Feature request: add interactive q&a process for proposal creation

[dchuk]

This is a great project good job!

Something I’ve found works really well when planning new features with ai is to have it ask me a series of questions to refine the scope of the idea before creating a plan. This project is an example of such a process: https://github.com/rizethereum/claude-code-requirements-builder

It would be a killer feature in your project to have something like “proposal-interactive” that triggers the same kind of workflow. What do you think?

If I have time today I can try and experiment with it to see if I can get it to work

[KeentGG]

Oh perfect, I've used this one months before, way before "spec" workflow became mainstream. This was effective for me on planning out complex features/issues. Stopped using when i moved of from CC.

The only caveat to this is it eats up credits for credit-based AI agents like Cursor, windsurf, augment... unless you're using feedback MCPs like me.

I'd use this if it were available.

[TabishB]

Hey @dchuk @KeentGG Thanks for suggesting this! I'm keen to try something like this out!

Like I said in the discord we do want to give people the option to choose between the 2 types of workflows, so maybe a seperate slash command here would work best.

I'd be keen to learn more about about the type of questions you'd like to ask. What do you mainly care about here? Is it the scope of the requirements? Getting the requirements right? Keeping them minimal?

Would love to know what questions you've found useful so far - ones that have made you go "I'm so glad it asked me that". Or ones you really wished it asked you.

If you guys could reach out to me through any of these channels, I'd appreciate it, so I can build this out in a way that makes sense!

Reddit: https://www.reddit.com/user/Narrow-Breakfast126/
X: https://x.com/0xTab
Discord: tab_dev

[KeentGG]

What i love about it before was that you'll be able to know more about your own feature because it asks you more info, and the "Targeted Context Gathering" as that repo calls it which is accurate. So it wasn't necessarily the scope of requirements, nor keeping it minimal, but it was good at finding necessary context before implementation.

I've had fond memories with it because i was testing before doing several feature/bug fixing with and without that framework, around 70% of the time i had better output when it's used. The 30% it did not, doesn't necessarily mean it had produced bad output, just wasted time reading/answering questions with little to no noticeable better output.

Would love to know what questions you've found useful so far

I cannot recall really since it was ~4 months ago i last used CC (i mean i can use it anywhere now since it's just a slash command, only CC had slash commands before - but i prefer optimizing my workflow one at time now which is openspec). However, just in a hindsight memory i think questions like these ones that i made up:

• "We already have this feature X that has Y function which is a similar pattern, should we follow Y function?"
• "Will user need to this X stuff?"
• "Just a recommendation, should we also implement X feature?"
• "Would you like if we add this error fallback?"
• "There's a possibility of having this edge case issue, should we mitigate it by this proposed X function?"
  etc...

Out of 10 questions it asks, maybe only half of it was good and not just common sense or obvious answers, and a fraction is "I'm so glad it asked me that" moments which is already sufficient to steer the AI agent with better guidance than without it.
Sometimes i was even able to improve the idea of the feature as i kept answering questions due to moments "oh you're right, you got a point, i missed that".

I totally agree exposing this as an optional command since not all have time to read and answer discovery questions.

I am open to ideas too while still thinking on how can we improved upon that idea without going overboard.

I'll share once i got new ideas.

[TabishB]

I've added a draft proposal for this here: #97

@KeentGG @dchuk Let me know if you guys have any thoughts on the proposal :)

[KeentGG]

i've added comments on the PR itself but yea i'm eager to see the proposal implementation to test it out

[SlavaKatiukha]

@TabishB , Thanks for the great tool! A q&a process is definitely something that’s missing right now.
As a workaround, I’m using a custom slash command: /openspec-plan-proposal [your initial proposal] (source: openspec-plan-proposal.md). You can run it in Plan mode in your preferred AI agent tool. For example, when run in Plan mode in Cursor, it opens a question wizard UI with clickable items. After the plan is generated, clicking Build automatically runs /openspec-apply with the clarified proposal, which usually gives better results for the generated proposals.