Inspiration
The hackathon brief named a pain I recognized immediately: asking an AI to help you code, then watching it repeat a mistake you already fixed, forget a decision you made ten prompts ago, or hand back code that “works” but you can’t explain. I’ve felt that. The problem usually isn’t the model; it’s that I started building before I had written down what I was building, for whom, and what was explicitly out of scope.
That’s what inspired SkillSprint: a small, local-first app that turns the “plan before you build” habit into a concrete workflow. Instead of another todo list, it walks you through a short interview, generates the same planning artifacts the Devpost Learn Skill Pack asks for — scope.md, prd.md, and spec.md — and then gives you a timeboxed build checklist so the plan actually survives contact with the afternoon.
What it does
SkillSprint takes a messy project idea and turns it into a one-afternoon MVP plan, end to end.
You create a project, answer seven guided questions about the problem, the user, the single core action, what is out of scope, and how you’ll know it works. SkillSprint then generates:
scope.md — the problem, target user, in-scope core action, and explicit non-goals
prd.md — user story, requirements, acceptance criteria, and success signal
spec.md — the implementation approach, data model, screens, and build steps
A timeboxed build checklist with must-have vs. nice-to-have steps
The app stores everything in localStorage, lets you export the three Markdown files into your repo, and includes a focus timer and progress tracker. It runs entirely in the browser with no account, backend, or API key. The demo video shows the full loop: idea → interview → generated docs → exported files → checked-off build steps.
It’s for AI-assisted builders, hackathon participants, and anyone who has ever opened a coding agent before deciding what “done” means.
How we built it
I started from an empty folder, installed the Devpost Learn Skill Pack with:
bash npx skills add challengepost/learn-ai-basics --all -y Then I opened Cursor in that folder and ran run 1-start. The skill interviewed me about the idea and helped cut it down before any code was written. The planning docs it generated — scope.md, prd.md, and spec.md — are included in the public repo, as required.
The app itself is a Vite + vanilla TypeScript front end. I kept the stack deliberately small: no backend, no auth, no external AI calls. The “interview” is a deterministic question graph, and the documents are generated from a single project-state object using Markdown templates. That single source of truth keeps scope.md, prd.md, and spec.md consistent instead of letting them drift into three unrelated documents.
I used a simple timebox equation to keep the build honest.
Challenges we ran into
The biggest challenge was scope creep. My first version wanted AI-generated docs, cloud sync, user accounts, and GitHub export. The skill pack forced me to cut it down. I ended up with a local-only app that generates deterministic templates — less magical, but reliable, testable, and finishable in an afternoon.
The second challenge was making template-generated documents feel specific rather than generic. I solved that by making the interview questions do more work: each answer maps directly to a section in the generated docs, and the app shows examples so users don’t stare at a blank box.
The third challenge was consistency. Early on, editing the scope didn’t update the PRD or spec. I refactored around one project-state object and derived all three documents from it. That made exports predictable and made the app easier to test.
Finally, Markdown export had the usual edge cases: escaping user input, filenames, and multi-file downloads. I kept the export simple and explicit rather than adding a zip dependency.
Accomplishments that we're proud of
I’m proud that SkillSprint is a real end-to-end loop, not a mockup. From an empty folder, I shipped a working app that interviews you, generates the three planning docs, exports them, and tracks the build checklist. The repo includes the scope.md, prd.md, and spec.md files the skills generated, and the demo video shows the app working.
I’m also proud of the constraint: no backend, no API keys, no account. It runs locally and is easy to explain. That felt like the right proof of concept for an afternoon build.
What we learned
The main lesson was that planning is a feature, not overhead. The skill pack’s interview step forced decisions early — especially the “what is out of scope” question — and those decisions kept the build from sprawling.
I also learned to treat generated artifacts as data, not afterthoughts. Once scope.md, prd.md, and spec.md came from one state object, the app became simpler and the docs stayed aligned.
And I learned that reusable skills are a practical way to keep an AI coding agent on track. The workflow is repeatable: interview, plan, build, review. I’ll use it on the next project.
What's next for SkillSprint
Next I’d add an optional LLM-backed refinement step for the interview, so the app can ask sharper follow-up questions while still falling back to deterministic templates. I’d also add more project templates — CLI tool, API, mobile screen — and a local folder export using the File System Access API. Longer term, I’d like a team mode for shared planning docs and a “plan vs. actual” view that compares the timebox to what actually happened. The goal is to keep SkillSprint small, useful, and focused on the habit that matters: decide before you build.
Log in or sign up for Devpost to join the conversation.