ns-tasks — Mobile implementation tasks

<objective> Turn `.agents/plan/PRD.md` + `.agents/plan/ARCHITECTURE.md` into a set of implementable task files under **`.agents/plan/tasks/`** plus a `.agents/plan/tasks/README.md` overview. Final step of the pipeline: PRD → architecture → **tasks**.

Each task is a vertical slice an agent can finish autonomously in ~1–3h with a clear deliverable and a way to verify it. Mobile slices usually run schema + index → Convex function (built with the project builders) → Expo screen/route → wire-up → verify. </objective>

<process> ## Phase 1 — Read specs + the codebase 1. Read `.agents/plan/PRD.md` and `.agents/plan/ARCHITECTURE.md` (require both; if missing → `/ns prd` / `/ns architecture` first). 2. Skim the real structure so tasks cite real paths: `convex/` domains + `convex/functions.ts` builders, `mobile-app/app/` route groups, `mobile-app/components/ui/`, `site-config.ts`, and the `.agents/rules/` that match (convex, mobile-app, auth-payments-storage, verification).

Phase 2 — Decompose into slices + dependency graph

• Cut the architecture into vertical slices, not horizontal layers — a task should deliver a usable thing end-to-end (data → backend → screen), not "all the schema".
• Order by dependency: schema/foundation → core Convex functions → screens → monetization/entitlements → polish. Note which tasks block which.
• Right-size: 1–3h each. Split anything bigger; merge trivial ones.

Phase 3 — Verify before writing

Confirm: every PRD must-have maps to ≥1 task; every architecture table/function/screen is covered; dependencies are acyclic; each task has a concrete verification step. If gaps, analyze more — don't emit tasks with holes.

Phase 4 — Write the files

1. Create .agents/plan/tasks/ (numbered: 01-…md, 02-…md).
2. One file per task using the template below.
3. Write .agents/plan/tasks/README.md: project summary, execution guidelines (read the matching .agents/rules/ first, follow convex/_generated/ai/guidelines.md), and the phased checklist with dependencies. </process>

<task_template>

# Task [NN]: [Action-oriented title]

## Context
[1–2 sentences linking back to the PRD feature / ARCHITECTURE section this delivers.]

## Scope
[The specific, end-to-end deliverable for this slice.]

## Implementation
**Files to create/modify**
- `convex/<domain>/<file>.ts` — [function(s), built with `authQuery`/`authMutation`/`adminQuery`…]
- `convex/schema.ts` — [table + indexes, if any]
- `mobile-app/app/(<group>)/<screen>.tsx` — [screen]
- [web `web-app/app/routes/…` if the `/app` surface is touched]

**Key functionality**
- […] (server-derived identity, validators on args, `.withIndex` not `.filter`)

**Patterns / rules to follow**
- `.agents/rules/convex.md`, `.agents/rules/mobile-app.md`, `.agents/rules/auth-payments-storage.md` as applicable
- Route-group paths (e.g. `/(flow)/paywall`); IAP iOS / Stripe Android-web; entitlement user-level + idempotent grants

## Success Criteria
- [ ] [Observable outcome 1]
- [ ] [Observable outcome 2]

## Verification
- `npx convex dev --once` (Convex changes)
- `cd mobile-app && npx tsc --noEmit` (+ `npm run lint` if shared UI)
- [Simulator/manual step for nav/auth/paywall/native behavior]

## Dependencies
- Blocked by: Task [N] — [name] (or "none")
- Blocks: Task [X]

**Phase**: Foundation / Core / Monetization / Polish

</task_template>

<constraints> - Require both `.agents/plan/PRD.md` and `.agents/plan/ARCHITECTURE.md`. - Vertical slices, not layer-dumps. Each task independently verifiable. - Tasks must reference REAL paths, the project builders, and the matching rule files — and carry the repo's verification commands, not generic "run tests". - Respect Convex rules (indexes, no unbounded scans, validators, server identity) and the purchase-safety rule in every relevant task. - Write under `.agents/plan/tasks/` with a `README.md` index. Don't dump tasks into the repo root. </constraints> <verification> - Every PRD must-have → ≥1 task; every architecture element covered; dependency graph acyclic. - Each task has files, success criteria, and a real verification command. - `.agents/plan/tasks/*.md` + `.agents/plan/tasks/README.md` written. Ready to implement (e.g. via the `apex` or `oneshot` skills). </verification>