◆ Vibe EngineeringSMIT · AI Agentic Engineering Urdu
Class 10 · 10.3

Planning V1: writing the Jira tickets

Plan the product the way a real team would — a backlog of clear tickets from PL-4 to PL-7, written in the stakeholder's voice, then handed one at a time to your most capable teammate: Claude Code.

🎯 Goal: a shippable backlog⏱ Read: 12 min🧭 Type: planning

The foundations are laid. Now we plan the actual product the way a real team would — by writing tickets. Imagine your engineering lead, product manager, business sponsor, even clients dropping these into Jira. Normally they would be shared out across a team; we are going to assign them all to one very capable teammate: Claude Code.

01From prototype to a real V1

Back in Jira, tidy up: the tiny “simple website” card (PL-1) has served its purpose, so delete it. The NDA prototype (PL-3) proved the concept. Now we break the journey to a full V1 into a handful of clear tickets, each a meaningful step a stakeholder would recognise.

02The four V1 tickets

Here is the plan — written deliberately in a mix of business and technical voices, because in real life they come from different people:

TicketTitleWhat it asks for
PL-4Build foundation of V1Upgrade the prototype into a proper technical foundation — frontend, backend and a temporary database with start/stop scripts — without adding new product features yet. (A technical-lead ticket.)
PL-5Add AI chat (still just NDA)Change how the product interacts: instead of a series of questions, a free-form chat with an AI that asks about the document, asks about the fields, and populates it from the answers.
PL-6All document typesExpand to support every document we have templates for. If a user wants an unsupported one, explain we cannot generate it — but offer the closest one we can.
PL-7Multiple users & final polishProper sign-up / sign-in; store and re-open previously generated documents; database may reset on restart; polish every screen to look like a professional SaaS; add a “draft, subject to legal review” disclaimer.
Write in the stakeholder's voice, on purpose

Notice PL-5 says nothing about Cerebras or structured outputs — that is deliberate. Tickets describe the what in human terms; the how comes from CLAUDE.md, the skill, and Claude's own judgement. Leaving room is a feature, not sloppiness.

03Start with a fake login

For PL-4 we are explicit about one thing: only a fake login screen for now, no real authentication — just bring the user into the platform. We do not want to bolt on auth up front; real sign-up/sign-in arrives in PL-7. Sequencing the hard parts for later keeps each ticket small and shippable.

04Think like a product owner

Do

✅ Put detail in the Description

Write the requirement in the ticket's Description field — that is what Claude reads. (It is surprisingly easy to type into the Comments box by mistake!)

Do

✅ Business-speak is fine

“Change the way the product interacts with a user” is a perfectly good ticket. You do not need to speak in code — that is the model's job.

With PL-4 through PL-7 on the board, our Kanban looks like a real product backlog. In a company these would fan out to different engineers. We are going to hand the lot to Claude Code, one at a time.

05Assign it to Claude — and check the skill loaded

Fire up claude and run /context. Two things to confirm before we build:

Your skill is recognisedcheck

You should see Cerebras inference listed as a project skill — proof the .claude/skills/cerebras folder is wired up correctly.

Plugins & MCP servers are presentcheck

FeatureDev, plus your Atlassian and GitHub MCP tools. Note they use a fair chunk of context — that is the MCP trade-off in action.

Re-authenticate Atlassianritual

You know the drill: /mcp → Atlassian → Re-authenticate → Approve → Accept. Skip it and the first Jira read will just hang.

✓ Key takeaways

  • Plan V1 as a handful of clear tickets (PL-4 to PL-7), each a step a stakeholder would recognise.
  • Write tickets in human/business language; leave the how to CLAUDE.md, the skill, and Claude's judgement.
  • Sequence the hard parts: a fake login first (PL-4), real auth later (PL-7).
  • Put requirements in the Description field, not Comments — that is what Claude reads.
  • Before building, confirm your project skill shows in /context, and re-authenticate Atlassian.