Use Alchemistic logo as favicon; fix and finish icon-generation pipeline

[aquarion]

Summary

The Alchemistic logo (yellow rounded square, black diagonal-stripe glyph) is now used for the site's favicon and Apple touch icon, generated from resources/branding/alchemistic.svg via a Vite-time icon-generation pipeline. The Istic logo in the sidebar (resources/views/components/app-logo-icon.blade.php) is untouched.

The icon-generation pipeline itself (bin/icons/*) was added in a prior commit on this branch but had several bugs preventing it from actually working; this PR fixes those and wires it up:

• Added the missing vitest devDependency and npm test script (test files imported from vitest but it was never installed, and there was no way to run them).
• Fixed generateFlatIcon (used by apps without a bloom.icon Apple Icon Composer bundle, like this one) — it was outputting a blank solid-color square instead of the brand glyph.
• Fixed glyph scaling for that fallback: it now renders at the source SVG's native 1200 viewBox and lets sharp downscale, instead of embedding the same path coordinates inside a mismatched smaller viewBox (which cropped and off-centered the glyph).
• Reconciled generate-web-icons.js's output filenames with its own test (logo-standard.*/logo-on-white.png) and updated docs/branding.md to match, plus corrected a couple of other doc inaccuracies (glyph color, regeneration instructions referencing a script that doesn't exist).
• Rewrote generate-apple-touch-icon.test.js, which previously mutated a real resources/branding/bloom.icon/icon.json file that doesn't exist in this repo (so it always failed); it now covers the flat-fallback path this app actually uses.
• Added vitest.config.js so npm test doesn't pick up vite.config.js's plugins and silently regenerate real icon files as a side effect of running tests.
• Deleted orphaned duplicate icon files (application-*, bloom-*) left over from earlier renames — some had been committed with the dev-only debug background color (#CC0000) instead of the real brand color.
• Gitignored resources/icons/* since it's fully regenerated per APP_ENV by the Vite plugin on every dev/build run — keeping it out of git is what prevents stale/wrong-environment-colored files from being committed again.
• Wired resources/views/partials/head.blade.php to the Vite-managed icons via Vite::asset(), and removed the now-superseded static files in public/.

Test plan

• npm test — all 17 tests pass
• npm run build — icon-generation plugin runs successfully (build itself fails later on an unrelated missing vendor/livewire/flux/dist/flux.css, because Composer dependencies aren't installed in this sandbox — pre-existing environment limitation, unrelated to this change)
• Visually inspected generated favicon.svg, favicon-96x96.png, apple-touch-icon.png, and logo-on-white.png — glyph renders centered and correctly proportioned
• Verify in a real Sail environment that the favicon/apple-touch-icon show up correctly in a browser tab and on iOS home screen

🤖 Generated with Claude Code

https://claude.ai/code/session_01StCfH1mTviG3UqQoL1bUW5

---

Generated by Claude Code

[aquarion]

PR Review Summary (multi-agent: code, tests, silent-failures, comments)

Critical Issues (1)

• bin/icons/generate-apple-touch-icon.js:65-78, 191-199 — glyphLayer() silently returns null if a referenced glyph asset is missing/unreadable, or its SVG doesn't match the naive <path d="..."> regex. The caller (generateFromIconFile) just skips it — no throw, no warning. A bad icon.json reference could produce a broken icon (background + effects, no logo) that ships with a clean exit code, regenerated on every build via vite-plugin.js. This path has zero test coverage today (no bloom.icon bundle in-repo yet). Recommendation: throw instead of returning null in both branches — a failed build beats a silently glyph-less icon in production.

Important Issues (3)

• docker/cloudflared/entrypoint.sh:31 — health check curls http://application:${APP_PORT}, but APP_PORT is the host-side forward port; inside the container network the app only listens on 80 (as config.yml:6 correctly hardcodes). Works today only because .env happens to set APP_PORT=80. If a dev changes APP_PORT to avoid a host port clash, the tunnel container will falsely report "application did not become ready" after exhausting all 60 retries. Fix: hardcode 80 for the internal health check like config.yml does.
• bin/icons/generate-apple-touch-icon.js — the entire Icon Composer rendering path (generateFromIconFile, backgroundLayer, glyphLayer, edgeGlowLayer, cornerSpecularLayer) has zero test coverage, since no bloom.icon bundle exists in this repo yet. Worth a minimal test fixture to exercise this code.
• bin/icons/generate-apple-touch-icon.js:74-78 — comment says translucency=0.5 yields "~0.70 opacity," but the formula 0.4 + translucency * 0.55 actually gives 0.675 — a real ~3.7% drift, invisible because of the untested-code gap above.

Suggestions

• bin/icons/vite-plugin.js has no test file — the APP_ENV ?? config.mode precedence and syncJson: !isServe behavior are untested.
• resources/branding/alchemistic.svg:3 — the new <g transform="translate(600 600) scale(1.0588235294) translate(-600 -600)"> wrapper (added to reduce icon padding) has no comment explaining the magic scale factor (18/17, targeting ~5% margin on the tall axis).
• New vitest suite (bin/icons/*.test.js) isn't run in CI — .github/workflows/tests.yml only runs Pest.
• docker/development/entrypoint.sh:39-49 unconditionally runs config:cache/route:cache/view:cache on every dev-container start — a dev footgun if .env edits are expected to take effect without a rebuild.
• Minor: docker/development/entrypoint.sh:8 id -u www-data failure isn't guarded under set -e; .env sourcing is fragile against unquoted special characters (pre-existing pattern); supervisord.pid is untracked and probably belongs in .gitignore.

Strengths

• Icon-pipeline JS tests are genuinely behavioral (real sharp output, real ICO byte layout, real pixel colors), not mocked.
• Error-handling elsewhere in the icon pipeline consistently favors throwing over silent defaults — generate-apple-touch-icon.js's glyphLayer is the one outlier.
• ngrok→Cloudflare migration is clean: no orphaned scripts, .gitignore updated coherently, no secrets committed.
• Prior review feedback on this branch (usermod error handling, retry-timeout handling, entrypoint logging gaps) has all been addressed.
• Blade head correctly uses config('app.name', …) rather than env().

🤖 Generated with multi-agent review (code-reviewer, pr-test-analyzer, silent-failure-hunter, comment-analyzer)

[aquarion]

Thanks — pushed fixes for the parts of this in scope for this PR (f70921e):

• Critical: glyphLayer() silent null — fixed. Both branches (missing asset, unmatched <path d="..."> regex) now throw instead of silently dropping the layer, so a bad icon.json reference fails the build loudly rather than shipping a logo-less Apple touch icon. Also dropped the now-dead if (composite) guard in the caller.
• Translucency comment drift (0.675 vs claimed ~0.70) — fixed.
• Vitest suite not run in CI — fixed, added npm test to ci.yml's test job.

Left for later, not urgent:

• Icon Composer path (generateFromIconFile and friends) has no test coverage — agreed, but this repo has no bloom.icon bundle at all yet, so there's nothing real to build a fixture against without inventing one. Worth doing once an app actually uses that path; will track separately rather than block this PR on it.
• vite-plugin.js untested — noted, same "not urgent" bucket.

Not acting on, out of scope for this PR:

• docker/cloudflared/entrypoint.sh and docker/development/entrypoint.sh items — neither file is touched by this PR's diff (they came in via the merge from Fix dev container HTTP server and replace ngrok with Cloudflare named tunnels #40, already merged to main). Happy to open a follow-up if these still stand, but won't fix files this PR doesn't own.
• resources/branding/alchemistic.svg:3 — the review describes a <g transform="translate(600 600) scale(1.0588235294)..."> wrapper that doesn't exist in the current file (it's just two <path> elements, no <g>). Looks like a mismatch on the reviewer's end; nothing to change here.

---

Generated by Claude Code