LLL · communication board

Verification and Constraints

● RUNNING · 2 / 24 criteria proved

Apply the revised sales prep skill to ablakbau.hu, testing and improving one asset at a time.

Live owner: /root/ablakbau_executor · Sol xhigh Run: ablakbau-sales-prep-trial-2026-09-28-v1 Updated: 2026-09-28T03:43:20.238468+00:00

Goal and finish line

End goal: Prove that sales-prep-kit can produce a clear, sourced, commentable sales-call kit for ablakbau.hu, and fix its templates or instructions whenever an asset fails review.

Done means: Each selected asset passes its required source and QA checks, has a live review URL and actual screenshot proof, and any reusable fix is in the canonical skill. An unresolved failure keeps the run incomplete. The review index links every asset. The single Figma file is editable and its handoff link opens for Matt.

2 / 24criteria verified

12deliverables

0open decisions

Rules that stay fixed

Preserve unrelated files, client facts and earlier reviews. Edit only the canonical sales-prep-kit and Luna skill, plus this run’s artifacts.
No security layer on hosted deliverables. Publish bare URLs reachable from any machine, with comment layers on each HTML and the index.
Stay within the shared USD 15 task spend limit unless Matt separately authorizes more. Do not send client messages or make commercial commitments.

Proof standard: Local source packets use provider receipts, paths/hashes and coverage. Rendered HTML uses source readback, the actual published URL and desktop/phone screenshots. For all client-facing webpages, including boards, catalog, index and A/B, follow canonical Luna QA: at least two independent Luna xhigh workers, Luna max fixes, complete section/tall-segment inventory including headers, footers, meaningful open states and overlapping 3–4-section runs, then fresh full-coverage postdeployment QA and actual model/effort receipts. Independent descriptions must not be coached with expected results. Unavailable evidence stays unverified.

Deliverable flow

Inputs → outputs · maximum eight nodes

Read left to right. Each node names why it exists. Add an extra-next line only when it describes a later handoff that is not represented by the next visible node.

Input 1 · confirmed or open

ablakbau.hu

Why: Prospect site and its current public evidence are the only company-specific starting point.

Input 2 · constraint or source

Revised sales-prep-kit

Why: Canonical templates and the stricter Luna QA contract define the repeatable trial.

Deliverable 1 · main review item

Research + board assets

Why: Measured keywords, visibility, site health, competitor and ad evidence support the sales conversation.

Deliverable 2 · evidence or artifact

Two homepages

Why: A and the replaced B template test premium design and working controls.

Output · decision or next action

Index + accessible Figma

Why: One review entrypoint and one editable design handoff make the kit usable.

Next: Review this contract. A later explicit “Launch” starts the executor.

Parent ↔ child messages

Newest message first

PARENT · PREPARE2026-09-28

The skill changes and trial contract are ready for Matt’s review. No Ablakbau execution has started.

Parent read setSmall files first

Required executor inputs: canonical /Users/agency/.agents/skills/sales-prep-kit/SKILL.md and its source-workflow, board-templates, gemini, design-contract and delivery references, paid-ads/SKILL.md and TEST-PLAN.md; canonical /Users/agency/.agents/skills/luna-visual-qa-section-screehoshots/SKILL.md. Resolve dependency paths through these skills. Keep the dated run inventory, prompts, source receipts and session ID here after Launch.

Work asset by asset. Establish the required inventory from the canonical source workflow first. Produce each asset, inspect its live output, repair a failed run artifact and the reusable rule/template responsible, then rerender and attach fresh proof. Preserve original generated responses. Back up canonical changes and run TEST-PLAN.md checks. A missing external input blocks only dependent acceptance, while independent cards may proceed. Never waive a failed criterion or launch from this Prepare board.

Work and proof

One card per deliverable

Collect measured source evidence

A dated, reusable packet establishes every client claim.

COMPLETE2 / 2

Show criteria, constraints, and evidence

Observable tests

When ablakbau.hu is inspected with current scraping, keyword and provider checks, then a dated source packet with measurements, screenshots and unknowns appears in the run folder.
When the packet is reviewed against the visible site and provider readbacks, then a source coverage receipt and any corrected extraction rule appear beside it.

Protect

Do not invent search volumes, reviews, active ads or rankings.
Do not overwrite valid prior source files; record dates and provenance.

Verified 2026-09-28: 36 preserved raw pages plus six dated browser supplements, 393 exact keyword inputs measured separately for Hungary and Pest County, 12 current priority SERPs, both real PageSpeed views, two actual reviews, 32 individually inspected originals and 15 + 15 library ads. Source coverage receipt has 14 byte-verified source paths and explicit unknowns. The reusable extraction fix preserves raw bytes and provider nulls. 70 canonical regression tests pass. Private evidence remains in the run folder.

Local proof: source-coverage-receipt.json, source-enriched.json, keyword-work/measurement-receipt.json, seo-research/, paid-ads/, data/pagespeed.json and data/gbp.json. Source-only criteria are complete. Rendered assets remain subject to full Luna QA.

Review catalog and preserved assets

Every unchanged source asset remains useful and inspectable.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When the source workflow is applied to Ablakbau images and baseline evidence, then a hosted image catalog plus every unaffected baseline/annotation/ad-presence/diagram asset appears in the final artifact inventory.
When each inventoried asset is opened at its actual review URL or inspected as a raster, then a separate receipt row proves correct content, working images and required comment layer; attach screenshots and Luna descriptions for GUI criteria.

Protect

Do not silently omit unchanged sales-prep outputs.
Keep credentials, private source packets and logs out of public pages.

Independent Luna xhigh opened all 757 pre-fix research/catalog screenshots at measured desktop and phone sizes. The 32 generated previews load and all 36 catalog records are reachable. Original-site capture artifacts, two library paths, four expanded phone panels and the catalog review control are in the required repair/fresh-QA loop. Live acceptance and the separate Luna description remain pending.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Verify site health and presence

One readable asset joins performance, technical and discovery evidence.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When mobile and desktop PageSpeed, technical, GBP, analytics and ad checks are supplied, then one dated HTML asset shows the screenshots, review count/themes, profile quick wins and verified tag/ad status.
When its live URL is opened on desktop and phone, then all evidence panels render, captions identify sources, and the review layer is usable; attach both screenshots and independent Luna description.

Protect

Treat absent or unavailable checks as unknown, not failure.
Keep screenshots with the correct date, viewport and source; do not call an ad active without library evidence.

Evidence pending: source/provider readback; published URL; desktop and phone screenshots; independent Luna description and parent comparison for GUI criteria; skill/template diff and re-test if failed.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Publish measured keyword opportunities

A Google-style table lets the prospect compare real searches quickly.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When the exact measured keyword set is rendered, then national and verified-region monthly volume, average CPC and low/high top-of-page bids appear in the keyword HTML.
When supported opportunities are reviewed in the live table, then their rows are light yellow with every cell bold, only those rows show the exact first three organic rankers, and no explanatory paragraphs remain; attach desktop/phone screenshots and Luna description.

Protect

Use provider null as missing, never zero or an estimate.
Add unusual categories only when offer fit, volume and SERP evidence support them.
Verify the prospect county from sources and provider-supported location code. Record geography, language, date, currency and any dated HUF conversion. Low/high top-of-page bids are not minimum/maximum CPC. Keep null unavailable fields explicit.
Use separate national and supported county location-code tasks over the same keyword set. Retain location/language response codes and raw search_volume, cpc, low_top_of_page_bid and high_top_of_page_bid fields. Never use omitted location or country-only coordinates as regional proof. Disclose grouped similar-keyword volume rather than treating it as unique demand; split calls only when needed within budget.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Map measurable service pages

Each proposed page ties to a search and concrete content.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When keyword measurements and proposed pages are combined, then each sitemap keyword displays monthly volume and average CPC beside the page it supports.
When the live sitemap is opened at desktop and phone widths, then page cards remain readable and the review layer works; attach screenshots and Luna description.

Protect

Keep keyword values identical to the keyword board.
Do not add pages without a clear client offer and page purpose.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Explain current search visibility

One short finding shows what is known now.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When dated ranking evidence is checked for the domain, then a concise current-visibility asset states the observed result and what it means.
When its live URL is reviewed, then the finding is understandable without the redundant search-by-search panel; attach screenshot and Luna description.

Protect

Do not turn sampled searches into complete ranking claims.
Keep source dates and uncertainty visible.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Explain practical competitor lessons

Readers see why each comparison matters.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When three priority searches and ranked landing pages are inspected, then both competitor views explain their purpose: ranked page elements and proposed actions in A, and up to three evidenced example/gap/improvement lessons in B. Record why any unsupported panel is omitted.
When the live competitor views are opened, then a first-time reader can state the takeaway without oral context; attach screenshots and Luna description.

Protect

Keep rank order tied to exact query, place, device and date.
Do not equate an observed page pattern with proven conversion performance.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Sort sourced ad examples

Ad examples stay concise and verifiable.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When at least 15 eligible ads per channel are collected from Google and Meta ad libraries and rendered, then the combined board shows actual creative captures, copy, dates, destination and inference labels, sorted longest first by observed date span within each channel.
When the published board is checked on desktop and phone, then source links and creative screenshots work and its comment layer remains usable; attach screenshots and Luna description.

Protect

Do not call first-to-last span uninterrupted delivery.
Do not count repeated creatives or invent missing ad text.
Eligibility includes distinct creative/library IDs and a verified observed interval of at least 35 days. Preserve Meta original-library captures with IDs. A partial board is explicitly incomplete and cannot pass the 15-per-channel criterion.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Refine homepage A visually

The premium concept must survive interaction and crop checks.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When the A template is populated from verified Ablakbau sources, then the live page has the stronger white hero overlay, fully visible portrait heads, Rólunk alone without repeated eyebrow/greeting, service icons about 48 px, and two gate halves using the identical full image/crop and exact offsets so the closed seam is uninterrupted.
When gate, expanding forms and chat are tested at desktop and phone widths, then controls work without clipping, open chat has three contrasting outline strokes and a top-right X, close/reopen works and visible demo status is absent; attach every section and overlapping 3–4-section screenshots plus fresh Luna descriptions.

Protect

Do not replace client facts with donor-template copy.
Keep all working form and gallery controls; label sample reviews.
Strong accents belong to CTAs, icons and logo. Large surfaces must be neutral, dark or desaturated. Use meaningful sourced Hungarian copy. Preserve the explicit labeled sample-testimonial design in design-contract.md.
Use the canonical Gemini A/B generation process with three source attachments and unchanged fixed prompt. Resolve current provider models, inspect dry-run requests and bound aggregate spend before generation. Preserve originals and model/request/cost receipts. Apply run fixes in final-html and reusable fixes only within the owned canonical package.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Test replacement homepage B

The user-supplied template becomes a distinct usable option.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When the pre-generation SHA256 check confirms that /Users/agency/Desktop/website_template_v1.html and canonical sales-prep-kit/resources/template-b.html match, and that template is populated with verified Ablakbau facts, then a complete B page appears at its live review URL with the template structure, forms and interactions preserved. Final client HTML need not be byte-identical to the donor.
When each section and overlapping 3–4-section runs are visually inspected on desktop and phone, then the page reads as a coherent, elegant design and defects are repaired and rechecked; attach post-deploy screenshots and Luna description.

Protect

Keep the replacement template as B’s canonical structural source.
Follow the required labeled before/after process: inspect a real client reference, generate the AI before image from those actual bytes, preserve viewpoint/geometry, and retain model/source/hash/cost receipts. Show the AI-before/real-after disclosure and verify slider alignment. Never substitute unrelated photographs.
Strong accents belong to CTAs, icons and logo. Large surfaces must be neutral, dark or desaturated. Use meaningful sourced Hungarian copy. Preserve the explicit labeled sample-testimonial design in design-contract.md.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Publish the final review index

One entrypoint makes every accepted asset reviewable.

PENDING0 / 2

Show criteria, constraints, and evidence

Observable tests

When all selected HTML is published with stable comment documents, then the index links every verified live asset, the working parameterized ROI calculator and the collapsed CTR explainer below its footer.
When final HTML and companion assets are copied to the actual Desktop lead folder, then matching file hashes, its index and a folder-open receipt appear in the delivery record; attach live-index screenshots and Luna description.

Protect

No login or review token may be added to public HTML.
Do not hand off local or staging URLs as verified live reviews.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Verify the editable Figma handoff

The actual file must be usable by Matt.

RUNNING0 / 2

Show criteria, constraints, and evidence

Observable tests

When every final HTML and required board raster is imported into the saved single Design file, then complete nonempty frames, editable Hungarian text and loaded images appear in Figma readback.
When the exact final file URL opens in Matt’s accessible account, then the intended file and frames load and edit permission is verified. Also open that same URL in a fresh unauthenticated browser to verify bare-link viewing. Record sharing settings and both access receipts. If provider policy prevents required access, leave only Figma incomplete with the exact limitation.

Protect

Reuse one saved file key during resume.
Do not mark Figma done from a create/import acknowledgement alone; keep access failures explicit.

One saved Design file contains editable Hungarian research frames, all 30 sourced ads across six native parts, all 36 catalog records across six native parts and 14 loaded legacy raster panels. The coherent source images now replace the existing original-site native fills. The initial owner edit and fresh anonymous view were verified. Annotation, both final homepages, the final index, final frame readback and final access verification remain pending.

Screenshot proof pending

Desktop, phone and required interaction states. Add the actual captured image, independent description and criterion comparison here after the test.

Open decisions and blockers

Only unresolved items

None recorded.

Recovery state

Only inspect on stall or completion

0consecutive stalls

activerecovery state

Source evidence → build → live verificationresume point

Current work: repairing the independently proved capture, library-image and expanded-state defects, then fresh live desktop/phone QA and final native Figma assembly.

Parent action: passive wait for the executor terminal report. Launch is authorized and execution is active.