Improve documentation SEO and agent retrieval

[ayushflow]

Description

Improves site-wide SEO, accessibility, freshness governance, and AI-agent consumption for the FlutterFlow documentation.

• Normalizes metadata and accessibility across 376 source documents without treating metadata-only changes as new freshness evidence.
• Generates canonical llms.txt, llms-full.txt, bounded topic packs, page-level Markdown, an agent manifest, clean sitemap metadata, and TechArticle structured data.
• Adds copy and agent-consumption controls while loading the third-party chatbot only after interaction.
• Adds deterministic SEO and retrieval audits, CI enforcement, and scheduled freshness checks.
• Corrects and verifies high-value task and troubleshooting guidance for accurate agent answers.
• Replaces 16 oversized referenced GIFs with optimized MP4s and removes duplicate or unused media, reducing repository payload by approximately 116 MB.

Type of change

• Typo fix
• New feature
• Enhancement to current docs
• Removed outdated references
• Update assets

Validation

• npm run typecheck
• npm run build
• npm run audit:seo-agent:built
• npm run eval:agent-retrieval
• git diff --check origin/main...HEAD

Build output covers 379 canonical pages and 12 bounded topic packs. All eight static audit categories score A+ at 100%. The fixed retrieval evaluation covers 50 intents and 300 query variants with 100% top-1 accuracy, recall@5, and evidence support.

Public repository safety

The complete branch delta, final tree, and branch history were scanned before publication. No private implementation paths, private code excerpts, credentials, new secret material, symlinks, submodules, hidden payloads, or local audit reports are included.

[johnbhiggins7]

The new fallback Markdown exports have broken links and omit some useful content. Could we fix this before merging?

I reproduced this with a production build:

• In build/marketplace.md, [add and purchase](adding-purchasing-item.md) retains the source-file path. From the published /marketplace.md URL, that resolves to /adding-purchasing-item.md, which does not exist in the build. The creator-guide links have the same problem.
• build/troubleshooting.md retains source-relative links and image paths that do not resolve to the published files.
• build/index.md loses the navigation-card content, including the first-app/quickstart link, because sourceToAgentMarkdown() removes entire <InfoCards> blocks.

The affected conversion is in scripts/postbuild-seo-agent.mjs around lines 163–169. These exports are also copied into the topic packs and llms-full.txt, so the problems propagate into the AI-readable corpus.

Could we use the rendered-page conversion pipeline for these pages, or preserve the card content and resolve document/image links to their published URLs? A regression check covering these landing pages would help. The current build and SEO audit pass despite these issues.

[pinkeshmars]

@ayushflow, a few additional issues to address before merge:

• In the Google Login guide, the instruction to add a Firebase __/auth/handler redirect URI for each preview origin is misleading. The handler URL should use the Firebase authDomain configured for the app, unless a preview domain has explicitly been configured to serve the auth handler.

• All 16 GIFs have MP4 replacements, but their page Markdown exports show the videos as empty links such as . The MP4 files work; the issue is that the exports lose the videos’ labels and use root-relative URLs in standalone text.

[pinkeshmars]

@ayushflow could you please look into the review comments?