Major update. A new site-audit workflow, automatic site fetching, clickable deliverables in the browser, and an audit-to-redesign path.

• Mapping existing sites into wireframes, now more accurate and automatic. The agent reads your site layout straight from the browser instead of guessing from screenshots, so the wireframe reflects the real structure: hero, sections, columns, cards, navigation. It also works on heavily JavaScript-driven sites that used to lose most of their content.
• Web (Claude / ChatGPT / Gemini), no copy-paste. The agent fetches the site by URL on its own and builds a mobile wireframe (around 400px wide). It does not interrupt you to paste content. The full desktop version is available in the IDE editions (Claude Code, Codex, Antigravity).
• New workflow: Site audit. Drop in a URL. The agent compares your site against knowledge of 17 page types (B2B Industrial, SaaS, ecommerce, portfolio, NGO, data-driven content, and more) and produces a single HTML file with two deliverables: an architecture analysis plus annotations on specific elements (where a call to action is missing, whether the conversion path is consistent, whether the hero communicates the value proposition). No brief, no analytics, no surveys required.
• You decide what to audit. Homepage only (the recommended, fastest option), the full site (complete coverage), or your own list of subpages. The agent proposes a candidate list based on the sitemap and domain knowledge. You accept, edit, or skip.
• Audit findings feed the redesign. If you want to rebuild the site after the audit, your findings (accepted / rejected / under discussion) flow automatically into the next steps: information plan, page archetype selection, anatomy of each template. You do not have to retype anything.
• Deliverables in the browser with clickable UI. Open the audit or wireframe in your browser. Every finding is clickable (the detail opens in the left panel), has three buttons (accept / reject / mark for discussion), and at the bottom you get an export-to-Markdown button for the decision brief. Findings are anchored to specific elements (for example a hero, a button, a particular section).
• Free Audit Demo, a lightweight variant of the skill for portfolios and digital products. Fetches a URL, classifies the site, produces a short audit of the homepage. No project setup, no context. Ideal for demos or quick feedback.
• Better copy quality in Messaging Foundations. Target-audience naming (ICP) is picked from 6 methodologies: by role, industry, life situation, or trigger. Main headlines are drawn from 8 proven patterns. The "Why us" section comes in three approaches: comparison with competitors, complement to existing tools, or alternative to avoiding the problem. Pillars with evidence in 3 tiers: Hard (metrics, case studies) / Soft (demos, capability) / Hypothesis (logic, early signals).

New workflow: Messaging Foundations. Build a complete messaging proposal with mapped evidence.

• Standalone Messaging Foundations workflow. From competitors through pillars and evidence to finished headlines and an HTML canvas. 11 steps across 4 phases with 6 STOPs
• The orchestrator scores the quality of the messaging foundations (compact WPQS /25) at the Discovery to Strategy handoff and picks the path automatically: full workflow, refine, or skip
• Two paths in step 1: standard (starting from competitors) and backward (starting from capabilities, for technical founders)
• A hybrid 4-row HTML canvas, built incrementally section by section alongside the MD document
• Pillars with evidence in 3 tiers: Hard (metrics, case studies), Soft (demos, capability), Hypothesis (logic, early signals)
• Strategy and Wireframing automatically consume the messaging-foundations file when it exists: segment, category, pillars, headlines, narrative
• A shared CSS base (html-craft.md) extracted for wireframes and the messaging-foundations canvas

Existing-site mapping now runs as a full process, with screenshots as the visual reference.

• Full process in 5 phases: subpage inventory, screenshots, structure mapping, context and strategy, wireframes. Three checkpoints where you can approve progress
• Automatic full-page screenshots with cookie banners dismissed. If the screenshot tool is missing, the agent walks you through the install (globally / scoped to the agent / skip)
• Every page template is processed in parallel, not sequentially. You get results faster on larger sites
• Every subpage is described in 6 fields per section: type, purpose, user question, layout, elements, copy from the original site
• Context and UX strategy are filled in automatically from the site analysis, not from a conversation with the user
• Self-check: after the wireframe is built, the agent verifies that every section and element from the map made it into the final HTML

New workflow: map an existing site into lo-fi wireframes. Available only in the IDE versions (Claude Code, Codex, Antigravity).

• Standalone entry point. No prior brief or strategy required. A site URL or a file (HTML, sitemap, screenshots) is enough
• Automatic setup for the content-fetch tool. The agent checks availability, installs missing components, and runs them in the background. If your environment is bare, it walks you through the install step by step
• Page-template recognition with confidence levels: Confident / Uncertain / Skipped. The agent shows what it considered and why it excluded a candidate, and you approve the list before further work
• The wireframe reflects the current state without judging the UX. Same quality and format as a new-site design, with the original copy from your site

Orchestrator fixes from the second live test. Unified greeting and consolidated communication rules.

• Unified startup template. One consistent greeting that adapts to context (new user, returning user, multiple projects)
• Communication rules consolidated into a single section, which removes inconsistencies between parts of the skill

Instead of a numbered workflow menu, the skill asks about your goal and picks the right path for you.

• The skill asks about your goal and selects the right path. You describe what you want to achieve and the skill leads from there
• A dependency map between workflows. The skill knows that strategy needs context and a wireframe needs strategy, and routes you to the missing steps automatically
• Removed the /lang and /preferences commands. Everything is handled conversationally

Wireframing moves to HTML-first. Mockups are built as lo-fi HTML you view in the browser, not Markdown documents.

• HTML wireframe is the default and primary format. The mockup grows section by section in a single HTML file. You preview it in the browser and give feedback live
• Iterative build in batches: pages with 4+ sections are split into logical narrative groups, and you review each batch before the next one starts
• Six design lenses: User Question, Narrative Arc, Content Priority, Persuasion, Task, Awareness
• New knowledge base wireframe-craft.md: rules for building lo-fi HTML, CSS architecture, section patterns, anti-patterns
• Markdown wireframe is still available as an alternative deliverable, not the only format

Communication-quality fixes. Consistent terminology, cleaner Polish copy, better reasoning behind recommendations.

• Fixed the confidence-level labels for consistent terminology across sections
• Removed anglicisms from Polish-language recommendations
• Improved reasoning transparency with concrete examples inside the confidence blocks

From discovery to wireframe, with Polish specifics baked in (the 1.5% PIT mechanism, OPP platforms, GDPR requirements).

• Expanded donation patterns: form variants, recurring giving mechanics, fee coverage, thank-you page
• Storytelling and donor journey: emotional trigger model, donor path architecture, defenses against compassion fatigue
• Polish specifics: OPP status and 1.5% PIT, Polish donation platforms, trust signals, compliance checklists

Three big changes: a copy-quality filter, a new Data Content domain, and deliverables exported as Word and HTML.

• Built-in anti-AI filter. 32 rules that strip generic, machine-sounding language from every piece of copy the skill produces. Covers both English and Polish patterns
• New domain: data-driven content (reports, dashboards, infographics, data stories). 4 narrative archetypes, 7 reader personas, a 17-item chart catalog with selection rules
• Deliverables in Word and HTML at the key workflow checkpoints. The client gets presentation-ready files without manual formatting

The skill runs discovery better, signals recommendation confidence more clearly, and loads domain knowledge more precisely.

• Discovery now opens with an inputs guide. The skill tells you which inputs are critical and which are optional
• Recommendations are tagged with a confidence level (confirmed / probable / speculative)
• Workflow steps now state which domain knowledge they draw on

Two domains expanded: B2B Industrial (ordering and configuration) and Portfolio (branding and creative agencies).

• B2B Industrial: order workspace, approval patterns, punchout/eProcurement, parametric configurator
• Portfolio: Agency-Studio-Led archetype, anatomy of a branding case study (7 sections), creative process gallery
• CMO / Marketing Director persona: trust-building sequence, browsing behaviors

The B2B domain is split into a shared base plus three specializations: SaaS, Services, Industrial.

• Shared B2B base: routing for two personas (technical vs. business buyer), positioning strategy through specialization
• B2B Services: full discovery (11 assumptions), referral typology, 4 strategic archetypes
• B2B Industrial: full discovery (12 assumptions), CAD ecosystem integration, industry compliance

Big expansion. The skill is now a generalist covering 14 website types (previously B2B only).

• 14 page types: B2B, E-commerce, Marketplace, SaaS, Services, Community, Course, Documentation, Media, Government, Enterprise, Event, Nonprofit, Portfolio
• Wireframes include 5 UI states (default, empty, error, loading, success) and 4-layer handoff annotations
• Message hierarchy and brand voice brief in the strategy step; tone tags inside wireframes

Simpler working modes, cleaner checkpoints, startup logic moved to a dedicated file.

• Simpler mode system: the Teacher mode is gone. The skill now runs in a single mode with full checkpoints
• Standardized STOP checkpoint format (deliverable, decisions, position, next steps)
• Startup logic moved to a dedicated file, so the skill loads faster

First release of the UX Design Skill. Full journey from discovery to decision log, built on architecture spec v3.2.