This repo contains the first MVP skeleton for Cursor for Product Managers:
- Flutter mobile app (iOS/Android) for product discovery + specs + build plan + PR scaffolds
- Backend API + workers to index GitHub repos, ingest evidence, generate opportunities/specs/plans/prompts, and create PR scaffolds
Cursor for Product Managers
Teams struggle to reliably go from messy discovery inputs (interviews, tickets, notes) to clear, evidence-backed product decisions and implementation-ready plans. AI tools help write code once requirements are known, but there’s no system that supports the full loop: evidence → decision → spec → build plan → agent-ready prompts → PR scaffold, with traceability.
- Seed startups: founders / early PMs / eng leads who want to decide what to build next and ship fast.
- Enterprise PM org pilots: a small working group (PM + design + eng lead) that needs traceability and review gates, without full enterprise procurement features on day 1.
An AI-native product discovery system that:
- connects to a GitHub repo from day 1 (including mobile apps),
- turns evidence into ranked opportunities with citations,
- generates a repo-aware spec and agent-ready task prompts,
- and can create a safe PR scaffold (previewed, human-triggered).
Mobile app (Flutter)
- GitHub connect → repo select → indexing status
- Repo Map (best-effort): screens/navigation, networking layer, persistence layer, analytics wrapper
- Evidence upload (text/markdown/CSV initially) + evidence viewer with extracted “atoms”
- Opportunities list (ranked) with “why” citations
- Spec v1 (structured editor) with citations + repo touchpoints
- Build Plan (tasks + dependencies) and “Generate agent prompts”
- PR scaffold preview + create PR (conservative levels)
Backend
- GitHub App/OAuth (GitHub first)
- Repo indexing pipeline (Flutter + Kotlin + Swift supported day 1; best-effort heuristics)
- Evidence ingestion + atom extraction + clustering (v1)
- Generation endpoints for opportunities/spec/plan/prompts (v1)
- PR scaffold creation:
- Level 0: PR with body + checklist only (no code changes)
- Level 1: add
specs/<feature>/artifacts (spec/plan/prompts) (minimal repo changes) - Level 2 (later): minimal wiring (feature flag + placeholder screen) only when confidence is high
- Safety rails: denylist sensitive paths, diff preview, audit log
- Full Figma authoring
- Deep product analytics connectors (Amplitude/GA4) beyond manual import
- Auto-merge or autonomous implementation
- Full enterprise suite (SSO/SCIM/data residency) beyond lightweight gates + audit log
- Connect GitHub → pick repo
- Index repo → browse Repo Map
- Upload evidence → view extracted atoms/themes
- Ask “what should we build next for onboarding?” → ranked opportunities w/ citations
- Select opportunity → Spec v1 (repo-aware)
- Generate Build Plan + per-task agent prompts
- Preview PR scaffold → create PR (Level 0/1)
- Activation: % of new workspaces that connect GitHub + complete first index
- Time-to-first-value: median time to first opportunity list < 15 minutes
- Trust: % of “why” claims with citations (target > 90% or labeled “Assumption”)
- Downstream: % that generate prompts and/or create a PR scaffold within 7 days
- Retention: weekly active workspaces generating specs/plans
- Repo inference wrong (mobile diversity) → show confidence + “why”, default to safe PR scaffolds (Level 0/1)
- Hallucinated rationale → citations required for claims; explicit assumptions section
- Dangerous repo writes → explicit user trigger + diff preview + denylist + audit log
- Latency on mobile → async jobs w/ progress; caching; partial results
- TestFlight + Play internal testing with a Discord cohort
- In-app Discord invite + “first spec in 15 minutes” guided flow
product-app/backend/Node.js Express API using SQLiteproduct-app/frontend/React app built with Vite