criterion, test — as its own Markdown file with YAML frontmatter. The directory tree mirrors the hierarchy, so LLMs get focused context per file instead of one massive spec blob.

Full disclosure: I'm the creator. It's free and the essentials repo has sample specs and docs. Happy to share what worked and what didn't if you're exploring this pattern.