The voice agent explains any task, answers questions grounded in this plan, and records a brain dump. Voice feedback is read-only and cannot approve or execute work. Allow the microphone when your browser asks.
The explicit Sonnet rollback keeps the existing capsule-only protocol. A lost connection retries the same Sonnet engine twice and reloads the complete persisted conversation. It never changes models during a conversation.
Sonnet 5 · no fallbackVoice answers are review evidence. Approval, sending, spending, and production changes stay in the originating chat. Voice trouble in this page? Open the standalone voice page.
Plan: /Users/agency/Documents/Agty/lovasember-project/action-plans/2026-08-10-lovasember-persona-workbook-build.md
J/K, to open it.1 delegate, 2 drop, 3 idea, 4 ask first, 0 clear, J/K to move between tasks (and open the one you land on).! to make it a hard requirement it cannot ignore.Produce the design inputs the workbook build consumes: the lovasember.hu brand tokens for colours, type, spacing and imagery mood, a concise $100M Offers doctrine card set from the Hormozi knowledge base, and a comparison of persona-mapping frameworks from which exactly three are chosen for the three template layouts.
Without this, the build guesses at brand and framework choices, producing a workbook that neither looks like the client's site nor rests on a defensible offer methodology.
A generic-looking tool the client does not recognize as premium work, and template structures that cannot be traced to any framework when asked why the questions are what they are.
Every template question and visual choice is grounded: brand-matched, Hormozi-aligned, framework-named. The three layouts are genuinely different lenses, not three skins.
A research inputs file exists with the brand token table, 10 to 15 doctrine cards each carrying a query-manifest reference, and the three chosen frameworks with a one-paragraph rationale each. No raw knowledge-base passages. 80/20. Brand colours plus the value equation plus one persona-mapping framework comparison table.
flowchart LR
A[Build guesses at brand and method] --> B[Research pack: tokens, doctrine, frameworks] --> C[Grounded design inputs for every template]
flowchart TD
S1[Extract brand tokens from lovasember.hu] --> S2[Query Hormozi knowledge base for offer doctrine]
S2 --> S3[Write concise doctrine cards with query manifest]
S3 --> S4[Compare persona-mapping frameworks and pick three]
S4 --> D{Inputs file has all three sections}
D -- yes --> DONE[Done]
D -- no --> S2
Build the Hungarian single-page interactive workbook in framework layout 1: six guided persona slots prefilled from the market analysis clusters and the persona hypotheses, an add-persona control, each slot walking problem, dream outcome, solution levels, offer sketch, with an editable ranking view that orders personas but never declares a winner. Submit POSTs the answers back, an export button downloads the same JSON as fallback.
Sándor and Dóri agreed to rank a top-3 persona set before any campaign build, but have no structured, low-effort way to do that thinking. The workbook is that instrument.
The persona decision keeps living in calls and message threads, the autumn season approaches with no chosen audience, and the funnel proposal has nothing verified to stand on.
One link the client can complete in a sitting, producing structured persona-offer data that flows straight into the prompt pack and the funnel proposal.
The workbook page renders all six prefilled slots with evidence tags, add-persona works, POST and export both produce the full answer JSON, all copy is Hungarian with no em dashes and no semicolons, and no commercial figures anywhere in the page. 80/20. Six slots plus a working export button; POST wiring can follow.
flowchart LR
A[Persona thinking scattered across calls] --> B[Guided workbook with 6 prefilled slots] --> C[Structured persona-offer answers returned]
flowchart TD
S1[Ground in research pack and market analysis clusters] --> S2[Build 6 prefilled slots plus add-persona]
S2 --> S3[Each slot: problem, dream outcome, solution levels, offer sketch]
S3 --> S4[Ranking view, POST submit, JSON export, autosave]
S4 --> D{Both return paths emit identical complete JSON}
D -- yes --> DONE[Done]
D -- no --> S2
Using the two remaining frameworks chosen in T-001, build two visually distinct template kinds and instantiate each for the same two personas, producing four extra templates. Every template is at least 1.5 A4 pages of visual structure: cards, arrows, tables, varied section shapes. The extras are selectable, self-contained sections that can be deleted as units.
Which framing works best for this client cannot be judged from a single layout. Three side-by-side framings of the same two personas make the choice concrete.
The workbook ships in whatever the first layout happened to be, with no evidence a different framework would have pulled better answers from the client.
Ten total templates to choose from, with the two comparison personas rendered in all three frameworks so the pick is an informed one.
The two alternate template kinds exist for the same two personas as four sections, each visually distinct from layout 1 and from each other, each at least 1.5 A4 pages equivalent when printed, reachable from the workbook navigation, deletable as units. 80/20. One alternate kind for one persona proves the pattern.
flowchart LR
A[One layout, no comparison] --> B[Two alternate framework template kinds] --> C[Ten templates, informed pick of the best]
flowchart TD
S1[Take frameworks 2 and 3 from the research pack] --> S2[Pick two of the six personas]
S2 --> S3[Build both alternate kinds for both personas]
S3 --> S4[Wire into navigation and the shared answer JSON]
S4 --> D{Print preview confirms length and JSON round-trip complete}
D -- yes --> DONE[Done]
D -- no --> S3
Publish the workbook to a staging URL, then verify it live: browser QA at 1280, 768 and 390, every slot opened, add-persona exercised, a full test fill submitted and confirmed to land with status code and readback, export downloaded and diffed against the POST payload, test data cleaned up. Iterate until clean. Nothing is forwarded to the client, the one-shot reveal stays Matt's call.
An unQA'd interactive page can fail silently in front of the client and spoil the first impression this deliverable is meant to make.
A broken input on a phone means the workbook gets abandoned mid-fill and the persona data never arrives.
A verified live link Matt can forward the moment he decides to, with proof the answers actually come back.
Staging URL live, QA findings table clean at all three viewports, one full submit verified server-side and deleted, export and POST payloads identical. 80/20. Mobile 390 pass plus one verified submit.
flowchart LR
A[Workbook exists only locally] --> B[Staging deploy plus live QA loop] --> C[Verified link ready to forward]
flowchart TD
S1[Deploy workbook to staging, record URL] --> S2[QA at 1280, 768 and 390: every template, test fill]
S2 --> S3[Verify POST lands, diff export against payload]
S3 --> S4[Fix findings, redeploy, re-test, delete test data]
S4 --> D{Findings table clean at all three widths}
D -- yes --> DONE[Done]
D -- no --> S4
Create the Activity Log entry on the client's CRM page linking the staging URL, the plan and the review page. Create one task in Matt's main tasks DB with deadline Friday 2026-08-14 10:00 containing links to every artifact from this run, the paste-ready post-fill prompts, and the send-to-client checklist.
The run's outputs and the exact next-step prompts otherwise live only in a chat session that expires from memory.
When the client returns the filled workbook, the next steps have to be reconstructed from scratch, and the Friday send deadline has no anchor anywhere.
One Notion task holds the deadline, every link, and the copy-paste prompts, and the client timeline shows the work.
Both Notion pages exist and are verified by readback, the task deadline is Friday 2026-08-14 10:00, and the prompts render as copyable code blocks. 80/20. The tasks-DB task with deadline and links.
flowchart LR
A[Outputs live only in a chat session] --> B[Two Notion records: activity log and deadline task] --> C[Deadline, links and prompts anchored in one place]
flowchart TD
S1[Log activity entry on the client CRM page] --> S2[Create task in main tasks DB due Friday 10:00]
S2 --> S3[Body: artifact links, prompt pack, send checklist]
S3 --> D{Readback shows correct deadline and links}
D -- yes --> DONE[Done]
D -- no --> S2
Write the exact prompts Matt pastes into any coding agent the moment the client returns the filled workbook: first ingest and normalize the returned JSON, second score and rank the personas and produce the top-3 recommendation memo, third generate the funnel and service-delivery proposal draft from the winner. Each prompt is self-contained with absolute paths and restates the house copy rules.
The gap between client replied and proposal drafted is where momentum dies. Ready prompts close it to minutes.
Days of lag while context gets reassembled, exactly when the client is warmest.
Client replies, Matt pastes three prompts, a grounded proposal draft exists the same day.
A prompt-pack file with the three prompts, each runnable standalone, referencing real existing paths, restating the Hungarian copy rules and the never-auto-choose guardrail. 80/20. Prompts one and two, ingest plus rank.
flowchart LR
A[Client replies, context must be rebuilt] --> B[Three paste-ready prompts] --> C[Grounded proposal draft the same day]
flowchart TD
S1[Prompt A: ingest and normalize returned JSON] --> S2[Prompt B: score, rank, top-3 memo that recommends only]
S2 --> S3[Prompt C: draft funnel and service-delivery proposal]
S3 --> D{Every referenced path verified to exist}
D -- yes --> DONE[Done]
D -- no --> S1
Write the short internal roadmap that sequences everything after the client returns the workbook: rank, top-3 review call, pilot spec, visual offer pack, then the full proposal with funnel and service-delivery system, including what may be claimed at each step under the evidence rules and where the commercial conversation fits.
The workbook is a means. Without the roadmap, its output does not connect to the proposal Matt actually wants to send.
The persona data arrives and sits, the autumn window narrows, and the older plan's follow-on tasks float unanchored.
A one-page internal sequence from filled workbook to a compelling offer, with the evidence and commercial gates marked.
A one-page roadmap file sequencing the steps with owners, gates, and links to the old plan's task IDs. 80/20. The ordered step list with gates.
flowchart LR
A[Workbook output with no next-step map] --> B[One-page gated roadmap] --> C[Clear path from answers to signed offer]
flowchart TD
S1[Sequence: workbook returned, rank memo, review call] --> S2[Then pilot spec, visual offer pack, full proposal]
S2 --> S3[Mark evidence gates and the commercial gate]
S3 --> S4[Link the old plan task IDs so nothing double-dispatches]
S4 --> D{One page, sequenced and gated}
D -- yes --> DONE[Done]
D -- no --> S1