Changelog

Start from what you already have, then design how to validate the market

PurplePrint coaching is now at v2.68.0. It starts from your real materials, checks that your environment can preserve the work, and closes the validation decisions that must exist before launch.

• If you have a plan, Notion page, or repository, the coach reads it first and only asks for material it cannot open
• If your chat surface cannot create files, it warns you about losing artifacts and shows the supported path before coaching begins
• Phase 7 now defines use, willingness-to-pay, and retention hypotheses plus one experiment with explicit success and failure criteria

The current coaching version is v2.68.0. Existing projects do not need a restart; revisit only the Phase that needs reinforcement.

Sign in with Google or GitHub — it's the same account either way

Until now each sign-in method created its own account, so coming back through the other one hid your work. If the email matches, they're now one account.

• Your projects, reviews and growth records show up whichever way you sign in
• Billing attaches to that same account — your sign-in method no longer splits it

For safety, only verified email addresses are linked. An unverified address never connects to an existing account.

Finished projects can be archived — the slot frees up, the record stays

There was no way to tidy up projects, so someone ended up reusing a finished project's slot for a completely different topic. Instead of deleting, you can now archive.

• Say "archive [name]" and it leaves the list, freeing one project slot
• Nothing is deleted — "restore it" brings it back exactly as it was
• "Rename it" works too, so a project that changed direction doesn't need a new slot

Your output files live in your own folder and are never touched by archiving.

Your next single step now has its own place on the growth page

After opening a place to see the curve comes a place to read it. Pick a measurement and what you actually did — and what to do next — appears beside it.

• Pick a measurement on the left and that record opens on the right — clicking a point on the curve does the same
• "Your next step" is no longer buried inside the assessment sentence
• Trend, history and detail each have their own pane, so you always know what you are looking at

Older measurements have no next step recorded. We leave it blank rather than invent one — measure again and it will be there.

Your growth curve is now on the web

Until now a growth measurement showed once inside the session and vanished. Watching the curve build is the whole point, and there was nowhere to watch it. Sign in and your records are gathered in one place.

• Lead-stage trend as a curve — from the second measurement you can see what changed
• The seven behavioral signals from your latest measurement, at a glance
• Which projects you worked on between each measurement

Your conversations never reach the server — only the tallies. Project attribution covers work done through PurplePrint.

Session hand-off notes always land in the same place + growth can be measured in depth

When a long conversation had to be cut short, the coach invented a different filename each time for the hand-off note — so the next session couldn't find it. The name and location are now fixed. Growth measurement can also go much deeper on request.

• Hand-off notes live at `docs/review/handoff-DATE-keyword.md` — the next session restores from that one file
• The growth table now shows which project each entry came from, so the curve lines up with what you remember
• Ask "look at my growth in detail" and it reads far more sessions and cites the evidence behind each call

The detailed pass costs more time and tokens. The default stays light — ask for depth only when you want it.

Parallel projects no longer bleed into each other + finished designs hand off to build

With several windows open on different projects, a save could land in the wrong one. And people who had finished their design still only got "want to continue?". Both are fixed.

• Each window records only to its own project — the other one no longer gets dragged along
• A project whose design is complete gets build hand-off guidance (pass the handoff and task docs to your dev agent) instead of a resume prompt
• Growth records tell you when it's worth measuring again — the curve appears from the second measurement

With two or more projects, the coach always tells you which project it wrote to.

Design isn't decided by feel — measure references, then use the numbers

In the design step (Phase 5) the coach could skip actually measuring reference sites and go by feel. The result works but looks flat. Now the step doesn't close without measured numbers.

• Open 2-3 reference sites, measure type scale, section spacing and key element sizes, and set the design from those numbers
• A helper agent (Codex) can do the measuring — you just take the numbers
• Adjectives like "modern, clean" alone count as incomplete. The review step (Phase 6) checks against the same bar
• Three more "made by AI" patterns get filtered — color removed without filling the gap, every section on the same rhythm, the signature card shaped like every other card

Projects already past the design step don't need a redo. Just fill in the numbers in your design system.

A helper agent takes the research — your coaching thread stays clean

When the coach researches competitors, personas and markets itself, tens of KB of raw material piles up in the thread, and the design decisions that follow get dragged along by it. Now a separate helper agent does the research and the coach only reads back the result.

• With a Codex subscription — the coach walks you through install and login, then runs the helper for you. No need to open the Codex app
• Without one (or not sure) — the coach does the research itself. We don't push you to subscribe
• Model and research depth are preset — you don't need to know what to pick
• You don't wait around — it runs in the background, and the coach picks up the result when it's done

This applies to competitor, marketing, persona and GTM research. Journey simulation and code audit stay with the coach, since those need the design context in hand.

Tighter screen design — a structure you don't get lost in, and guidance for first-timers

Three defaults landed in the screen-design coaching (Phase 4~5). Until now the coach decided these case by case.

• Structure first, screens second — decide how screens group, what they're called, and how many taps deep they sit, then draw
• Walk your journeys back over that structure — wherever “what do I tap next?” stalls, that's the gap
• First-timer guidance by default — empty-state nudges, a first-run tour, inline hints, progress indicators
• Icon buttons carry an explanation — no more buttons nobody dares to press
• Blocks the “made by AI” look — purple gradient heroes, endless rounded cards, and friends now need a stated reason

Already mid-design? That's fine. If you're shaping screen flow, reinforce the structure once; if you're settling the visuals, just add the icon and guidance specs. The review stage picked up the same checks.

Hyper-Review on the web — no install, connect GitHub or Notion

You can now get the 39-axis review straight from the web, without installing anything in your editor. Sign in and it's free, twice a week.

• Paste — drop in your plan or design as-is
• Attach a file — design PDFs and wireframe images are read too
• Connect GitHub — pick a repo and it reviews the code and docs as they are
• Connect Notion — no need to move your planning pages anywhere

Connecting is one button, and only the repo or page you pick is read, read-only. Disconnect any time. The report states what was read and what was left out — because no findings on a file that was never read doesn't mean it's fine.

Hyper-Review series — from service design to code alignment

Ahead of the July 23 web launch, we finalized the three-layer review contract from service design → dev design → code:

• Hyper-Review — checks your service design across 39 axes, recursing until zero findings
• DAR (Deep Architecture Review) — reviews your dev-design architecture across 8 rotating lenses
• Code Audit — checks whether the code matches the design

Hyper-Review and DAR run asynchronously on the server. Code Audit keeps raw code local and runs through your agent. Standard is the base pass; Ultra recurses more deeply.

4 ops commands — recover on your own when stuck

server-health (self-diagnosis)·manual (detailed guide)·oauth (account·trial status)·handoff (artifact package). The agent diagnoses “connected but not working” and tells you the next step.

Hyper-Scenario — user-journey simulation

Replays your design as the persona, step by step, so you see where users stall and drop off before you build. Pulls User Flow·Story Map·test seeds too. Runs on your AI agent's web — zero cost.

Hyper-Research family — competitor·marketing·persona·GTM

Fills in what you can't know from your head with real research — competitor strategy·marketing channels·persona validation·go-to-market. In the Phase flow, the coach proposes it at the right moment. Runs on your AI agent's web — zero cost, content stays local.

Connection stability — slash visibility + 30-day login

/purpleprint:* slash shows reliably in Claude Code·Cursor. One login lasts 30 days, so no re-auth every time.

2-week free trial + beta billing

Sign up and a 2-week free trial starts with no card. After that, beta $13/mo (list $20).

Builder-capability measurement — my-growth

Measures how much you've grown as a builder while shaping your design. Content-free — it never reads what you wrote, only behavioral signals (like the decisions you made on your own). It shows 'Zero to Builder' as measured evidence, not a promise.

PurplePrint MCP beta — first release

Phase 0~9 conversational design coaching inside your coding tool. From idea → a build-ready design.