RunLoyal — How We Operate

Phase 01 · Intake & Clarification

The PO writes
the requirement.

A one-liner in Slack is enough to start. The AI takes that raw input and returns a structured feature definition, user stories, and measurable success criteria — ready for the whole team to work from.

OWNED BY

📋 Product Owner

01

STEP 01 Product Owner

Write the requirement

No template needed. Drop it in Slack, create a Linear issue, or write it in Notion. Even a single sentence works — that's the input to the AI prompt.

Can be as short as one sentence
Posted in #product-requirements Slack or Linear
No format required at this stage

💬 Slack

📋 Linear

📓 Notion

📝 Raw requirement text

STEP 02 Product Owner · AI assisted

Run Stage 01 AI prompt → Feature Clarification

PO pastes the raw requirement into the Stage 01 Claude prompt. AI returns a complete, structured feature definition. PO reviews every line and corrects anything wrong.

Open stage-01-feature-clarification.md, copy the prompt
Paste requirement into the placeholder field
Review output — PO is accountable, AI is a drafting tool
Save the 3 output files to the feature folder

AI PRODUCES

01-feature-definition.md · Field specs, who uses it, data model
01-user-stories.md · "As a KO, I want to..."
01-success-criteria.md · Measurable pass/fail conditions

🤖 Claude

🗂️ Linear AI

📄 01-feature-definition.md

📄 01-user-stories.md

📄 01-success-criteria.md

🚦

Gate 1 — PO Sign-off

PO confirms the feature definition matches intent. All open questions answered. Nothing moves forward until Gate 1 is cleared. Tech Lead is notified to start Phase 2.

GATE 1 of 5

Phase 02 · Technical Assessment

Tech Lead maps
the risk.

Before anyone writes code, the Tech Lead maps where the feature touches the codebase and assesses every risk. No HIGH risks can be unmitigated before Gate 2 closes. This is also where the timeline estimate goes back to the PO.

OWNED BY

🔧 Tech Lead

02

STEP 03 Tech Lead · AI assisted

Run Stage 02 → Check Existing Flow

Map exactly where this feature plugs into the codebase. Entry points, exit points, shared components at risk, file-level dependency table.

Paste Stage 01 outputs + relevant code snippets into Stage 02 prompt
Cursor searches codebase for every dependency named in the output
Mermaid diagram pasted into mermaid.live → shareable link for team

AI PRODUCES

02-flow-map.md · entry/exit points, dependency table, Mermaid flowchart block

🤖 Claude

⚡ Cursor

🔷 Mermaid Live

📄 02-flow-map.md

🔷 Flow diagram

STEP 04 Tech Lead · AI assisted

Run Stage 03 → Impact Analysis

Full risk register: DB migration safety, API changes, frontend blast radius, performance. Every HIGH risk must have a mitigation before Gate 2 closes.

Paste Stage 01 + 02 outputs + DB schema into Stage 03 prompt
Tech Lead personally reviews every HIGH risk item
Timeline estimate generated and shared with PO
Risk register goes into Notion with owners assigned

AI PRODUCES

03-impact-list.md · DB / backend / frontend / perf impacts
03-risk-mitigation.md · Risk register with likelihood, severity, fix

🤖 Claude

🏗️ dbdiagram.io

📓 Notion AI

📄 03-impact-list.md

📄 03-risk-mitigation.md

🚦

Gate 2 — Tech Lead Sign-off

No HIGH risks unmitigated. DB migration strategy confirmed. PO notified with timeline estimate. Green light to start design.

GATE 2 of 5

Phase 03 · Feature Design

Full technical spec
before anyone codes.

Stage 04 produces the complete blueprint — API contracts, DB schema and migration, Angular component tree, and the UI stitch. The whole team reviews async in Notion. No engineer picks up a ticket until Gate 3 is approved.

OWNED BY

🔧 Tech Lead

+

💻 Senior Engineer

03

STEP 05 Tech Lead + Senior Eng · AI assisted

Run Stage 04 → Feature Design

The single most important document in the pipeline. Everything engineers build from lives here. AI drafts it, humans own every decision in it.

🗄️ Database Design

SQL migration (additive-only). TypeORM entity decorators. Index strategy. Rollback plan.

🔌 Backend Design

NestJS API endpoints. Request/response DTOs. Auth requirements. Validation rules.

🅰️ Frontend Design

Angular component tree. Full UI stitch prompt. NgRx state actions. Form validation logic.

Paste all prior stage outputs into Stage 04 prompt
Figma AI generates mockups — matched to Angular Material component patterns
Claude generates Angular UI stitch: full module, component, service wired together
DB migration reviewed and approved by Tech Lead before it's written
Design doc shared in Notion for async team review — minimum 24h review window

AI PRODUCES

04-backend-design.md · API contracts + DTO shapes
04-database-design.md · Migration SQL + TypeORM entity additions
04-frontend-design.md · Component tree + Angular stitch prompt

🤖 Claude

🎨 Figma AI

⚡ Cursor

🏗️ dbdiagram.io

📄 04-backend-design.md

📄 04-database-design.md

📄 04-frontend-design.md

🚦

Gate 3 — Design Review Approved

Whole team async-reviews in Notion. Tech Lead resolves any conflicts. Engineers assigned to tasks in Linear. Sprint begins. No ticket picked up before Gate 3 closes.

GATE 3 of 5

Phase 04 · Build

AI scaffolds,
Engineer owns.

Cursor + Claude generates all the files from the design doc — Angular components, NestJS service, TypeORM entity, migration. The engineer reviews every line, fixes what's wrong, and opens the PR. AI cuts scaffolding time by ~50%. The engineer owns quality.

OWNED BY

💻 Engineer

04

STEP 06 Engineer · AI assisted

Run Stage 05 → Development

The engineer pastes the frontend design doc into Cursor Composer — and gets back every file, wired together, matching the existing codebase style.

Cursor Composer: paste 04-frontend-design.md → generates Angular module, components, service, all imports wired
Claude Code handles NestJS backend: writes controller + service, runs ng build, self-fixes TypeScript errors
Migration file written — Tech Lead reviews before it runs on any environment
PR opened — GitHub Copilot fills PR description from 05-pr-summary.md
Tech Lead code review — mandatory, even for AI-generated code

AI PRODUCES

All Angular component files · NestJS controller + service · TypeORM entity additions · Unit test stubs · 05-pr-summary.md

⚠️ Non-negotiable rule

Code review by Tech Lead cannot be skipped — AI-generated code still requires human review. No merge without approval.

⚡ Cursor Composer

🤖 Claude Code

🐙 GitHub Copilot

🅰️ Angular components

🗄️ Migration file

🐙 PR opened

🚦

Gate 4 — Code Review Approved

Tech Lead reviews and approves PR. Migration reviewed separately. CI must pass (lint + unit tests). No merge without approval. Migration never runs on prod without Tech Lead sign-off.

GATE 4 of 5

Phase 05 · Quality Assurance

QA sign-off is
the deploy trigger.

Claude generates Cypress e2e tests and Angular unit tests directly from the success criteria written in Stage 01. The V1 regression suite runs unchanged — confirming nothing broke. Security checklist cleared. QA sign-off is what starts the deploy, not a formality.

OWNED BY

🧪 QA

+

💻 Engineer

05

STEP 07 QA + Engineer · AI assisted

Run Stage 06 → Testing & Regression

The success criteria from Stage 01 become the test case table. Claude generates runnable Cypress scripts and Angular TestBed unit tests from that table automatically.

Paste 01-success-criteria.md + 03-risk-mitigation.md into Stage 06 prompt
Claude generates the full test case table + Cypress .cy.ts file
Run ng test (unit) + ng e2e (Cypress) — all must pass
V1 regression suite runs unchanged — any failures block the release
Security checklist: JWT auth, input validation, XSS, no sensitive data in responses

New feature tests

All success criteria covered · HIGH priority tests must pass · Performance within threshold

Regression (V1 suite)

All prior tests run unchanged · Zero regressions allowed · Existing tabs / flows verified

AI PRODUCES

06-test-cases.md · Full test case table
feature.cy.ts · Runnable Cypress e2e tests
06-test-report.md · APPROVED / BLOCKED verdict

🤖 Claude

🧪 Cypress

🅰️ Angular TestBed

🛡️ Snyk

📈 Datadog

📄 06-test-cases.md

🧪 feature.cy.ts

📄 Test report: APPROVED

🚦

Gate 5 — QA Sign-off (Deploy Trigger)

Test report: APPROVED. All HIGH-priority tests pass. Zero regressions. Security checklist cleared. Performance within threshold. PO completes final UAT on staging. This sign-off starts the deploy.

GATE 5 of 5

Phase 06 · Release & Deployment

Staging first.
Then production.

AI generates the deploy runbook, GitHub Actions CI workflow, release notes, and customer-facing Intercom announcement. Staging deploy runs smoke tests before production is touched. Monitoring is live from minute one.

OWNED BY

🚀 DevOps

+

🔧 Tech Lead

06

STEP 08 DevOps + Tech Lead · AI assisted

Run Stage 07 → Release & Deployment

Two-stage deploy. Staging first — migration runs, smoke tests run, health check passes — then production. Customer announcement goes out automatically on success.

🧪 Staging Deploy

→ Run DB migration
→ Deploy to staging
→ Run smoke tests
→ Health check ping ✓

🚀 Production Deploy

→ Run DB migration
→ Deploy to production
→ Monitor 15 minutes
→ Slack #releases notified

Claude generates deploy.yml GitHub Actions workflow from deployment checklist
Intercom AI rewrites technical release notes into customer-friendly in-app message
Datadog dashboard and alert rules go live immediately on deploy
PO notified — feature is live, customers see the announcement

AI PRODUCES

07-release-notes.md · Technical + customer versions
deploy.yml · Full GitHub Actions CI workflow
Datadog dashboard JSON · PagerDuty alert config

🤖 Claude

⚙️ GitHub Actions

💬 Intercom AI

📡 Datadog

🔔 PagerDuty

📣 Release notes

⚙️ deploy.yml

📡 Monitoring live

🎉 Feature live

🎉 Feature is live in production

PO notified. Customers see the announcement. Monitoring active for 24h. All .md files committed to features/[date]-[slug]/ — full decision history preserved.

6 phases complete
5 gates cleared
3–4 days total

⏱ Typical Timeline

Simple feature end-to-end (like the About Section V1). Each bar is proportional to actual time spent.

Phase 1 — Intake

PO · Steps 1–2

~2h

2 hours

Phase 2 — Assessment

Tech Lead · Steps 3–4

~1 day

1 day

Phase 3 — Design

TL + Eng · Step 5

~1 day

1 day

Phase 4 — Build

Engineer · Step 6

1–2 days (AI cuts ~50%)

1–2 days

Phase 5 — QA

QA + Eng · Step 7

~4 hours

4 hours

Phase 6 — Release

DevOps + TL · Step 8

~2 hours

2 hours

Total — simple feature, end to end

vs. 1–2 weeks without the AI workflow

3–4 days

~50% faster with AI

🏛 Non-Negotiable Rules

These six rules hold the entire workflow together. Breaking any one of them creates unpredictable outcomes.

📋

PO owns Stage 01 — always

The feature definition is a product decision. AI drafts it, PO signs off on it. Engineers don't start until Stage 01 is approved by the PO and Gate 1 is cleared.

PHASE 1 GATE

🔧

Tech Lead owns Gates 2, 3 and 4

No HIGH risks unmitigated (Gate 2). No build without approved design doc (Gate 3). No merge without code review (Gate 4). All three are mandatory checkpoints.

PHASES 2–4 GATES

🤖

AI generates. Humans own.

Every AI output is a first draft. The person who runs the prompt is accountable for the result. Never ship AI output without reading and reviewing every line.

ALL STAGES

🧪

No deploy without APPROVED test report

Stage 06 must produce a test report with APPROVED status before Stage 07 begins. QA sign-off is the deploy trigger, not a formality that follows the decision to deploy.

PHASE 5 GATE

📁

All .md files committed to the repo

Every stage output lives in features/[date]-[slug]/. The feature folder is the single source of truth. No decisions lost in Slack threads.

ALL STAGES

♻️

V2s carry V1 files forward

When a new requirement extends an existing feature, keep all prior stage outputs. Only re-run stages that are impacted by the delta. Never start from scratch on an extension.

DELTA RULE