Images: 80 posts ship 1.6 MB PNG covers (mobile LCP ≈ 8 s on slow 4G), 120 marketing pages eager-load a 1.7 MB hero, and 209 posts use an SVG og:image that social networks cannot show

Found by the seat's site-wide audit of www.objectos.ai on 2026-10-06 (production build of main @ 53764e0). The measurements used Playwright Chromium at 390×844 with 4× CPU throttling and a slow-4G profile. Priority: P1. HTML, CSS and JS are already lean (no web fonts, about 2 KB of inline JS, 5–7 KB of gzipped CSS, CLS 0), so images are almost the whole cost.

Findings

1. Blog covers

80 posts use a PNG cover of 1.5–1.9 MB at 1672×941, and it is the LCP element. Another 50 use JPGs of 240–370 KB.
Card thumbnails use the full-size file too, so a full scroll of the blog index downloads about 18 MB in en and zh-Hans.
Measured LCP:
- /en/blog/ai-agent-workbench/ (PNG cover): 8.2 s;
- the same template with an SVG cover: 0.74 s.
Converted with sharp, a cover drops to about 52 KB as a 1200-wide WebP, or about 11 KB at card size.
Cause: src/layouts/BlogPost.astro:26-34,165-176 and src/components/ArticleCard.astro:26-33 use the raw cover.src.

2. Marketing hero image

120 marketing pages eager-load /images/objectos-home-hero.png (1,732,355 bytes, confirmed by the seat), which renders at only 415×340.
/en/ai/ has an LCP of 8.8 s at 1366 on slow 4G. As WebP it is about 32 KB.
Cause: src/components/MarketingEvidencePage.astro:40-48.
The same image is the hero on home and on all 15 marketing pages, and in dark mode it stays a glaring white block.

3. Social share images

On 209 posts (41 of them English), og:image is an SVG. X, LinkedIn, Facebook and Slack show no image card, which works against the AGENTS.md goal of distributing posts as threads.
BlogPosting.image is the same SVG (BlogPost.astro:34,108).
531 pages share the generic 1.69 MB og:image.

Done looks like

Covers: generated through astro:assets (<Image> / getImage) as WebP or AVIF with srcset and sizes, in the post header and in card thumbnails.
Hero: moved into src/assets and rendered through <Image> with sizes, lazy where it is below the fold. A dark-mode treatment (a dark variant or a muted overlay) is the dev's call, stated in the PR.
Social images: a 1200×630 raster (PNG or JPG) per post for og:image, twitter:image and BlogPosting.image, generated at build time from the cover, with og:image:width, og:image:height and og:image:alt.
Before/after table in the PR: bytes per template and LCP on the same profile, for home, pricing, platform, the blog index and 2 PNG-cover posts.
Browser check: the six viewports per AGENTS.md, covers looked at for quality loss.
Gates: green.
No content edits.

Surface

src/layouts/BlogPost.astro, src/components/{ArticleCard,MarketingEvidencePage,MarketingHome}.astro, src/assets/** (the moved hero), an OG-image build step, and astro.config.mjs if image service config is needed.

Order

After the indexing card.

Generated by Claude Code

Claim: PM loop, repo:www.objectos.ai seat
Session: session_01FeA1nwBz1ohH65dvffUGKr
Account: hotlong (the seat's linked user, as GET /user answers it; always the card's assignee)
Branch: claude/pm-dispatch-objectos-ju9td1, the session's one designated push branch for this repo. The card's order is "after the indexing card", so it is staged on #219's local branch staged/issue-219, which sits on stack/www-wave1. At push time only this card's own commits are replayed (rebase --onto).
Worktree: www-issue-220
Domain: n/a (single-lane repo)
Seat: repo:www.objectos.ai (objectstack-ai/objectstack#12224)
File surface: the card's own Surface section:

src/layouts/BlogPost.astro;
src/components/{ArticleCard,MarketingEvidencePage,MarketingHome}.astro, image markup only;
src/assets/**, for the moved hero;
an OG-image build step;
astro.config.mjs, image-service hunks only, kept separate from the markdown: key.

Make no content edits. Stay out of #223's layout CSS, and keep the hero's dark-mode treatment to the image itself.
Stop on breach and explain in the report.
Container & model: M, mode:subagent, model: default.
Clause-②: no
Thread-read: none
Serial constraints cleared: no. Pushes are serial on the one branch, and this card follows #219.

Generated by Claude Code

os-dev-report

{
  "issue": 220,
  "status": "staged",
  "branch": "staged/issue-220",
  "worktree": "/home/user/www-issue-220",
  "base_sha": "5d685cf995e2d09a7f3a699d1d06e6148b733877 (staged/issue-219 head, on stack/www-wave1)",
  "final_commit": "bec5251e3794e9ebe0e15ee7309ae23a0d70770d",
  "commits": [
    "76563bc perf(images): serve raster blog covers through astro:assets as WebP with srcset",
    "8c49132 feat(seo): render a 1200x630 JPEG share image per post from its cover",
    "2221e0e perf(images): move the hero into src/assets, serve it through Image, dim it in dark mode",
    "bec5251 perf(images): size card thumbnails for the width object-fit draws them at"
  ],
  "pr": null,
  "pr_body_draft": "/tmp/claude-0/-home-user/2224823d-3484-535e-94a5-c071ba6af395/scratchpad/issue-220/pr-body.md (first line: Fixes #220)",
  "session": "session_01FeA1nwBz1ohH65dvffUGKr",
  "premise_still_valid": true,
  "summary": "Raster covers (80 PNG + 50 JPG posts; 17 unique files) now go through astro:assets as WebP with widths and sizes in the post header (eager, fetchpriority high), ArticleCard thumbnails (feature card eager) and the home latest-posts grid; SVG covers keep a plain img on the original. The hero moved to src/assets and renders through the Image component (WebP 400-1586w): eager on the 120 marketing evidence pages, where it is in the first screen at all six viewports and the LCP element on a desktop, and lazy on home; in dark mode it is dimmed on the image itself (brightness 0.6, contrast 1.12). A build step (scripts/og-images.mjs, run in pnpm build after astro build) renders one 1200x630 JPEG per cover for og:image, twitter:image and BlogPosting.image, with width/height/alt. SVG text is drawn with bundled OFL faces and a cmap glyph gate fails the build on any character no face has. Premise verified: 80 PNG-cover posts, a 1,732,355 B hero, and 209 posts with an SVG og:image; the card's line numbers had moved after #219, but the code it describes was still there.",
  "files_changed": [
    "package.json (og:images script, build chain, 7 devDependencies)",
    "pnpm-lock.yaml",
    "scripts/og-images.mjs (new: OG build step)",
    "scripts/lib/og-image.mjs (new: shared URL rule)",
    "scripts/lib/og-fonts.mjs (new: bundled faces, fonts.conf, cmap glyph gate)",
    "public/images/objectos-home-hero.png → src/assets/objectos-home-hero.png (git mv)",
    "src/layouts/BlogPost.astro",
    "src/components/ArticleCard.astro",
    "src/components/MarketingHome.astro",
    "src/components/MarketingEvidencePage.astro",
    "src/layouts/Base.astro (DEFAULT_IMAGE block only; outside the card Surface, see deviations)"
  ],
  "before_after": {
    "profile": "Playwright Chromium; phone rows 390x844 at 3x, desktop rows 1366x625; 4x CPU throttle; DevTools Slow 4G (562.5 ms RTT, 1.44 Mbps down, 675 kbps up); cold cache; LCP = median of 3 runs. Bytes = encodedDataLength from a local python static server (no gzip, so HTML/CSS uncompressed); no-scroll = everything the load started incl. lazy images in Chrome's load-in distance, full scroll = after scrolling to the bottom. before = 5d685cf, after = bec5251, measured back to back under one verify lock on a shared 4-core box (load avg about 10).",
    "rows": [
      "home /en/ | LCP 1944 → 1916 ms (text both) | no-scroll 1,821,615 → 170,412 B | full scroll 1,835,942 → 184,739 B",
      "pricing /en/pricing/ | LCP 1560 → 1548 ms (text) | 46,504 → 46,534 B | same",
      "platform /en/platform/ | LCP 1796 → 1808 ms (text) | 1,805,784 → 129,837 B | same",
      "blog index /en/blog/ | LCP 2708 → 1808 ms (SVG feature cover both) | 142,787 → 148,368 B | 18,826,110 → 1,167,398 B",
      "post /en/blog/ai-agent-workbench/ (PNG cover) | LCP 9956 → 1700 ms (PNG → WebP) | 1,685,249 → 120,959 B | same",
      "post /en/blog/ai-expense-audit/ (PNG cover) | LCP 11120 → 1752 ms (PNG → WebP) | 1,903,685 → 135,214 B | same",
      "platform @1366x625 | LCP 2060 → 1928 ms (h1) | 1,805,784 → 92,303 B | same",
      "ai /en/ai/ @1366x625 | LCP 10620 → 1964 ms (hero PNG → hero WebP) | 1,804,910 → 91,429 B | same"
    ],
    "build_time": "pnpm build wall, cold Astro image cache both, shared box: 87.5 s → 108.4 s (astro image generation 0.6 s → 6.3 s; og-images.mjs 14.3 s for 87 images in 3 workers); warm-cache rebuild 90.0 s",
    "dist_size": "57,657,630 B / 1,082 files → 68,197,374 B / 1,321 files: +6.05 MB cover WebP variants (143), +5.46 MB share JPEGs (87; median 52 KB, max 152 KB), +0.4 MB hero variants, -1.73 MB hero PNG. 18.86 MB of original raster covers (17 files) stay in dist, referenced by no page (Astro keeps content-collection image originals)"
  },
  "og_raster_checks": [
    "87 share images, all 1200x630 JPEG (step 4 of the script plus an independent sharp metadata pass: 0 wrong), 5.46 MB total",
    "glyph gate: 70 SVG covers, 721 distinct drawn characters, all covered; its first run on this tree found U+21B6 and U+2298, so Noto Sans Math was added",
    "gate can fail: a synthetic SVG with Hangul and Thai reported those 6 code points and passed CJK, check mark and an entity-encoded emoji; on a copy of dist, og:image:width 1201 and an unmapped share URL made the step exit 1 with 3 issues",
    "looked at full size: zh-Hans/zh-Hant cover.svg rasters of ai-wrote-your-app-dare-to-merge, beyond-agentforce-copilot-open-runtime, natural-language-to-app-metadata, airtable-omni-vs-governed-ai-app-platform, self-hosted-ai-app-platform (1.6:1); ja/ko use cover-en.svg (AGENTS.md: translated locales reuse the English cover): why-ai-agent-pilots-fail-four-layers (circled digits, U+3000), when-ai-agent-deletes-production-database (no-entry sign, collision emoji, check mark), automation-governed-flow-changes (bottom caption kept). All 87 in six contact sheets. No tofu anywhere",
    "no SVG cover has Japanese or Korean text today; kana and Traditional Chinese are covered (checked), Hangul is not, so a future Korean-text SVG cover fails the build naming its code points",
    "emoji are monochrome (Noto Emoji): COLRv1 colour emoji decode but cairo draws nothing",
    "SVG covers are drawn whole at 1120x630 and widened by repeating edge columns (two covers have a caption in the bottom 23px a centre crop would cut; no cover draws in its outer 40px); PNG/JPG covers are centre-cropped",
    "in-process parallel SVG text rendering crashed (SIGSEGV or SVG rendering failed in 4 of 6 runs); three worker processes rendering one image at a time passed 5 of 5"
  ],
  "tests": "All at bec5251 (tree identical to the last build). pnpm check: 'Result (150 files): 0 errors, 0 warnings, 0 hints'. pnpm content:lint: 'content lint passed (339 files, 44 glossary terms checked)'. pnpm content:lint --published: same line. pnpm build (cold cache, under os-verify-lock, VERDICT command-exit 0): BUILD_EXIT=0, including 'svg-text-fit: OK ... (112 SVGs, 2244 text boxes, tolerance 2px, 1 allowlisted)', 'share images: 87 rendered at 1200x630 (5.2 MB, 14.3 s, 3 workers) for 879 pages; 125 bundled font files, no uncovered glyph', 'content lint --dist passed (478 built blog pages checked)', 'content lint --dist links passed (93199 internal links in 880 built pages resolve to built files)'. pnpm seo:smoke: 'SEO smoke test passed (870 HTML pages checked; 569 sitemap URLs against 301 noindexed pages; 5424 canonical/hreflang annotations; 15654 prose blocks language-checked; 93189 internal links redirect-checked)'. pnpm svg:lint: 'svg-text-fit: OK'. objectstack scripts/check-nul-bytes.mjs run from the www worktree: 'OK (scanned 621 text file(s) ... no raw ASCII control bytes)'. Negative proofs: glyph gate on a synthetic SVG (6 code points reported); og-images.mjs on a mutated dist copy (exit 1, 3 issues). WebP quality: 1280w cover vs the original resized to 1280, PSNR 39.06 dB, 1:1 crop shows no visible difference. No repo test suite exists in www; www has no scripts/pm/dispatch-gates.mjs, so the gate list was derived by hand from package.json and .github/workflows/cloudflare-pages.yml (check, content:lint --published, build, seo:smoke).",
  "screenshots": "/tmp/claude-0/-home-user/2224823d-3484-535e-94a5-c071ba6af395/scratchpad/issue-220/shots/after/: 36 states (six viewports x PNG-cover post /en/blog/ai-agent-workbench/, /en/blog/, /en/ x light, dark), each NAME-WxH-THEME-first.png and -full.png after scrolling every lazy image into view; report.json records 0 unloaded images and 0 px horizontal overflow in all 36. Close-ups: cover-closeup-ai-expense-audit-1920/390.png, blog-index-raster-card-*.png (1366 light, 1920 dark, 390 dark, 360 light), thumb-1366-before-vs-after.png, home-hero-media-*.png, ai-hero-panel-*.png. OG contact sheets: /tmp/claude-0/-home-user/2224823d-3484-535e-94a5-c071ba6af395/scratchpad/issue-220/og-check/sheet-0..5.jpg. Covers looked at for quality loss: none visible.",
  "deviations": [
    "Base.astro edited, outside the claimed Surface: moving the hero (card Done) removes /images/objectos-home-hero.png, which DEFAULT_IMAGE named as every non-post page's og:image. Replaced with getImage of the moved asset at 1200x630 JPEG (94 KB). Only the frontmatter DEFAULT_IMAGE block changed; it is in the same commit as the move (2221e0e), so the two replay together. staged/issue-218's Base.astro hunks are CSS tokens only and do not overlap",
    "ClusterPage.astro not touched although its reading-path cards are card thumbnails: it is outside the Surface and staged/issue-218 holds edits to it. Reported below as a finding",
    "New files outside the listed component files, all part of 'an OG-image build step': scripts/og-images.mjs, scripts/lib/og-image.mjs, scripts/lib/og-fonts.mjs, and the package.json build-chain line",
    "7 new devDependencies: wawoff2 (MIT; the FreeType in sharp's libvips reads neither WOFF2 nor WOFF, measured), @fontsource-variable/arimo, @fontsource/cousine, @fontsource-variable/noto-sans-sc, @fontsource/noto-sans-symbols-2, @fontsource/noto-sans-math, @fontsource-variable/noto-emoji (OFL-1.1; each needed by at least one character in today's covers). sharp was already a dependency",
    "astro.config.mjs untouched (no image-service config needed); no markdown key added or altered; no content edits (git diff 5d685cf..HEAD -- content astro.config.mjs is empty); no layout CSS changed: the only CSS added is the dark-theme filter on the two hero images",
    "Not pushed and no PR, per the staged-mode dispatch. That conflicts with agent-contract rule 1 (push the empty branch before editing); I followed the dispatch",
    "status 'staged' is not one of the contract's status values; used as the dispatch asked",
    "commit trailers are the model-free pair (Co-Authored-By: Claude, Claude-Session) per the agent contract and the sibling staged branches, not the harness reminder's model-named Co-Authored-By",
    "measurements and build times are shared-box readings (load avg about 10 on 4 cores); the before run used the same box and lock, back to back",
    "one background job was started (a 5-run stability loop the shell moved to background); I stopped it by its own PIDs. All static servers I started are stopped (ports 4801/4802 closed). Worktree kept as instructed"
  ],
  "mcp_calls": "0",
  "api_writes": "1 - the os-dev-report comment on objectstack-ai/www.objectos.ai#220, via scripts/pm/with-fleet.sh --via dispatch (one repository_dispatch POST to objectstack-ai/objectstack, which executes POST /repos/objectstack-ai/www.objectos.ai/issues/220/comments as objectstack-fleet[bot]). Reads only otherwise: GET issues/220 and its comments; npm registry reads for the new packages",
  "open_questions": [],
  "out_of_scope_findings": [
    "class: a · reach: public door - GET /en/ai-native-app-platform/ (built dist at bec5251) serves img src /_astro/cover.BauJbMSI.jpg, cover.ChJFVxqy.jpg, cover.Dn04X1um.jpg, cover.rFSi4jxB.jpg (original JPGs, 240-370 KB each, about 1.1 MB) in 180x190 reading-path card slots; 40 cluster pages use src/components/ClusterPage.astro coverSrc() on the raw cover · evidence: ClusterPage.astro lines 36-39 and 96-104 still use the original cover src; same defect as #220 finding 1 (card thumbnails) but the file is outside #220's Surface and staged/issue-218 edits it · suggested as a sub-issue of #220 · dedupe words: ClusterPage cover thumbnail, cluster page original JPG, reading-path article-card image, astro:assets ClusterPage",
    "carrier: 承接者:无 · noted, not filed - 17 original raster covers (18.86 MB) stay in dist/_astro, referenced by no page after this change; Astro keeps the originals of content-collection images (vite-plugin-assets marks query-string imports as referenced). Dead weight in the deploy, no wrong answer at a public door",
    "carrier: #218 (staged/issue-218 adds the dark-theme --proof-* overrides) · noted, not filed - on this base the marketing hero panel's HUMAN label (--proof-blue #3157d8) is dark blue on the dark panel in dark mode, seen in ai-hero-panel-390x844-dark.png",
    "carrier: #223 (layout CSS) · noted, not filed - the home hero's media-note overlay stays a near-white box in dark mode; with the hero now dimmed it stands out more. It is layout CSS, which this card stays out of"
  ]
}

os-dev-report-addendum

{
  "issue": 220,
  "status": "staged",
  "addendum_to": "os-dev-report https://github.com/objectstack-ai/www.objectos.ai/issues/220#issuecomment-6020532870",
  "branch": "staged/issue-220",
  "worktree": "/home/user/www-issue-220",
  "previous_tip": "bec5251e3794e9ebe0e15ee7309ae23a0d70770d",
  "final_commit": "95ada6245124b95e856bc5f01c7d0a2e7768c6e0",
  "new_commit": "95ada62 perf(images): serve cluster reading-path thumbnails through astro:assets",
  "pr": null,
  "session": "session_01FeA1nwBz1ohH65dvffUGKr",
  "summary": "Folded in the first out-of-scope finding, as the coordinator asked. The reading-path cards in src/components/ClusterPage.astro (40 cluster pages) now serve a raster cover through the astro:assets Image component as WebP, 400-1080w; SVG covers keep the plain img. sizes = '(max-width: 640px) calc(100vw - 2rem), 400px'. It names the drawn width, the same way as the ArticleCard commit: above 640px the image is a 180px column (150px at 641-900px) as tall as the card's text, measured at 190-227px; object-fit: cover fills it by height, so the cover is drawn 304-364px wide. On a phone the card stacks and the 16:9 image is drawn at the column width. Only the frontmatter (the import, and a width list replacing coverSrc) and the card markup change; there is no CSS change.",
  "files_changed": [
    "src/components/ClusterPage.astro"
  ],
  "replay_with_218": "git merge-tree --write-tree staged/issue-218 staged/issue-220 exits 0 with tree 1b733e9ad9c636ec39712b51632611912a5cef62. That merged ClusterPage.astro carries both #218's dark kicker colour (#3fbf8f) and this card's CARD_WIDTHS. #218's ClusterPage hunks are CSS around line 151; this card's are the import at line 2, the frontmatter at lines 34-49 and the card markup at lines 100-127, with no overlap. The whole-branch merge-tree also covers the other files both branches touch (Base.astro, BlogPost.astro, MarketingHome.astro, MarketingEvidencePage.astro), all clean",
  "numbers": {
    "profile": "same as the os-dev-report: Playwright Chromium, phone 390x844 at 3x / desktop 1366x625, 4x CPU, DevTools Slow 4G, cold cache; LCP is the median of 3 runs; bytes from a local static server with no gzip. before = bec5251 (ClusterPage unchanged since 5d685cf), after = 95ada62, back to back under the verify lock",
    "cluster /en/ai-native-app-platform/ phone": "no-scroll 903,617 → 230,655 B; full scroll 1,182,014 → 282,059 B; LCP 1688 → 1680 ms (text)",
    "cluster /en/ai-native-app-platform/ @1366x625": "no-scroll 1,182,014 → 124,226 B; full scroll same; LCP 1848 → 1828 ms (h1)",
    "dist": "68,197,374 → 68,235,078 B, 1,321 files both: the cards reuse WebP variants the ArticleCard thumbs already emit; the growth is srcset bytes in the HTML",
    "build": "pnpm build 90.5 s wall with a warm image cache (shared box); og-images 13.3 s"
  },
  "gates": "At 95ada62. pnpm check, under the lock: 'Result (150 files): 0 errors, 0 warnings, 0 hints'. pnpm build, under the lock, VERDICT command-exit 0, BUILD_EXIT=0; it includes 'svg-text-fit: OK ... (112 SVGs, 2244 text boxes, tolerance 2px, 1 allowlisted)', 'share images: 87 rendered at 1200x630 (5.2 MB, 13.3 s, 3 workers) for 879 pages; 125 bundled font files, no uncovered glyph', 'content lint --dist passed (478 built blog pages checked)' and 'content lint --dist links passed (93199 internal links in 880 built pages resolve to built files)'. pnpm seo:smoke: 'SEO smoke test passed (870 HTML pages checked; 569 sitemap URLs against 301 noindexed pages; 5424 canonical/hreflang annotations; 15654 prose blocks language-checked; 93189 internal links redirect-checked)'. pnpm content:lint and pnpm content:lint --published: 'content lint passed (339 files, 44 glossary terms checked)'. pnpm svg:lint: 'svg-text-fit: OK'. check-nul-bytes: 'OK ... no raw ASCII control bytes'. The four original JPGs are now referenced by 0 built pages (they were referenced by 8-40 before).",
  "screenshots": "/tmp/claude-0/-home-user/2224823d-3484-535e-94a5-c071ba6af395/scratchpad/issue-220/shots/cluster/: /en/ai-native-app-platform/ at 1366x625 and 390x844, light and dark, full page (*-after-*-full.png) plus the reading-path section before and after (*-reading-path.png), all after scrolling every lazy image into view. 0 unloaded images and 0 px horizontal overflow in all 8. The card image boxes are unchanged (180x227 desktop, 356x200 phone). card-image-before-vs-after-1366/390.png compare the first card's image side by side and show no visible loss. One artifact: in the 1366 section screenshots the sticky site header overlaps the section's heading; it comes from the element capture, not the page",
  "pr_body_draft": "/tmp/claude-0/-home-user/2224823d-3484-535e-94a5-c071ba6af395/scratchpad/issue-220/pr-body.md, updated: covers bullet, before/after cluster rows, browser check, gates at 95ada62, and a ClusterPage acceptance note with the merge-tree proof; first line still Fixes #220",
  "deviations": [
    "ClusterPage.astro was outside the claimed Surface. It was added at the coordinator's request; the replay overlap with #218 was checked by merge-tree, see replay_with_218",
    "Not pushed, no PR, no other branch touched. git merge-tree --write-tree writes objects only, no refs"
  ],
  "mcp_calls": "0",
  "api_writes": "1 - this addendum comment on #220 via scripts/pm/with-fleet.sh --via dispatch (one repository_dispatch to objectstack-ai/objectstack, executing POST /repos/objectstack-ai/www.objectos.ai/issues/220/comments as objectstack-fleet[bot])",
  "open_questions": [],
  "out_of_scope_findings": []
}

perf(images): WebP covers with srcset, JPEG share images, hero and th…

Landed and live · 2026-10-15T12:22Z

PR perf(images): WebP covers with srcset, JPEG share images, hero and thumbnails through astro:assets #247, squash 311a349, merged by the seat. It was replayed onto main from the seat's linear stack, patch-equal to the staged branch the seat reviewed.
Checks: PR CI Build and deploy was green on the head. On main, run 38733894996 Build and deploy succeeded on 311a349, with check, content lint, build, SEO smoke and the Pages deploy.
Housekeeping: the seat removes pm:dispatched here.
Next on the branch: Blog SVG drop shadows cut off by their filter region in 12 SVGs (hard edges at the fold), and placeholder "concept diagram" alt text in 5 locales of 4 posts #224.

Generated by Claude Code

Navigation Menu

Repository navigation

Images: 80 posts ship 1.6 MB PNG covers (mobile LCP ≈ 8 s on slow 4G), 120 marketing pages eager-load a 1.7 MB hero, and 209 posts use an SVG og:image that social networks cannot show #220

Findings

Done looks like

Surface

Order

Landed and live · 2026-10-15T12:22Z

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Images: 80 posts ship 1.6 MB PNG covers (mobile LCP ≈ 8 s on slow 4G), 120 marketing pages eager-load a 1.7 MB hero, and 209 posts use an SVG og:image that social networks cannot show #220

Description

Findings

Done looks like

Surface

Order

Activity

objectstack-fleet commented on Oct 6, 2026

objectstack-fleet commented on Oct 6, 2026

objectstack-fleet commented on Oct 6, 2026

objectstack-fleet commented on Oct 15, 2026

Landed and live · 2026-10-15T12:22Z

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions