[yia-mw-agent]

Change

A newcomer can now read one guide in about ten minutes and see what saga does, and the setup guide is a short common path with two tracks.

Before. docs/build-tic-tac-toe-with-react.md was a draft opening that ended at the first click. docs/getting-started.md mixed what every app needs with the Lit composition, evidence and actions material, so a reader met package structure and evidence boundaries before an app.

After.

• docs/build-tic-tac-toe-with-react.md is the whole guide: the layout and the four operations, what saga handles, one guided change (a move list), connecting to Claude, ChatGPT and Alexa+, and a closing "what else this design allows" section. The closing section labels each arrangement runnable, a straightforward composition, or an architectural sketch, and adds two diagrams (diagrams/tic-tac-toe-arrangements, diagrams/tic-tac-toe-a2a-delivery). The approved opening is kept.
• The three host sections each say this game is unverified there and link docs/host-compatibility.md. They describe a voice or conversation move as a different interaction from clicking the board. Nothing in them reports a host run.
• docs/getting-started.md keeps what every app needs: prerequisites, published and candidate installs, saga init with the offered pairs, HTTP defaults, attaching to an existing backend, troubleshooting, and the choice of track.
• docs/build-notes-with-lit.md is the Lit track. The composition, content policy, identity and evidence, actions and "what is implemented" sections moved here byte for byte (checked against the previous text), with a short lead-in and a closing hand-off to docs/actions.md.
• The README's first steps point to both tracks, and links into the moved sections are updated (docs/public-install.md, docs/development-reference.md, AGENTS.md). The README sentence saying the starters "currently use Lit" now says five of the six use Lit and one uses React on Hono.

The "Component, definition, template and instance" section stays in the setup guide. It is renderer-neutral vocabulary, and the interactive tour and the action guide link to it.

Design. Getting-started tracks, section "Setup guide and two tracks", approved at revision 6c3638b3f372c27d12b406a0c4800ea01f93cf24 (approval record, #324). The delivery coordinator records the landed revision in the Blueprint Status after merge.

The ten-minute claim is reading time (about 2,700 words of prose plus code excerpts), not install or host setup time. The guide says so, and it states that the game needs the source packages (0.8.0-rc.1) because appendMessage and operations without a view are not in the published 0.7.7-rc.1.

Validation

Run on head 9b58e6e (Node 26.7.0, Linux x64), from a clean worktree of current main:

• Route from a clean checkout. npm ci, npm run build:tic-tac-toe, then starting apps/lit-host/dist/tic-tac-toe/server.mjs and the guided build's server.mjs: both printed their MCP endpoint line. This found that building apps/lit-host clears apps/lit-host/dist/tic-tac-toe, so the guide's browser-test group builds the host app first.
• Guided change in Chromium on the local host fixture. npm run build -w apps/lit-host, npm run build:tic-tac-toe, then npx playwright test e2e/tic-tac-toe.spec.ts -g "adds a move list": 1 passed. This is local qualification: the test clicks a square and plays O by calling the server, and no assistant took part.
• Document checks. npm run test:docs: 5 files, 171 tests passed. This covers scripts/guide.test.ts (setup guide, Lit track and generated recipes, which are current), the extended scripts/tic-tac-toe-document.test.ts (commands, quoted source in order, the patch line for line, quoted sentences, the operations table against the server, links, anchors, images and diagram sources, host claims against the host compatibility record, and the arrangement labels) and scripts/guide-actions-document.test.ts.
• tsc -p tsconfig.json clean; npm run lint on the changed test files reports nothing; Prettier applied to the three changed TypeScript files; npm run hooks:check passes.
• Version policy. python3 scripts/mw-version.py check --base origin/main --title '<this title>' passed with no bump: documentation, diagrams and tests only.
• LFS. python3 scripts/check-lfs.py reports 25 valid pointers; the two new PNGs are LFS pointers and git lfs push ran before the branch push.
• Full headless suite (npm run test:headless): 2,842 of 2,843 tests passed. The one failure, a 5-second timeout in scripts/verify-safeguards.test.ts under 140 parallel workers, passes alone (23 of 23). apps/lynx-fixture/src/adapter/admission.test.ts could not load because the retained Lynx fixture's own dependencies were not installed in this checkout. Neither touches these changes.
• The diagrams were rendered with @mermaid-js/mermaid-cli 12.0.0 from the committed .mmd sources and checked by eye for legibility.

Limits. Local source and Chromium evidence only. Claude, ChatGPT and Alexa+ have not been run against this game, and the guide says so. The guide's statements about each host rest on the existing host compatibility record, which records no host accepting a message a card appends. The wait's cadence and limit remain proposals for host runs to settle. Source readiness is not publication, installation or host acceptance. For review: please check the technical claims in the closing "what else this design allows" section.

[timeleft--]

Writing and technical review completed at this exact head. The six moved Lit sections are byte-for-byte unchanged. Independently verified 171 documentation checks, eight packed example checks with canonical TMPDIR=/private/tmp, root TypeScript checking, tracked-clean normal/guided packed builds, current-base version policy, and 25 valid LFS pointers. The diagrams distinguish runnable examples from composition and architectural sketches, and the host instructions keep the game unverified in each remote host. One bounded accuracy correction is required below; no runtime or architectural expansion is requested.

[timeleft--]

Independently re-reviewed exact head 15c02bb. The repair changes only the operations bullet and accurately distinguishes a bounded wait and cancellation signal from stopping an uncooperative handler or undoing effects. The original finding is resolved. All 171 documentation checks and current-main version policy pass at this head; current main 5015042 is an ancestor and the review checkout is clean. Reused unchanged evidence from the full writing/technical review: six byte-identical moved Lit sections, root TypeScript, eight packed-example checks with TMPDIR=/private/tmp, tracked-clean normal/guided builds, the passing guided move-list Chromium walkthrough, and 25 valid LFS pointers. Both new diagrams were inspected at full size and in a rendered article; runnable/composition/sketch distinctions and remote-host limits remain clear. No further findings. Current-head application CI is still running and must pass before merge.