Kandr training playbook

How we build apps, from idea to launch

Talk messy in Cursor. Lock in a plan. Then loop build → QA → review for every phase until you ship.

Cursor — Plan Mode

Ask me c

The whole process on one map

Plan once in a straight line. Then loop build → QA → review for every phase until you ship.

Part 1 — Get to a plan

Do this once at the start — or again when scope changes.

Linear · 4 steps
1

Talk

Voice mode or long-form chat. Jump around — the agent fills gaps with follow-up questions.

2

Synthesize

Plan Mode (Shift+Tab). Messy thoughts become a reviewable markdown plan.

3

Review

Edit the plan. Ask what most builders skip — auth, admin, personas, AI layer.

4

Segment

Break the plan into phases, each with clear done criteria.

You have a phased plan — now execute one phase at a time

Part 2 — Build, QA, repeat

Run this loop for every phase. Do not start the next phase until QA passes.

Loop per phase

Build

Implement the current phase with the agent. One phase at a time — no jumping ahead.

QA

Code review, quality review, and sanity checks. Catch bugs before you stack on top.

Review & tweak

Design review, fix issues, adjust scope. Does it match what you planned?

Loop until this phase passes QA

Phase passes QA

Move to the next phase in your plan — or ship when you hit Launch.

Serious builds: Cursor Pro (medium tier) or higher — Plan Mode, agent runs, and longer context matter once you are past a toy demo. Prompts for each step live in the .

How modern apps are layered

Think bottom-up when you plan. Click any layer or capability chip for detail.

Bottom → top

Plan-level checklist

Ask these when you review your plan — before you write feature code. Full guidance lives in the Approach tab.

  • Do users need to log in?

    Marketing sites usually do not. Apps with accounts do — decide how (email, Google, Apple).

  • Account management and profiles?

    What can users edit? What is stored per user vs shared?

  • Admin or internal tools?

    You cannot run a real product from the database console forever.

  • Personas or customer segments?

    Different roles may need different screens, permissions, or data.

  • Intelligence layer built in?

    AI features, agents, RAG, or MCP tools — design them in, not bolted on late.

Tools in the loop

Click any chip for setup detail. Toolbox checklists are in Approach.

Three tiers of agent knowledge

Global skills teach the how. Project rules teach the what. Both work together.

Tier 1

Global process skills

~/.cursor/skills/kandr-*

How Kandr plans, deploys, QA’s, and integrates Firebase, Stripe, email, and more.

Tier 2

Generic shareable skills

cursor-build-playbook repo

Same workflows with org-agnostic placeholders — safe to share with teammates.

Tier 3

Project rules & overlays

.cursor/rules/, AGENTS.md

This repo’s IDs, routes, collection names, deploy targets, and safety rules.

Kandr skillGeneric equivalentProject layer
kandr-developmentPhased planning in BUILD_PLAYBOOK.mdAGENTS.md + extend-before-create gate
kandr-firebaseFirebase setup in project-setup-guideproject-context.mdc + kandr-overlay.mdc
kandr-deployDeploy checklist in project-setup-guideDeploy table in overlay rules
kandr-qaVerification steps in BUILD_PLAYBOOKProject-specific test commands
Download the shareable skills pack

Copy skills to ~/.cursor/skills/ and rules to ~/.cursor/rules/

Ready for prompts and checklists?

Approach has copy-paste prompts, the phased plan example, toolbox setup, and your first-session checklist.

Questions about the playbook?

This is the exact process we use on our own products. If you want help applying it to yours, we’re happy to talk.