Release 0.18.3 — native screenshot renderer + cross-platform OCR

[ronaldeddings]

Release 0.18.3 — native screenshot renderer + cross-platform OCR

Bumps every surface to 0.18.3. Two headline changes.

Screenshots — native DOM renderer (html-to-image removed)

The default DOM-render screenshot hung for the full CLI timeout on any backgrounded tab — html-to-image resolved its image load inside requestAnimationFrame, which Chrome suspends for hidden tabs, so the render never completed. Replaced the dependency with a direct native renderer (getComputedStyle → inline cssText, XMLSerializer → <svg><foreignObject>, Image/decode, Canvas, FileReader/fetch for embedding images, canvas snapshots, and background-image resources).

• Works fully backgrounded, no focus required.
• Much faster on large pages: inline <svg> subtrees are deep-cloned wholesale instead of inlining styles per descendant, so SVG-heavy pages (a full Wikipedia article, ~11k px, ~2.6k icons) render in ~7s on a hidden tab instead of timing out.
• DOM-render timeout guard fails fast with a clear error instead of a silent 45s hang.
• html-to-image, its vendored runner, and its patches are removed entirely.

OCR — native-first + cross-platform (ocrad.js removed, Tesseract.js added)

• canvas ocr now returns the canvas's native accessible text (aria + fallback subtree + figcaption) plus the page's semantic textbox model, instead of low-quality pixel OCR. ocrad.js (an unmaintained pixel-OCR blob) is removed.
• New interceptor ocr <selector|element|region>: renders the target natively and OCRs it with a bundled Tesseract.js engine — offline, cross-platform, no native bridge required; returns a deterministic text string with a confidence score. canvas ocr falls back to it for pixel-only canvases. WASM core + worker + English data are bundled and loaded from extension-local URLs (wasm-unsafe-eval added to the extension CSP).

Notes

• Net +841 / −1218 (removed the html-to-image runner and ocrad).
• All surfaces bumped to 0.18.3. Test suite: 484 pass / 10 skip / 0 fail.

🤖 Generated with Claude Code

Summary by CodeRabbit

New Features

• Added OCR command to the CLI for extracting text from page elements, regions, and CSS selectors.
• Improved canvas OCR to intelligently prioritize native accessible text sources before using pixel-based recognition.

Bug Fixes

• Enhanced DOM screenshot rendering with better timeout handling and improved error guidance.

Chores

• Version updated to 0.18.3.
• Upgraded OCR processing for better accuracy and performance.

[coderabbitai]

Caution

Review failed

Pull request was closed or merged during review

📝 Walkthrough

Walkthrough

Replaces the html-to-image vendor library with a native SVG-foreignObject DOM-cloning renderer in the content script. Swaps the ocrad.js OCR engine for tesseract.js with extension-local WASM assets. Adds a new top-level ocr CLI command and background action. Changes canvas OCR to prefer accessibility text (ARIA/fallback subtree/figcaption) before falling back to pixel OCR. Bumps to v0.18.3.

Changes

Native DOM Renderer + Tesseract OCR Pipeline

Layer / File(s)	Summary
Native DOM renderer replaces html-to-image extension/src/content/dom-screenshot.ts, extension/dist-mv2/content.js, extension/src/background/capabilities/screenshot.ts, extension/dist-mv2/background-electron.js, package.json, scripts/build.sh	Removes the screenshot-runner.js injection of html-to-image and replaces the rendering pipeline with a native clone + inline-styles + SVG foreignObject rasterizer. handleDomScreenshot is rewritten to use nativeRenderToDataUrl with mode-aware dimensions and optional region cropping via cropDataUrl.
Tesseract.js OCR engine replaces ocrad.js extension/src/offscreen.ts, extension/manifest.json, scripts/build.sh, package.json	offscreen.ts gains a lazily-initialized Tesseract.js worker loading WASM/traineddata from bundled tesseract/ extension assets. ocrImage is rewritten to use the worker, returning normalized text with confidence. ocrad.js and html-to-image are removed from dependencies. Manifest gains wasm-unsafe-eval CSP and build bundles Tesseract assets.
Canvas OCR: accessibility-first, pixel OCR fallback extension/src/background/capabilities/canvas.ts, extension/dist-mv2/background-electron.js, .agents/skills/interceptor-browser/references/command-catalog.md	canvasAccessibleText() aggregates aria-label, aria-labelledby, fallback subtree textContent, figcaption, aria-describedby, and title for a canvas element. The canvas_ocr handler now selects accessibility text first, then semantic textbox text, and only runs pixel OCR as a last resort. Diagnostics fields and docs updated accordingly.
New ocr background action + DOM-render timeout guard extension/src/background/capabilities/screenshot.ts, extension/src/background/router.ts, extension/dist-mv2/background-electron.js	handleOcr renders PNG via handleDomRenderScreenshot, sends the data URL to the Tesseract offscreen worker, and returns text with confidence/dimensions. The DOM-render path is wrapped in withCaptureTimeout(DOM_RENDER_TIMEOUT_MS=30s). SCREENSHOT_ACTIONS and the dispatcher are updated to route "ocr".
CLI ocr command, transport timeouts, and routing cli/commands/screenshot.ts, cli/help.ts, cli/index.ts, cli/transport.ts	parseScreenshotCommand gains an "ocr" case parsing selector/ref/region/scale/target-max-long-edge. CLI routing adds "ocr" to known commands and introduces unwrapResult. Help text adds interceptor ocr entries and updates the canvas ocr description. ACTION_TIMEOUT_OVERRIDES_MS sets canvas_ocr and ocr to 60s.
Tests, version bumps, and manifest test/screenshot-minimized-preflight.test.ts, cli/version.ts, package.json, extension/dist-mv2/manifest.json, extension/manifest.json	Preflight tests switch from tracking scripting.executeScript injection to tracking sendMessage dispatch. Minimized-window case asserts zero sendMessage calls; non-minimized asserts dispatch proceeds. All versions bumped to 0.18.3.

Sequence Diagram(s)

sequenceDiagram
  participant CLI
  participant Background as screenshot.ts
  participant Content as dom-screenshot.ts
  participant Offscreen as offscreen.ts / Tesseract

  rect rgba(100, 149, 237, 0.5)
    note over CLI,Offscreen: New top-level OCR flow
    CLI->>Background: action { type: "ocr", selector/region }
    Background->>Content: sendToContentScript dom_screenshot (PNG, withCaptureTimeout 30s)
    Content->>Content: clone DOM, inline styles, fetch resources as data URLs
    Content->>Content: serialize SVG foreignObject → Image.decode() → canvas.toDataURL()
    Content-->>Background: { dataUrl, width, height }
    Background->>Offscreen: { type: "ocr", dataUrl }
    Offscreen->>Offscreen: getOcrWorker() lazy-init Tesseract WASM
    Offscreen-->>Background: { success: true, text, confidence }
    Background-->>CLI: { text, source, confidence, width, height }
  end

Loading

flowchart LR
  A[canvas_ocr action] --> B[canvasAccessibleText]
  B --> C{ARIA / figcaption found?}
  C -- yes --> G[return accessibility text]
  C -- no --> D[hostCanvasSignals semantic textbox]
  D --> E{semantic text found?}
  E -- yes --> H[return semantic text]
  E -- no --> F[canvas_read + Tesseract pixel OCR]
  F --> I[return OCR text or no-text hint]

Loading

Estimated code review effort

🎯 4 (Complex) | ⏱️ ~60 minutes

Possibly related PRs

• Hacker-Valley-Media/Interceptor#44: Introduced the original DOM-render screenshot pipeline (html-to-image runner injection) that this PR replaces wholesale.
• Hacker-Valley-Media/Interceptor#97: Modified handleDomRenderScreenshot and the surrounding DOM-render flow in screenshot.ts, the same code path extended with the timeout guard and OCR handler here.

Poem

🐇 No more borrowing libraries from the shelf,
I draw the DOM myself, myself, myself!
Tesseract reads the pixels with a trained eye,
While ARIA text skips the OCR — oh my!
Canvas speaks first through its accessible voice,
Version 0.18.3 — what a wonderful choice! 🎉

🚥 Pre-merge checks | ✅ 4 | ❌ 1

❌ Failed checks (1 warning)

Check name	Status	Explanation	Resolution
Docstring Coverage	⚠️ Warning	Docstring coverage is 10.64% which is insufficient. The required threshold is 80.00%.	Write docstrings for the functions missing them to satisfy the coverage threshold.

✅ Passed checks (4 passed)

Check name	Status	Explanation
Description Check	✅ Passed	Check skipped - CodeRabbit’s high-level summary is enabled.
Title check	✅ Passed	The title accurately describes the main changes: a native screenshot renderer replacing html-to-image and cross-platform OCR replacing ocrad.js, matching the core focus of the entire changeset.
Linked Issues check	✅ Passed	Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check	✅ Passed	Check skipped because no linked issues were found for this pull request.

_{✏️ Tip: You can configure your own custom pre-merge checks in the settings.}

✨ Finishing Touches

📝 Generate docstrings

• Create stacked PR
• Commit on current branch

🧪 Generate unit tests (beta)

• Create PR with unit tests
• Commit unit tests in branch release/0.18.3

---

Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out.

❤️ Share

• X
• Mastodon
• Reddit
• LinkedIn

_{Comment @coderabbitai help to get the list of available commands and usage tips.}