Most PMs use AI the same way: paste some notes, ask for a PRD, get generic output, manually fix it for an hour. The problem isn’t the AI — it’s that you’re feeding it a blob of unstructured context and hoping for magic.

What if your AI assistant had the same structured methodology you use? What if it pushed back when you jumped to solutions, enforced requirement syntax, and carried forward every decision you made?

That’s what ContextForge’s Project Kickstart workflow does. It’s an 8-step PM pipeline that takes you from raw idea to sprint-ready user stories, with a built-in kill gate to prevent wasted effort.

The pipeline at a glance

Step	Method	What you produce
L0: Brief	CIRCLES framework	Problem statement + appetite
L1: IRD	EARS syntax	Testable requirements
GATE	Decision framework	GO / PIVOT / KILL
L2: Personas	Proto-personas	2-3 behavioral archetypes
L3: Scenarios	Journey mapping	User flows with pain scores
L3.5: Mocks	Fat-marker sketches	Key screen layouts
L4: Conceptual Model	Domain modeling	Glossary + relationships
L5: Stories	Story mapping	Prioritized vertical slices

Each step has its own ContextForge session with pre-loaded templates and reference cards. A PM persona system prompt shapes every AI interaction — it enforces structured thinking and pushes back on feature-first reasoning.

How the zones work

Every session in the pipeline has three zones:

Permanent contains the PM persona — a system prompt that enforces CIRCLES, EARS, and user-oriented thinking. It stays the same across all 8 steps. This is what prevents the AI from giving you generic MBA-speak.

Stable swaps with each step. At L0 you get the CIRCLES reference card and a brief template. At L1 you get EARS syntax patterns and an IRD template. Each step gets exactly the methodology it needs.

Working is where you create. It starts empty and accumulates your outputs as you advance through the pipeline. By L5, your Working zone contains the full chain: brief, requirements, gate decision, personas, scenarios, and conceptual model — everything needed to write stories.

Walking through a step: L0 Brief

You open the first step. Your zones are ready:

Paste raw materials into Working — meeting notes, client emails, product ideas, whatever you have
Open Brainstorm and say: “Walk me through CIRCLES for this project”
The PM persona guides you through each letter of CIRCLES — Comprehend, Identify, Report, Cut, List, Evaluate, Summarize
It will push back if you jump to solutions too early. That’s intentional.
Set the appetite — a time budget, not an estimate. “This is worth 6 weeks” is a decision. “This will take 6 weeks” is a guess.
Save the completed brief to Working

Before advancing, check:

Is the problem stated from the user’s perspective (not as a feature request)?
Is the appetite a decision, not a guess?
Do you have at least one customer segment specific enough to find and interview?

The kill gate

Step 3 is the hardest. After writing your brief and requirements, you run the GATE decision: GO, PIVOT, or KILL.

The AI evaluates evidence, risks, and decision criteria. It’s honest. If your evidence is weak, it says so. If your requirements are vague, it calls it out.

This is where most PM workflows fail — they don’t have a formal decision point. Without a gate, you invest weeks in personas and stories for a project that should have been killed at the brief stage.

Carry-forward: context that compounds

The key mechanic is carry-forward. When you advance from one step to the next:

Permanent carries — the PM persona stays active
Working carries — your outputs accumulate
Stable swaps — each step loads fresh templates

By L5 (Stories), the AI has your full context: the problem brief, testable requirements, gate decision, personas, and journey maps. When it writes user stories, every story traces back through the chain. No story exists in a vacuum.

Requirements that are actually testable

The L1 step uses EARS syntax — a pattern system for writing requirements that can be turned directly into test cases:

WHEN [trigger], THE SYSTEM SHALL [response] — event-driven
WHILE [state], THE SYSTEM SHALL [behavior] — state-driven
IF [condition], THEN THE SYSTEM SHALL [response] — conditional
THE SYSTEM SHALL [behavior] — ubiquitous (always true)

Every requirement follows one of these patterns. If you can’t write a test case from the sentence alone, the requirement isn’t specific enough.

Getting started

Import project-kickstart-workflow.zip into ContextForge using Import Skill
Name your project and start the workflow
Walk through each step — the templates and reference cards guide you
Use Brainstorm at each step with your full context loaded
Advance when your quality checklist passes

The entire pipeline exports as a Claude Code skill, so your team can run the same methodology outside of ContextForge too.

Why structure beats raw prompting

Without this pipeline, a PM asking AI for a PRD gets a template filled with platitudes. With it, you get output grounded in your specific problem, your specific users, and your specific constraints — because every piece of context was built methodically through 8 steps of structured thinking.

The AI doesn’t replace your judgment. It amplifies it — by holding the full context while you focus on the current decision.