feat(page-context): collect page-type signals (jsonLdType, ogType, lang) by anpete · Pull Request #2770 · duckduckgo/content-scope-scripts

anpete · 2026-06-15T18:09:56Z

Add an optional pageTypeSignals field to page-context collection, gated behind the includePageTypeSignals feature setting (off by default). It extracts schema.org JSON-LD @type, Open Graph og:type, and the declared page language, riding the existing collectionResult message so consumers (e.g. Duck.ai sidebar shortcuts) get cheap page-type hints with no extra round-trip and no LLM.

Export extractPageTypeSignals(document) for unit testing (mirrors domToMarkdown); JSON-LD parsed defensively (malformed/oversized skipped), @type collected as string or array and from @graph, deduped, casing preserved, capped.
Add jasmine unit specs.

Asana Task/Github Issue: https://app.asana.com/1/137249556945/project/1209072953775241/task/1215655051685860

Description

Adds page signal collection to the page context collection script.

Testing Steps

Added unit tests

Checklist

Please tick all that apply:

I have tested this change locally
I have tested this change locally in all supported browsers
This change will be visible to users
I have added automated tests that cover this change
I have ensured the change is gated by config
This change was covered by a ship review
This change was covered by a tech design
Any dependent config has been merged

Note

Low Risk
Feature is config-gated off by default and only adds best-effort metadata to existing collection; JSON-LD handling includes explicit bounds to avoid breaking full page collection.

Overview
Adds optional pageTypeSignals to page-context collectionResult payload when includePageTypeSignals is enabled (default off), giving consumers cheap page-type hints without an extra round-trip.

New extractPageTypeSignals harvests JSON-LD @type (string/array, @graph, deduped, order preserved), og:type, and html lang. JSON-LD parsing is defensive: malformed/oversized blocks skipped, type count and block size capped via config (maxPageTypeSignals, maxJsonLdBlockLength), recursion bounded, case-insensitive script type, hasOwnProperty for @type/@graph.

Coverage includes JSDOM unit specs, a Playwright end-to-end spec with fixture HTML/config, and registration of that spec in the Windows Playwright project.

^{Reviewed by Cursor Bugbot for commit 911adcf. Bugbot is set up for automated code reviews on this repo. Configure here.}

github-actions · 2026-06-15T18:11:30Z

[Beta] Generated file diff

Time updated: Wed, 17 Jun 2026 07:28:03 GMT

Android

    - android/contentScope.js

File has changed

Apple

    - apple/contentScope.js

File has changed

Integration

    - integration/contentScope.js

File has changed

Windows

    - windows/contentScope.js

File has changed

cursor

Stale comment

Injected PR Evaluation: Web Compatibility & Security

Web Compatibility Assessment

File Lines Severity Finding

injected/src/features/page-context.js 253–265 info Read-only DOM access (querySelector / querySelectorAll on <head> elements). No API overrides, prototype patches, or descriptor mutations — zero toString() / instanceof fidelity concerns.

injected/src/features/page-context.js 275–290 info collectJsonLdTypes only walks arrays, top-level @type, and @graph. Nested entities (e.g. mainEntity.@type) are intentionally skipped for a cheap signal. Consumers should not assume exhaustive schema.org coverage.

injected/src/features/page-context.js 593–595, 600–602 info pageTypeSignals is included in the same content cache as markdown body text. The MutationObserver watches document.body only, so head-only updates (late-injected JSON-LD / og:type changes) may not invalidate the cache until another trigger fires. Typical static <head> metadata is unaffected.

injected/src/features/page-context.js 724–728 info getPageTypeSignals() runs during collectPageContent() in init()-time collection paths — consistent with existing optional fields (includeHeadings, includeLinks, etc.). No load()-before-config hazard.

injected/unit-test/page-context-signals.spec.js all info Good unit coverage for malformed JSON, oversized blocks, deduping, and caps. Tests pass locally (13 specs, 0 failures).

No warning or error-level web-compat findings.

Security Assessment

File Lines Severity Finding

injected/src/features/page-context.js 260 info Uses page-reachable JSON.parse instead of captured JSONparse from captured-globals.js. A hostile page could substitute JSON.parse after injection; risk is limited here (parsed output is walked defensively, failures are caught, only @type strings are retained). Aligning with captured globals would be defense-in-depth consistency, not a blocker.

injected/src/features/page-context.js 243 info new Set() is uncaptured. Tampered Set could break deduping or throw; impact is localized to this optional, config-gated sub-field.

injected/src/features/page-context.js 282–289 info Bracket reads (obj['@type'], obj['@graph']) on JSON.parse output. Prototype-polluted Object.prototype could theoretically surface spurious types when own properties are absent; only trimmed strings are collected, limiting blast radius.

injected/src/features/page-context.js 593–595, 754–757 info New data flows through the existing sendContentResponse → messaging.notify('collectionResult', { serializedPageData }) path. No new nativeData fields, no bridge/origin changes, no postMessage usage. Appropriately gated behind includePageTypeSignals (default disabled).

injected/src/features/page-context.js 256–257, 726–727 info maxBlockLength (default 100k) and maxTypes (default 10) bound CPU/memory per collection. Remote config can raise these (same pattern as maxContentLength / upperLimit elsewhere in this feature).

No warning, error, or critical security findings.

Risk Level

Low Risk — Config-gated, read-only DOM/metadata extraction with bounded parsing; no API shims, messaging transport changes, or captured-globals modifications.

Recommendations

(info) When enabling includePageTypeSignals in privacy-configuration, add an integration-test case under integration-test/test-pages/page-context/ to verify end-to-end collection (mirrors existing page-context IT coverage).

(info) Consider import { JSONparse } from '../captured-globals.js' in extractJsonLdTypes for consistency with hardened features (optional hardening).

(info) Document for consumers that jsonLdType is a shallow @type/@graph harvest, not a full JSON-LD graph walk.

(info) If SPA head updates matter for your use case, consider invalidating cache on document.head mutations or re-collecting signals outside the body-only observer (not required for initial rollout).

_{Sent by Cursor Automation: Web compat and sec}

Dismissing stale approval — new commits pushed, awaiting Cursor re-review.

github-actions · 2026-06-15T18:13:12Z

Build Branch


Branch	`pr-releases/apeters/page-type-signals`
Commit	`59500c67d0`
Updated	June 17, 2026 at 7:27:42 AM UTC

Static preview entry points

Docs: docs/index.html
Static pages: build/integration/pages/index.html
Integration pages: injected/integration-test/test-pages/index.html

QR codes (mobile preview)

Entry point	QR code
Docs
Static pages
Integration pages

Integration commands

npm (Android / Extension):

npm i github:duckduckgo/content-scope-scripts#pr-releases/apeters/page-type-signals

Swift Package Manager (Apple):

.package(url: "https://github.com/duckduckgo/content-scope-scripts.git", branch: "pr-releases/apeters/page-type-signals")

git submodule (Windows):

git -C submodules/content-scope-scripts fetch origin pr-releases/apeters/page-type-signals
git -C submodules/content-scope-scripts checkout origin/pr-releases/apeters/page-type-signals

Pin to exact commit

npm (Android / Extension):

npm i github:duckduckgo/content-scope-scripts#59500c67d0806767e2e44ce9b6a167bdede3c4de

Swift Package Manager (Apple):

.package(url: "https://github.com/duckduckgo/content-scope-scripts.git", revision: "59500c67d0806767e2e44ce9b6a167bdede3c4de")

git submodule (Windows):

git -C submodules/content-scope-scripts fetch origin pr-releases/apeters/page-type-signals
git -C submodules/content-scope-scripts checkout 59500c67d0806767e2e44ce9b6a167bdede3c4de

cursor

Stale comment

Injected PR Evaluation: Web Compatibility & Security

Reviewed extractPageTypeSignals and its integration into page-context.js (+ unit tests). This is read-only DOM/metadata extraction behind a new config gate (includePageTypeSignals, default disabled). No API overrides, prototype patches, or messaging transport changes.

Web Compatibility Assessment

File Lines Severity Finding

injected/src/features/page-context.js 253 info querySelectorAll('script[type="application/ld+json"]') is case-sensitive for the attribute value. HTML type is case-insensitive; pages using APPLICATION/LD+JSON or mixed case may be missed. Low impact — signals are best-effort.

injected/src/features/page-context.js 275–290 info collectJsonLdTypes only walks arrays, top-level @type, and @graph. Nested entities (e.g. mainEntity: { "@type": "Recipe" }) are not traversed. Common on recipe/article pages; won't break sites but may under-collect.

injected/src/features/page-context.js 275–290 warning Recursive traversal has no depth cap. A page can embed deeply nested JSON-LD (within maxJsonLdBlockLength) that triggers RangeError during collection. Caught by handleContentCollectionRequest, but when includePageTypeSignals is enabled it can fail the entire content collection for that page.

injected/src/features/page-context.js 593–595 info Feature is config-gated with default disabled and follows the same optional-field pattern as includeMetaDescription / includeHeadings. Safe rollout path.

injected/unit-test/page-context-signals.spec.js all info Good unit coverage for happy paths, malformed JSON, caps, and deduping. No integration test for live DOM injection timing (acceptable for this scope).

Security Assessment

File Lines Severity Finding

injected/src/features/page-context.js 260 warning Uses uncaptured JSON.parse. captured-globals.js exports JSONparse for hostile-page hardening. A tampered JSON.parse could throw or behave unexpectedly; wrapped in try/catch so impact is limited, but inconsistent with security-sensitive injected patterns.

injected/src/features/page-context.js 243 warning Uses uncaptured new Set(). Captured Set is available from captured-globals.js. Low exploitability here, but a tampered Set could break dedup logic.

injected/src/features/page-context.js 282–289 warning Reads obj['@type'] and obj['@graph'] without own-property checks. Object.prototype pollution with @type / @graph keys can inject spurious signals into the native payload. Prefer hasOwnProperty from captured globals (see broker-protection / api-manipulation patterns).

injected/src/features/page-context.js 593–595, 754–757 info New data flows through existing serializedPageData: JSON.stringify(content) path. No new notify fields, no nativeData spread, no bridge changes.

injected/src/features/page-context.js 726–727 info maxPageTypeSignals / maxJsonLdBlockLength are read from config without upper-bound clamping. Misconfiguration could increase parse work; not a direct exploit but worth capping in code (e.g. Math.min(config, 50)).

Risk Level

Medium Risk — New injected-runtime code that reads and parses page-controlled JSON-LD, but no API shims or trust-boundary changes; properly config-gated with sensible defaults and bounds.

Recommendations

(warning) Import JSONparse and Set from captured-globals.js instead of using page-mutable globals.

(warning) Guard @type / @graph reads with captured hasOwnProperty.call(obj, key) before access.

(warning) Add a recursion-depth limit to collectJsonLdTypes (e.g. 32–64) to prevent stack overflow from adversarial nesting when the feature is enabled.

(info) Consider case-insensitive JSON-LD script discovery (script[type="application/ld+json" i] where supported, or normalize type in a manual loop).

(info) Clamp config-driven limits to safe maximums in getPageTypeSignals().

(info) Ensure includePageTypeSignals, maxPageTypeSignals, and maxJsonLdBlockLength are documented in privacy-configuration before enabling in production.

Overall this is a well-scoped, opt-in addition with solid unit tests. The main actionable items are captured-global hygiene, prototype-pollution guards, and recursion bounding before wide rollout.

_{Sent by Cursor Automation: Web compat and sec}

moon0326

I've left one minor comment, but LGTM otherwise.

@moon0326

Address PR #2770 review feedback: - Use `??` (not `||`) for maxPageTypeSignals/maxJsonLdBlockLength so an explicit 0 is honored (@moon0326). - Import captured JSONparse/Set/hasOwnProperty so a hostile page can't subvert parsing or dedup. - Guard @type/@graph reads with hasOwnProperty.call to avoid Object.prototype pollution surfacing spurious types. - Bound JSON-LD recursion (MAX_JSON_LD_DEPTH=32) so adversarial nesting can't fail the whole collection. - Clamp config-driven limits (Math.min) to safe ceilings. - Case-insensitive ld+json script discovery (APPLICATION/LD+JSON). - Add unit specs (depth cap, prototype pollution, casing, explicit 0) and a windows-project integration test (page + config) asserting end-to-end pageTypeSignals. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Dismissing stale approval — new commits pushed, awaiting Cursor re-review.

@graph

…harvest Follow-up to PR #2770 re-review: - add() now returns early once maxTypes is reached so a single wide @graph stops iterating. - Document that jsonLdType is a shallow top-level/@graph harvest (no nested mainEntity.@type). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Dismissing stale approval — new commits pushed, awaiting Cursor re-review.

cursor

Stale comment

Web Compatibility & Security Review

Web Compatibility Assessment

File Lines Severity Finding

injected/src/features/page-context.js 267–330 info Read-only DOM queries (querySelector / querySelectorAll) with no API overrides or prototype mutation. Matches existing page-context patterns (getMetaDescription, getHeadings, etc.) and is unlikely to break third-party scripts.

injected/src/features/page-context.js 267 info Case-insensitive attribute selector ([type="application/ld+json" i]) is already used in web-compat.js for viewport meta tags, so this is an established repo pattern.

injected/src/features/page-context.js 220–228 info Shallow JSON-LD harvest (top-level + @graph only) is documented in JSDoc. Consumers must not treat jsonLdType as full schema.org graph coverage — this is a product contract note, not a runtime risk.

injected/src/features/page-context.js 549–557 info MutationObserver watches document.body only (pre-existing). Head metadata (og:type, JSON-LD scripts, lang) can be stale if injected into <head> after initial collection without a body mutation. Same limitation applies to includeMetaDescription; not introduced by this PR.

injected/integration-test/page-type-signals.spec.js — info Integration test is registered only on the Windows Playwright project. Broader page-context coverage lives in pages.spec.js (chrome-mv3); consider adding a signal assertion there if cross-platform regression signal is desired.

No warning or error-level web-compat findings.

Security Assessment

File Lines Severity Finding

injected/src/features/page-context.js 9, 274, 300–309 info Sensitive operations use captured globals (JSONparse, Set, hasOwnProperty) rather than re-reading page-tamperable builtins.

injected/src/features/page-context.js 300–309 info hasOwnProperty.call() guards on @type / @graph prevent polluted Object.prototype from injecting spurious types. Unit test explicitly covers this.

injected/src/features/page-context.js 14, 255–281, 292–293 info DoS mitigations are solid: maxBlockLength skip before parse, MAX_JSON_LD_DEPTH (32) bounds recursion, maxTypes short-circuits iteration. Malformed JSON blocks are skipped without failing collection.

injected/src/features/page-context.js 613–615 info New functionality is config-gated behind includePageTypeSignals (default disabled), allowing remote rollback without deploy.

injected/src/features/page-context.js 788–791 info Signals flow through the existing serializedPageData: JSON.stringify(content) notify path. This is page-derived metadata sent to native — same trust model as title/content/headings. No new postMessage or bridge surface. nativeData is not included.

injected/src/features/page-context.js 258 info Non-string @type values are filtered, preventing object-coercion surprises in downstream consumers.

No warning, error, or critical security findings.

Risk Level

Medium Risk — New runtime signal-extraction logic in injected page-context, but no API shims, no messaging transport changes, config-gated off by default, and defensive bounds on hostile JSON-LD input.

Recommendations

(info) Add includePageTypeSignals, maxPageTypeSignals, and maxJsonLdBlockLength to the privacy-configuration schema for pageContext so remote config authors have typed, documented settings.

(info) Ensure downstream consumers understand the shallow JSON-LD scope (documented in code; may warrant a brief note in consumer docs).

(optional) Extend chrome-mv3 pages.spec.js Page Context test to assert pageTypeSignals when the setting is enabled, for broader platform coverage.

_{Sent by Cursor Automation: Web compat and sec}

anpete · 2026-06-16T06:55:25Z

@moon0326 All review comments addressed.

Dismissing stale approval — new commits pushed, awaiting Cursor re-review.

cursor

Stale comment

Injected PR Evaluation: Web Compatibility & Security

Re-evaluation after dc38d866 — prior warning-level findings are resolved.

Web Compatibility Assessment

File Lines Severity Finding

injected/src/features/page-context.js 264–327 info Read-only DOM access (querySelector / querySelectorAll on <head> elements). No API overrides, prototype patches, or descriptor mutations — zero toString() / instanceof fidelity concerns.

injected/src/features/page-context.js 264 info Case-insensitive script[type="application/ld+json" i] matches the established web-compat.js pattern (meta[name=viewport i]). On engines without Selectors Level 4 i, querySelectorAll may throw; impact is limited because includePageTypeSignals defaults to disabled and is remotely gateable.

injected/src/features/page-context.js 226–228, 289–307 info Shallow @type harvest (top-level + @graph only). Now documented in JSDoc — consumers must not assume full schema.org graph coverage.

injected/src/features/page-context.js 610–611, 617–618 info pageTypeSignals is cached with body markdown. MutationObserver watches document.body only, so head-only late injections (JSON-LD / og:type) may not invalidate the cache until another trigger fires. Typical static <head> metadata is unaffected.

injected/integration-test/page-type-signals.spec.js all info End-to-end integration test added; good coverage alongside 17 unit specs.

No warning or error-level web-compat findings.

Security Assessment

File Lines Severity Finding

injected/src/features/page-context.js 9, 252, 271 info Uses captured JSONparse, Set, and hasOwnProperty — addresses prior defense-in-depth feedback. Parsed output is walked defensively; only trimmed string @type values are retained.

injected/src/features/page-context.js 289–307 info hasOwnProperty.call before @type / @graph reads mitigates Object.prototype pollution. MAX_JSON_LD_DEPTH (32) bounds recursion. Oversized blocks skipped via maxBlockLength.

injected/src/features/page-context.js 254–260, 266 info recordType short-circuits once types.size >= maxTypes; outer loops also break early. Resolves prior perf concern about wide @graph iteration past the cap.

injected/src/features/page-context.js 757–759 info clampedNumericSetting() coerces non-finite config to fallback before Math.min, preventing NaN from disabling caps. Resolves prior config-hardening warning.

injected/src/features/page-context.js 610–611 info Gated behind includePageTypeSignals (default disabled) — required remote kill switch present.

injected/src/features/page-context.js 278 info Array.from(types) uses uncaptured global; Arrayfrom is available in captured-globals.js. Tampered Array.from could throw; impact is localized to this optional sub-field. Cosmetic consistency only.

injected/src/features/page-context.js 785–788 info pageTypeSignals flows into existing serializedPageData: JSON.stringify(content) notify path. No nativeData spread; no messaging-boundary changes.

No warning, error, or critical security findings.

Risk Level

Low Risk — Config-gated, read-only DOM metadata extraction with defensive bounds, captured globals for hostile-environment parsing, and no API shims or messaging trust-boundary changes.

Recommendations

(info) Consider wrapping document.querySelectorAll('script[type="application/ld+json" i]') in try/catch so an unsupported selector on a legacy engine cannot take down the entire collectPageContent() path when the setting is enabled.

(info) Optional: switch Array.from(types) → captured Arrayfrom(types) for consistency with repo hygiene patterns.

(info) If consumers need fresh signals after SPA head mutations, consider extending cache invalidation to <head> changes (only if product requirements demand it).

_{Sent by Cursor Automation: Web compat and sec}

anpete · 2026-06-16T17:41:07Z

@jonathanKingston Thanks, addressed feedback.

@type

Add an optional pageTypeSignals field to page-context collection, gated behind the includePageTypeSignals feature setting (off by default). It extracts schema.org JSON-LD @type, Open Graph og:type, and the declared page language, riding the existing collectionResult message so consumers (e.g. Duck.ai sidebar shortcuts) get cheap page-type hints with no extra round-trip and no LLM. - Export extractPageTypeSignals(document) for unit testing (mirrors domToMarkdown); JSON-LD parsed defensively (malformed/oversized skipped), @type collected as string or array and from @graph, deduped, casing preserved, capped. - Add jasmine unit specs. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

@moon0326

Address PR #2770 review feedback: - Use `??` (not `||`) for maxPageTypeSignals/maxJsonLdBlockLength so an explicit 0 is honored (@moon0326). - Import captured JSONparse/Set/hasOwnProperty so a hostile page can't subvert parsing or dedup. - Guard @type/@graph reads with hasOwnProperty.call to avoid Object.prototype pollution surfacing spurious types. - Bound JSON-LD recursion (MAX_JSON_LD_DEPTH=32) so adversarial nesting can't fail the whole collection. - Clamp config-driven limits (Math.min) to safe ceilings. - Case-insensitive ld+json script discovery (APPLICATION/LD+JSON). - Add unit specs (depth cap, prototype pollution, casing, explicit 0) and a windows-project integration test (page + config) asserting end-to-end pageTypeSignals. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

@graph

…harvest Follow-up to PR #2770 re-review: - add() now returns early once maxTypes is reached so a single wide @graph stops iterating. - Document that jsonLdType is a shallow top-level/@graph harvest (no nested mainEntity.@type). Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…passing the cap Malformed config (e.g. "foo") previously produced NaN from Math.min, disabling the type cap. Clamp via a finite-number check that still honors an explicit 0. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

@jonathanKingston

…e Set Address @jonathanKingston review: a Set already dedupes and preserves insertion order, so it replaces the parallel types array + seen Set. Cap via .size, drop the now-redundant slice (recordType short-circuits at the cap), and rename add -> recordType. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

Dismissing stale approval — new commits pushed, awaiting Cursor re-review.

cursor

Web Compatibility & Security Review (automation)

Reviewed 911adcff on apeters/page-type-signals. This adds config-gated, read-only page-type signal extraction (jsonLdType, ogType, lang) to the existing page-context feature. No API overrides, messaging transport changes, or prototype patches.

Web Compatibility Assessment

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	264–327	info	Read-only `querySelector` / `querySelectorAll` on `<script>`, `<meta>`, and `<html lang>` — same DOM-read pattern as existing page-context collectors. No layout/DOM mutation risk.
`injected/src/features/page-context.js`	264	info	Case-insensitive `script[type="application/ld+json" i]` selector improves discovery of non-canonical script types without affecting page scripts. Supported on all DDG WebView targets.
`injected/src/features/page-context.js`	226–228	info	Shallow `@type` harvest (top-level + `@graph` only) is documented. Consumers must not treat `jsonLdType` as full schema.org graph coverage — acceptable contract limitation.
`injected/src/features/page-context.js`	741–745	info	Signals are collected from the main `document` only; same-origin iframe content is not scanned (unlike markdown collection's `includeIframes`). Intentional scope boundary — document for downstream consumers.
`injected/src/features/page-context.js`	610–612	info	Feature is opt-in via `includePageTypeSignals` (default `disabled`), so no compat impact until remotely enabled per-domain.

No web-compat warnings or errors identified.

Security Assessment

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	9, 271, 252, 297–306	info	Correct use of captured `JSONparse`, `Set`, and `hasOwnProperty` for hostile-page hardening. Prior review feedback addressed.
`injected/src/features/page-context.js`	14, 268, 256, 290	info	DoS mitigations are solid: `MAX_JSON_LD_DEPTH=32`, `maxBlockLength` (default 100 KB, ceiling 1 MB), `maxTypes` (default 10, ceiling 50), and per-block parse try/catch so one bad block cannot fail collection.
`injected/src/features/page-context.js`	757–759	info	`clampedNumericSetting()` prevents malformed config from producing `NaN` and bypassing caps — good config-trust hygiene.
`injected/src/features/page-context.js`	278, 291, 299	info	`Array.from` / `Array.isArray` are uncaptured globals. Low practical risk here (internal iteration only), and consistent with broader codebase patterns. Optional harden: import `Arrayfrom` from `captured-globals.js`.
`injected/src/features/page-context.js`	610–612, 785–788	info	`pageTypeSignals` is added as an explicit field on the content object (not spread from untrusted input). Outbound path reuses existing `collectionResult` / `serializedPageData` channel — no new exfiltration surface beyond page-context's existing trust model. Signals are public page metadata only.
`injected/src/features/page-context.js`	297–306	info	`hasOwnProperty.call` guards on `@type` / `@graph` prevent `Object.prototype` pollution from surfacing spurious types. Covered by unit test.

No security warnings, errors, or critical findings.

Risk Level

Low Risk — Config-gated, read-only DOM inspection with bounded JSON parsing; no API shims, messaging changes, or captured-global regressions in critical paths.

Recommendations

(info) Consider importing Arrayfrom from captured-globals.js for consistency with the JSONparse/Set hardening — not blocking.
(info) Ensure privacy-configuration schema documents includePageTypeSignals, maxPageTypeSignals, and maxJsonLdBlockLength with safe defaults before rollout.
(done) Prior review items (captured globals, depth cap, prototype pollution, numeric clamping, case-insensitive discovery, tests) are addressed in commits c02ea1541–911adcff3.

Verdict: Approve from a web-compat and security perspective. Well-bounded addition with appropriate hardening and test coverage (17 unit specs + windows integration test).

_{Sent by Cursor Automation: Web compat and sec}

anpete requested review from a team and noisysocks as code owners June 15, 2026 18:09

anpete assigned moon0326 Jun 15, 2026

github-actions Bot requested a review from moon0326 June 15, 2026 18:10

daxtheduck previously approved these changes Jun 15, 2026

View reviewed changes

cursor Bot reviewed Jun 15, 2026

View reviewed changes

anpete force-pushed the apeters/page-type-signals branch from b89b88d to 3a3341d Compare June 15, 2026 18:12

daxtheduck previously approved these changes Jun 15, 2026

View reviewed changes

cursor Bot reviewed Jun 15, 2026

View reviewed changes

moon0326 reviewed Jun 15, 2026

View reviewed changes

Comment thread injected/src/features/page-context.js Outdated

moon0326 approved these changes Jun 15, 2026

View reviewed changes

daxtheduck previously approved these changes Jun 16, 2026

View reviewed changes

This comment was marked as outdated.

Sign in to view

daxtheduck previously approved these changes Jun 16, 2026

View reviewed changes

This comment was marked as outdated.

Sign in to view

daxtheduck previously approved these changes Jun 16, 2026

View reviewed changes

cursor Bot reviewed Jun 16, 2026

View reviewed changes

anpete enabled auto-merge June 16, 2026 06:57

jonathanKingston reviewed Jun 16, 2026

View reviewed changes

Comment thread injected/src/features/page-context.js Outdated

jonathanKingston reviewed Jun 16, 2026

View reviewed changes

Comment thread injected/src/features/page-context.js Outdated

jonathanKingston reviewed Jun 16, 2026

View reviewed changes

Comment thread injected/src/features/page-context.js Outdated

daxtheduck previously approved these changes Jun 16, 2026

View reviewed changes

cursor Bot reviewed Jun 16, 2026

View reviewed changes

Comment thread injected/src/features/page-context.js

Comment thread injected/src/features/page-context.js

anpete and others added 5 commits June 17, 2026 09:26

anpete force-pushed the apeters/page-type-signals branch from dc38d86 to 911adcf Compare June 17, 2026 07:26

daxtheduck approved these changes Jun 17, 2026

View reviewed changes

cursor Bot reviewed Jun 17, 2026

View reviewed changes

shakyShane approved these changes Jun 17, 2026

View reviewed changes

anpete added this pull request to the merge queue Jun 17, 2026

Merged via the queue into main with commit 71ce27f Jun 17, 2026
45 of 46 checks passed

anpete deleted the apeters/page-type-signals branch June 17, 2026 14:48

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	253–265	info	Read-only DOM access (`querySelector` / `querySelectorAll` on `<head>` elements). No API overrides, prototype patches, or descriptor mutations — zero `toString()` / `instanceof` fidelity concerns.
`injected/src/features/page-context.js`	275–290	info	`collectJsonLdTypes` only walks arrays, top-level `@type`, and `@graph`. Nested entities (e.g. `mainEntity.@type`) are intentionally skipped for a cheap signal. Consumers should not assume exhaustive schema.org coverage.
`injected/src/features/page-context.js`	593–595, 600–602	info	`pageTypeSignals` is included in the same content cache as markdown body text. The `MutationObserver` watches `document.body` only, so head-only updates (late-injected JSON-LD / `og:type` changes) may not invalidate the cache until another trigger fires. Typical static `<head>` metadata is unaffected.
`injected/src/features/page-context.js`	724–728	info	`getPageTypeSignals()` runs during `collectPageContent()` in `init()`-time collection paths — consistent with existing optional fields (`includeHeadings`, `includeLinks`, etc.). No `load()`-before-config hazard.
`injected/unit-test/page-context-signals.spec.js`	all	info	Good unit coverage for malformed JSON, oversized blocks, deduping, and caps. Tests pass locally (13 specs, 0 failures).

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	260	info	Uses page-reachable `JSON.parse` instead of captured `JSONparse` from `captured-globals.js`. A hostile page could substitute `JSON.parse` after injection; risk is limited here (parsed output is walked defensively, failures are caught, only `@type` strings are retained). Aligning with captured globals would be defense-in-depth consistency, not a blocker.
`injected/src/features/page-context.js`	243	info	`new Set()` is uncaptured. Tampered `Set` could break deduping or throw; impact is localized to this optional, config-gated sub-field.
`injected/src/features/page-context.js`	282–289	info	Bracket reads (`obj['@type']`, `obj['@graph']`) on `JSON.parse` output. Prototype-polluted `Object.prototype` could theoretically surface spurious types when own properties are absent; only trimmed strings are collected, limiting blast radius.
`injected/src/features/page-context.js`	593–595, 754–757	info	New data flows through the existing `sendContentResponse` → `messaging.notify('collectionResult', { serializedPageData })` path. No new `nativeData` fields, no bridge/origin changes, no `postMessage` usage. Appropriately gated behind `includePageTypeSignals` (default `disabled`).
`injected/src/features/page-context.js`	256–257, 726–727	info	`maxBlockLength` (default 100k) and `maxTypes` (default 10) bound CPU/memory per collection. Remote config can raise these (same pattern as `maxContentLength` / `upperLimit` elsewhere in this feature).

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	253	info	`querySelectorAll('script[type="application/ld+json"]')` is case-sensitive for the attribute value. HTML `type` is case-insensitive; pages using `APPLICATION/LD+JSON` or mixed case may be missed. Low impact — signals are best-effort.
`injected/src/features/page-context.js`	275–290	info	`collectJsonLdTypes` only walks arrays, top-level `@type`, and `@graph`. Nested entities (e.g. `mainEntity: { "@type": "Recipe" }`) are not traversed. Common on recipe/article pages; won't break sites but may under-collect.
`injected/src/features/page-context.js`	275–290	warning	Recursive traversal has no depth cap. A page can embed deeply nested JSON-LD (within `maxJsonLdBlockLength`) that triggers `RangeError` during collection. Caught by `handleContentCollectionRequest`, but when `includePageTypeSignals` is enabled it can fail the entire content collection for that page.
`injected/src/features/page-context.js`	593–595	info	Feature is config-gated with default `disabled` and follows the same optional-field pattern as `includeMetaDescription` / `includeHeadings`. Safe rollout path.
`injected/unit-test/page-context-signals.spec.js`	all	info	Good unit coverage for happy paths, malformed JSON, caps, and deduping. No integration test for live DOM injection timing (acceptable for this scope).

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	260	warning	Uses uncaptured `JSON.parse`. `captured-globals.js` exports `JSONparse` for hostile-page hardening. A tampered `JSON.parse` could throw or behave unexpectedly; wrapped in try/catch so impact is limited, but inconsistent with security-sensitive injected patterns.
`injected/src/features/page-context.js`	243	warning	Uses uncaptured `new Set()`. Captured `Set` is available from `captured-globals.js`. Low exploitability here, but a tampered `Set` could break dedup logic.
`injected/src/features/page-context.js`	282–289	warning	Reads `obj['@type']` and `obj['@graph']` without own-property checks. `Object.prototype` pollution with `@type` / `@graph` keys can inject spurious signals into the native payload. Prefer `hasOwnProperty` from captured globals (see broker-protection / api-manipulation patterns).
`injected/src/features/page-context.js`	593–595, 754–757	info	New data flows through existing `serializedPageData: JSON.stringify(content)` path. No new `notify` fields, no `nativeData` spread, no bridge changes.
`injected/src/features/page-context.js`	726–727	info	`maxPageTypeSignals` / `maxJsonLdBlockLength` are read from config without upper-bound clamping. Misconfiguration could increase parse work; not a direct exploit but worth capping in code (e.g. `Math.min(config, 50)`).

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	267–330	info	Read-only DOM queries (`querySelector` / `querySelectorAll`) with no API overrides or prototype mutation. Matches existing page-context patterns (`getMetaDescription`, `getHeadings`, etc.) and is unlikely to break third-party scripts.
`injected/src/features/page-context.js`	267	info	Case-insensitive attribute selector (`[type="application/ld+json" i]`) is already used in `web-compat.js` for viewport meta tags, so this is an established repo pattern.
`injected/src/features/page-context.js`	220–228	info	Shallow JSON-LD harvest (top-level + `@graph` only) is documented in JSDoc. Consumers must not treat `jsonLdType` as full schema.org graph coverage — this is a product contract note, not a runtime risk.
`injected/src/features/page-context.js`	549–557	info	`MutationObserver` watches `document.body` only (pre-existing). Head metadata (`og:type`, JSON-LD scripts, `lang`) can be stale if injected into `<head>` after initial collection without a body mutation. Same limitation applies to `includeMetaDescription`; not introduced by this PR.
`injected/integration-test/page-type-signals.spec.js`	—	info	Integration test is registered only on the Windows Playwright project. Broader page-context coverage lives in `pages.spec.js` (chrome-mv3); consider adding a signal assertion there if cross-platform regression signal is desired.

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	9, 274, 300–309	info	Sensitive operations use captured globals (`JSONparse`, `Set`, `hasOwnProperty`) rather than re-reading page-tamperable builtins.
`injected/src/features/page-context.js`	300–309	info	`hasOwnProperty.call()` guards on `@type` / `@graph` prevent polluted `Object.prototype` from injecting spurious types. Unit test explicitly covers this.
`injected/src/features/page-context.js`	14, 255–281, 292–293	info	DoS mitigations are solid: `maxBlockLength` skip before parse, `MAX_JSON_LD_DEPTH` (32) bounds recursion, `maxTypes` short-circuits iteration. Malformed JSON blocks are skipped without failing collection.
`injected/src/features/page-context.js`	613–615	info	New functionality is config-gated behind `includePageTypeSignals` (default disabled), allowing remote rollback without deploy.
`injected/src/features/page-context.js`	788–791	info	Signals flow through the existing `serializedPageData: JSON.stringify(content)` notify path. This is page-derived metadata sent to native — same trust model as title/content/headings. No new `postMessage` or bridge surface. `nativeData` is not included.
`injected/src/features/page-context.js`	258	info	Non-string `@type` values are filtered, preventing object-coercion surprises in downstream consumers.

File	Lines	Severity	Finding
`injected/src/features/page-context.js`	9, 252, 271	info	Uses captured `JSONparse`, `Set`, and `hasOwnProperty` — addresses prior defense-in-depth feedback. Parsed output is walked defensively; only trimmed string `@type` values are retained.
`injected/src/features/page-context.js`	289–307	info	`hasOwnProperty.call` before `@type` / `@graph` reads mitigates `Object.prototype` pollution. `MAX_JSON_LD_DEPTH` (32) bounds recursion. Oversized blocks skipped via `maxBlockLength`.
`injected/src/features/page-context.js`	254–260, 266	info	`recordType` short-circuits once `types.size >= maxTypes`; outer loops also break early. Resolves prior perf concern about wide `@graph` iteration past the cap.
`injected/src/features/page-context.js`	757–759	info	`clampedNumericSetting()` coerces non-finite config to fallback before `Math.min`, preventing NaN from disabling caps. Resolves prior config-hardening warning.
`injected/src/features/page-context.js`	610–611	info	Gated behind `includePageTypeSignals` (default `disabled`) — required remote kill switch present.
`injected/src/features/page-context.js`	278	info	`Array.from(types)` uses uncaptured global; `Arrayfrom` is available in `captured-globals.js`. Tampered `Array.from` could throw; impact is localized to this optional sub-field. Cosmetic consistency only.
`injected/src/features/page-context.js`	785–788	info	`pageTypeSignals` flows into existing `serializedPageData: JSON.stringify(content)` notify path. No `nativeData` spread; no messaging-boundary changes.

Uh oh!

Conversation

anpete commented Jun 15, 2026 • edited by cursor Bot Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Testing Steps

Checklist

Uh oh!

github-actions Bot commented Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

[Beta] Generated file diff

Uh oh!

cursor Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Injected PR Evaluation: Web Compatibility & Security

Web Compatibility Assessment

Security Assessment

Risk Level

Recommendations

Uh oh!

github-actions Bot commented Jun 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Build Branch

Static preview entry points

Integration commands

Uh oh!

cursor Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Injected PR Evaluation: Web Compatibility & Security

Web Compatibility Assessment

Security Assessment

Risk Level

Recommendations

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

Uh oh!

moon0326 left a comment

Choose a reason for hiding this comment

Uh oh!

This comment was marked as outdated.

Uh oh!

This comment was marked as outdated.

Uh oh!

cursor Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Web Compatibility & Security Review

Web Compatibility Assessment

Security Assessment

Risk Level

Recommendations

Uh oh!

anpete commented Jun 16, 2026

Uh oh!

Uh oh!

Uh oh!

Uh oh!

cursor Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Injected PR Evaluation: Web Compatibility & Security

Web Compatibility Assessment

Security Assessment

Risk Level

Recommendations

Uh oh!

Uh oh!

Uh oh!

anpete commented Jun 16, 2026

Uh oh!

cursor Bot left a comment

Choose a reason for hiding this comment

Web Compatibility & Security Review (automation)

Web Compatibility Assessment

anpete commented Jun 15, 2026 •

edited by cursor Bot

Loading

github-actions Bot commented Jun 15, 2026 •

edited

Loading

cursor Bot left a comment •

edited

Loading

github-actions Bot commented Jun 15, 2026 •

edited

Loading

cursor Bot left a comment •

edited

Loading

cursor Bot left a comment •

edited

Loading

cursor Bot left a comment •

edited

Loading