feat(export): add Notion, Bilibili, Mastodon, Bluesky, Markdown round-trip targets

[wuwangzhang1216]

Summary

Adds five new export targets under src/lib/export/, filling out the matrix beyond WeChat / Zhihu / image / clipboard / download / deck:

• notion.ts — juice-inlined fragment for Notion paste. Drops <section>/<article> wrappers (Notion unwraps anyway) and strips class / data-* noise, but preserves language-* on <code> so Notion picks the right code-block mode.
• bilibili.ts — 专栏-compatible HTML subset. DOM walker through a tag whitelist (p, h1-6, lists, pre, code, a, img, figure, span, etc.). Images not on *.hdslb.com / *.bilibili.com get swapped for a placeholder SVG with data-original-src stashed for recovery.
• mastodon.ts / bluesky.ts — screenshot + short-caption share bundles. Reuses iframeToBlob for the image. Captions are code-point-truncated to 500 / 300 chars (no surrogate-pair / emoji slicing). Image and text are pushed through separate clipboard calls because browsers reject mixed-mime writes.
• markdown-roundtrip.ts — focused HTML → GFM walker for Hugo / 11ty / Obsidian users. Handles headings, nested lists, fenced code with language, tables, links, images, blockquotes, hr, br. Skipped pulling in turndown.

Notes

• Wire-up into export-menu.tsx is intentionally left for a follow-up so this PR stays mechanical/reviewable.
• tsc --noEmit clean.

Test plan

• Paste output from toNotionHtml into a Notion doc — verify headings, code blocks (with language), and inline color/weight survive
• Paste output from toBilibiliHtml into a 专栏 draft — confirm unsupported tags are gone and external images render as the placeholder
• buildMastodonShare / buildBlueskyShare against a sample preview iframe — verify caption truncation respects code points (try one with emoji past the limit)
• htmlToMarkdown on a deck-style document — verify nested lists, code fences with language hints, and tables render correctly in a Hugo preview

[wuwangzhang1216]

Pushed 8a40fe2 addressing the review feedback.

markdown-roundtrip.ts

• Inline code with backticks — wrapInlineCode now picks a fence longer than the longest backtick run in the body and pads with a space when the content starts/ends with ` (CommonMark rule). Backslash-escape attempt is gone.
• **Fenced code with embedded ** — `renderPre` uses the same `backtickFence(body, 3)` helper so a fenced block containing no longer terminates early.
• Link / image escaping — escapeHref wraps URLs with whitespace or parens in <…> (percent-encoding any literal </>); escapeTitle handles " and \ inside titles; escapeBrackets escapes [/] in link text and img alt without double-escaping the \ from prior escapeMd passes.
• Line-starting block markers — new escapeBlockStarts pass on paragraph + fallback inline-at-block content escapes leading #, >, -, +, *, digits. only when followed by whitespace/EOL, so a paragraph starting with 1. First no longer renders as an ordered list.
• Block content inside <li> — renderList now partitions list-item children into inline / nested-list / other-block buckets. Block children (paragraphs, code blocks, blockquotes, tables) are rendered through renderBlock with a continuation indent matching the marker width, so list items with embedded paragraphs or fenced code nest cleanly.
• escapeMd now also escapes ~ so stray ~~text~~ in plain prose doesn't become strikethrough.

bilibili.ts

• <code class> is filtered to keep only language-* tokens; arbitrary highlighter classes (hljs-keyword, etc.) are dropped.
• Removed the dead "Wrap loose inline children — caller decides" comment in the unwrap branch since no caller actually does that.

mastodon.ts

• Comment on truncateForPost now explicitly notes the function counts code points, not graphemes. ZWJ-joined emoji like 👨‍👩‍👧 count as 5 here instead of 1 — conservative, never overshoots the platform limit, but compound-emoji-heavy captions get trimmed more aggressively than strictly necessary.

tsc --noEmit is still clean. Tests deferred along with the menu wire-up.

[wuwangzhang1216]

Pushed b48e5b3 — adds 50 vitest tests for the new exporters in addition to the review fixes from the prior comment.

Test approach: "simulate real"

• happy-dom as the test environment so the exporters run against a real DOMParser / Element / TreeWalker (the same APIs the prod code uses at runtime).
• For markdown-roundtrip, every emitted Markdown sample is piped back through marked (already a runtime dependency for the editor's preview) and the resulting HTML is asserted to round-trip the structural intent. Closest "real" check possible without standing up a Hugo / 11ty / Obsidian preview.
• For mastodon.truncateForPost, tested with real surrogate pairs and a ZWJ-joined family emoji.

Coverage

File	Tests	Highlights
markdown-roundtrip.test.ts	25	Inline code containing backticks; fenced code with embedded ```; links with spaces/parens in href using the angle-bracket form; title quote escape; bracket escape in link text; paragraphs starting with `1.` / `#` / `>` / `-` staying as paragraphs; nested lists; list items with paragraph + code block; pipe escape in tables; full-document smoke.
bilibili.test.ts	9	<script>/<iframe>/<style> stripped; <div>/<section> unwrapped; class/data-*/onclick dropped; javascript: href dropped; language-* class retained on <code>; CDN vs. non-CDN image rewrite.
notion.test.ts	8	<section>/<article>/<header>/<footer>/<main> unwrap; class/data-* strip; inline style preserved; <code> language-* retention; language-plaintext fallback; <pre> without <code> gets wrapped.
mastodon.test.ts	8	Whitespace collapse, surrogate-pair safety, ZWJ-emoji conservative count, boundary cases.

pnpm test wired in package.json. tsc --noEmit clean.

Extra fixes uncovered while writing tests

• renderList: when an <li> has no inline children (e.g. <li><p>x</p><pre>…</pre></li>), the first <p>'s content is hoisted onto the marker line. Without this, marked saw 1. \n\n text and treated the next marker as a fresh list.
• notion.ts <code class> filter now keeps only language-* tokens (matches bilibili.ts); previously hljs / token-keyword / etc. could leak through.

Test plan items 1–2 (Notion / Bilibili paste) and item 4 (Hugo preview) still need manual verification with a real account, but the suite covers the underlying transformation logic that drives all three. Item 3 (buildMastodonShare / buildBlueskyShare) depends on iframeToBlob which needs a real iframe; the caption-truncation half is covered by the mastodon.test.ts suite.

[PerishCode]

@wuwangzhang1216 nice clean set of new emitters — the test matrices (DOMParser round-trip + marked validation for the Markdown emitter) are exactly the right level of coverage and caught the Unicode/codepoint edge cases I would have asked about. A few non-blocking correctness notes inline on three edge cases that the current fixtures don't cover yet: a CDN allowlist regex that admits lookalike domains, and two Markdown round-trip paths (image/code inside <a>, and <br> inside table cells) that the emitter silently mangles. Happy to land as-is if you'd rather follow up separately.

_{🔁 Powered by Looper · runner=reviewer · agent=claude-code · An autonomous AI dev team for your GitHub repos.}

[wuwangzhang1216]

Pushed d786299 — all three non-blocking findings from @PerishCode's review are addressed, with regression tests for each.

1. isBilibiliCdn lookalike-domain fix

bilibili.ts

• New regex: ^https?:\/\/(?:[^/]+\.)?(?:hdslb\.com|bilibili\.com)(?:[/?#]|$)/i
• Either an exact host match or a subdomain dot is required before the literal — evilhdslb.com, notbilibili.com, etc. now fall through to the placeholder swap.
• Tests in bilibili.test.ts:
  • rejects lookalike hostnames — asserts data-bili-placeholder="true" is set for https://evilhdslb.com/foo.png and https://notbilibili.com/foo.png.
  • accepts subdomains of the bilibili CDN — verifies i0.hdslb.com, album.bilibili.com, and bare apex hdslb.com / bilibili.com all pass through untouched.

2. <img> / <code> nested inside <a> no longer mangled

markdown-roundtrip.ts

• Took option A from the suggestion: moved [ / ] escape into escapeMd (text-node level) and dropped the post-composition escapeBrackets(inner()) wrapper. escapeBrackets helper deleted.
• Nested rendered structure now flows into […](href) untouched, so <a><img alt=a src=y></a> becomes [![a](y)](href) and round-trips back into <a><img> correctly.
• Tests in markdown-roundtrip.test.ts:
  • preserves <img> nested inside <a> (badge-style link) — asserts the round-tripped DOM has <a> containing an <img> with the original src/alt.
  • preserves <code> nested inside <a> — asserts the <code> survives and contains no stray backslashes inside its textContent.
• Existing escapes [ and ] inside link text test still passes — text-node bracket escape covers that case too.

3. Table cells with <br> no longer break the table

markdown-roundtrip.ts renderTable

• Cells now collapse \r?\n+ to literal <br> (inline HTML is allowed inside GFM cells) before pipe-escaping, so a literal \n from a <br> / <p> no longer terminates the table row early.
• Test in markdown-roundtrip.test.ts:
  • keeps a cell containing <br> on a single row — <table><tr><th>H</th></tr><tr><td>line1<br>line2</td></tr><tr><td>next</td></tr></table> round-trips with 3 <tr>s (header + 2 body), and the first body cell contains both text fragments plus a <br>.

pnpm test → 55/55 green (was 50). tsc --noEmit clean.

[lefarcen]

Thanks for the detailed fix notes, @wuwangzhang1216 — that makes the follow-up easy to track.

@PerishCode could you re-check the current head when you have a moment? The author says the three non-blocking findings from your review are addressed with regression coverage. @qiongyu1999, your product approval is now stale after the push as well, so a quick current-head re-approval would unblock the product gate once review/mergeability are otherwise ready.

[PerishCode]

Thanks for the PR and the follow-up fixes. I rebased this branch onto the latest main after the public workspace / CI harness landed.

What changed:

• Moved the new exporter modules from src/lib/export/... to next/src/lib/export/....
• Moved the Vitest coverage into next/src/lib/export/__tests__/... so it runs through the Next package test command.
• Kept the root package as workspace metadata only; no root scripts or root vitest.config.ts were reintroduced.
• Kept the feature scope library-only: the new targets are still not wired into ExportMenu, matching the original PR intent.

Validation:

• pnpm install --frozen-lockfile
• pnpm exec tsx scripts/guard.ts
• pnpm -F @html-anything/next typecheck
• pnpm -F @html-anything/e2e typecheck
• pnpm -F @html-anything/next test — 59 tests passed
• pnpm -F @html-anything/next build
• pnpm -F @html-anything/e2e test — 3 tests passed
• git diff --check

Risk: Low-to-medium. The rebase was limited to directory/package-shape alignment, and the exporter test suite stayed green. Product wiring for these targets remains out of scope for this PR.

[wuwangzhang1216]

Thanks @PerishCode for taking the time to rebase this onto the new next/ workspace layout — much appreciated. I pulled the rebased branch locally and re-ran the full validation matrix to confirm nothing regressed in transit:

• pnpm exec tsx scripts/guard.ts — passed
• pnpm -F @html-anything/next typecheck — clean
• pnpm -F @html-anything/next test — 59/59 passed
• pnpm -F @html-anything/next build — succeeded

I've also gone through and resolved the three outdated review threads from the original review pass. Each of the findings landed in 3a0f402 with regression coverage:

• isBilibiliCdn lookalike-hostname tightening — covered by the rejects lookalike hostnames / accepts subdomains of the bilibili CDN cases in bilibili.test.ts.
• Bracket-escape mangling of <img> / <code> nested inside <a> — covered by the preserves <img> nested inside <a> and preserves <code> nested inside <a> cases in markdown-roundtrip.test.ts.
• Table-cell <br> collapsing into a literal newline — covered by keeps a cell containing <br> on a single row.

With those threads closed, the PR is now APPROVED + CLEAN and the ruleset gates are satisfied. Ready for a squash-merge whenever it's convenient — thanks again to both reviewers for the careful pass.