Improving The LLM Wiki Setup Wizard And Public Guidance - Source Excerpt 03 - Prioritized recommendations
Back to Improving The LLM Wiki Setup Wizard And Public Guidance
Summary
This source excerpt begins near Prioritized recommendations and preserves the surrounding evidence from Wiki.FFTAC.org/agent-file-handoff/Archive/2026-05-11-improvement-concept-layer/Improving the LLM Wiki Setup Wizard and Public Guidance.md.
**Source path:** Wiki.FFTAC.org/agent-file-handoff/Archive/2026-05-11-improvement-concept-layer/Improving the LLM Wiki Setup Wizard and Public Guidance.md
| Benchmark | What it does well | Useful pattern for LLMWikis | Published onboarding proxy | Official evidence |
|---|---|---|---|---|
| Docusaurus | Ordered sidebar trees, autogenerated and hand-curated sidebars, next/previous paging, official search docs, official i18n support. | Make information architecture explicit and let metadata control labels/order instead of flattening everything into one long route list. | No completion rate published. Proxy: clear doc-tree model plus DocSearch support. | citeturn27search0turn27search24turn27search32turn28search4turn31search0turn31search8 |
| GitBook | Site sections, global search across sections, automatic and draft redirects, analytics by traffic/pages/search, built-in translation tooling. | Treat restructuring as a first-class operation; let redirects, sections, analytics, and localization live in the publishing model itself. | No completion rate published. Proxy: analytics segmented by section/device/search and draft-vs-live redirects. | citeturn33search1turn33search2turn33search3turn33search6turn33search9turn33search10turn31search1turn33search20 |
| Material for MkDocs | Emphasizes clear navigation, tabs/sections, and instant loading as a flagship feature. | Use fewer full-page reloads, stronger local nav, and lighter-feeling transitions in the public handbook. | No completion rate published. Proxy: instant loading as a documented UX feature. | citeturn27search2turn27search18turn28search13 |
| MediaWiki | Web config script plus CLI alternative, explicit install/config documentation, technical manual hierarchy. | Keep advanced install modes available, but clearly separate “first-run browser path” from “advanced operator path.” | Published time estimate: installation typically takes around 10–30 minutes. | citeturn34search0turn34search8turn34search11turn34search12turn34search7 |
| Drupal | Interactive installer, language selection, installation guidance, quick-start local demo path, maintained curated guides. | Use presets/profiles for common site types instead of asking everyone to assemble a path from scratch. | Proxy: quick-start command for local demo; official interactive installer path. | citeturn27search4turn34search1turn34search10turn34search13turn34search19 |
| WordPress | Built-in permalink settings and built-in Site Health with critical/performance issue reporting. | Surface URL policy and diagnostic health checks inside the setup flow, not as separate expert knowledge pages. | No funnel metric published. Proxy: built-in health screen and permanent-link model. | citeturn27search5turn31search15turn32search0turn32search1 |
| Read the Docs | Versions, translations, redirects, canonical URLs, sitemap/robots/404/`llms.txt` support in one documentation platform. | Treat versions, language, redirects, canonical rules, and discovery files as foundational platform concerns. | No completion rate published. Proxy: product-level support for versions/translations/redirects. | citeturn27search6turn27search14turn28search7turn28search23turn31search2turn31search6 |
| GitLab Wiki | Wiki as separate Git repo, custom sidebar, web/local editing, wiki-specific linking. | Keep the durable markdown model, but pair it with a stronger sidebar and browse structure. | No completion rate published. Proxy: sidebar customization and Git-backed wiki model. | citeturn27search7turn27search15turn27search31 |
A second, narrower benchmark from the LLM Wiki ecosystem reinforces the same lesson. The `lucasastorian/llmwiki` repo emphasizes a “one-command start” (`llmwiki open`) that initializes, serves, opens the browser, and prints the MCP configuration hint, while the repo itself had 861 stars and 135 forks at audit time. The `MehmetGoekce/llm-wiki` repo advertises “5-minute setup” via a single `./setup.sh` that creates schema, namespaces, and git tracking, and the repository had 86 stars at audit time. In contrast, the `obsidian-llm-wiki-local` project had a public bug where ingest could not resume from a failed chunk, forcing a full rerun. The adoption proxy and the pain signal point in the same direction: shorten time-to-first-success, then harden recovery. citeturn20view0turn20view1turn25view2
The benchmark takeaway is straightforward. The best documentation and setup systems do **three things extremely well**: they give users a short path to first value, they keep advanced configuration available but out of the beginner’s way, and they treat redirects/versions/search/localization/analytics as part of the product—not as afterthoughts. citeturn27search0turn27search1turn27search2turn33search2turn31search2turn31search6
## Prioritized recommendations
The table below compares the current experience with the recommended target state.
| Capability | Current state | Recommended state | Why it matters |
|---|---|---|---|
| Wizard entry point | One page carries explanation, form, packet, and AI digest. | Split into a short landing step plus a guided wizard; keep the AI digest in a collapsible/secondary panel. | Reduces cognitive load and preserves dual-use behavior. |
| Wizard steps | Seven visible steps with advanced architecture/policy inputs early. | Five core steps for most users; advanced branches only when selected. | Improves completion and reduces abandonment. |
| Presets | Setup path choices exist, but still expand rapidly into expert detail. | Presets such as **Personal wiki**, **Team handbook**, **Docs migration**, **Evidence archive**. | Gives newcomers a safe default. |
| Validation | Local draft save and packet generation are visible; public validation/error copy is not. | Required-field validation, path/URL examples, preflight checks, inline recovery text, and blocked export states. | Prevents bad packets and boosts trust. |
| Navigation | Repeated global nav + menu + source-map rail; shallow breadcrumbs. | One primary nav, one local nav, contextual “related pages,” structured breadcrumbs. | Reduces clutter and improves findability. |
| Metadata | Current public standard covers ownership/review/sensitivity well. | Add `seo_title`, `meta_description`, `canonical_url`, `page_type`, `primary_category`, `tags`, `locale`, `schema_type`, `noindex`. | Makes governance and SEO work together. |
| Discovery files | `llms.txt` exists; sitemap signaling is inconsistent. | One canonical sitemap strategy, validated discovery files, and explicit redirect/canonical coverage. | Prevents crawl ambiguity and broken discovery. |
| Internationalization | Explicitly unsupported today. | Locale-ready URL/content model now; translation and language switcher later. | Avoids rework when multilingual support arrives. |
| Analytics | No public instrumentation path is described for the wizard. | Funnel events, search analytics, zero-result tracking, packet export, and error telemetry. | Lets the team improve based on evidence. |
| Accessibility | Skip link exists; dense option blocks and unclear public validation text. | One-question pages, proper fieldsets/legends, inline error summaries, reflow-safe layouts, keyboard QA. | Improves completion for everyone, not only disabled users. |
The evidence base for these recommendations draws from the current LLMWikis pages, comparable documentation systems, and official accessibility/SEO guidance. citeturn4view1turn4view3turn6view0turn27search0turn27search1turn27search2turn31search2turn29search0turn29search1turn29search15turn30search1turn30search3turn30search6
The **single most important UX change** is to refactor the wizard around user intent rather than implementation policy. The first question should not be “scope and audience” with 20 downstream architecture fields visible. It should be “What are you trying to set up?” followed by a concise explanation of what each path produces. The quickest way to do that is to turn the existing setup-path choices into presets with opinionated defaults, and ask the advanced questions only when the preset requires them. That matches the current wizard’s own latent branch model—direct setup paths and optional Project Handoff already exist in the AI digest—but it is not yet expressed cleanly in the public form flow. citeturn5view0turn36view1turn29search1turn29search2