Plain Jane's Task Ask

Plain English in. Builder-ready spec out.

Calgary Cursor — May 2026 buildathon · calgary-cursor-2026-may

Live demo

Surface	URL
Production	https://calgary-cursor-2026-may.vercel.app
Stakeholder intake	https://calgary-cursor-2026-may.vercel.app/request
Developer Inbox	https://calgary-cursor-2026-may.vercel.app/inbox

Deployed on Vercel (Express @vercel/node, framework preset Other). Request data on serverless uses ephemeral /tmp storage — fine for demo; durable storage is out of scope for v1.

How it was built

The entire buildathon submission — product scope, UI, Express API, tests, specs, and deploy config — was authored with Cursor only. No Claude Code, Codex, GitHub Copilot, Cline, Kilo Code, Antigravity, Kiro, or other agentic coding harness was used to build this repository.

Summary for reviewers

Criterion	How this project addresses it
Everyday pain point	Vague, typo-ridden stakeholder asks hide scope before any code is written.
AI-native	Output is a minified PRD + ADR + Cursor prompt block, not prose for humans only.
Cursor fit	Stakeholders never touch GitHub; builders Copy for Cursor from the inbox.
Working demo	Live URLs above; core loop testable in under two minutes (see below).
Quality	npm test — 16 Vitest tests (normalization, spec writer, API).

Problem

Developers lose time before implementation: messy language, missing success criteria, and context trapped in chat or email. Plain Jane's Task Ask converts that intake into a spec-shaped, agent-ready brief stakeholders can submit without learning GitHub.

What it does

1. Request form (/request) — three steps, plain English, optional screenshot, optional Polish with AI (OpenRouter openrouter/free when OPENROUTER_API_KEY is set).
2. Normalization — rules-based cleanup (typos, goal minification, vagueness flags); optional LLM polish on the same pipeline.
3. Spec file — .spec-workflow/specs/requests/<id>.md with PRD (minified), ADR, Cursor prompt (minified), verbatim raw intake, GitHub issue draft.
4. Developer Inbox (/inbox) — numbered requests, preview, Copy for Cursor, status (new → in-cursor → done).

Example input: sales need to exprot dashbord to csv asap → cleaned goal, scoped open questions, paste-ready Cursor block.

Judge checklist (core loop)

1. Open /request — complete all steps (try intentionally messy text).
2. Open /inbox — confirm the new request appears with a stable # number.
3. Preview the spec — verify PRD, ADR, and Cursor prompt sections (not just raw text).
4. Copy for Cursor — clipboard should contain the minified agent brief.
5. (Optional) Run locally: npm install && npm test && npm start (Node 24+, see .nvmrc).

Architecture

/request  →  POST /api/requests  →  *.md spec on disk  →  /inbox  →  Cursor

Stack: Node 24 · Express · static HTML/CSS/JS · Vitest · Clay design tokens (DESIGN.md)

Deep dive (agents & implementers): AGENTS.md — API routes, spec schema, file map, conventions.

Environment variables

See .env.example. None required for the rules-only demo. Optional: OPENROUTER_API_KEY (+ OPENROUTER_SITE_URL on Vercel) for the AI polish toggle.

License

Apache License 2.0 — see LICENSE.