Disclose complete typed-data signing requests

[HDauven]

Summary

Refs #113 and dusk-network/typed-data#4. Keep the bounded approval preview and add Full signing request (escaped JSON): native details plus a read-only, keyboard-scrollable textarea exposing full schema, strings, bytes, arrays, empty values and the trusted origin.

• Both views use one JSON snapshot checked by the shared hasher against the pending digest. Serialization, mismatch or disclosure-budget failure disables Sign while keeping Reject available.
• Neutralize Bidi_Control; visibly escape formatting controls/line separators and flag non-NFC text. Preserve original Unicode and literal escape spellings; never normalize or feed display substitutions into signing. The background still signs its captured original digest after lifecycle/context checks.
• Disclose the implicit 32-zero-byte contract default and distinguish unused metadata from hashed inputs.

The full view has an approximate early construction-work budget and a strict 2 MiB final escaped-text limit (2,097,152 ASCII characters). Serialization and escaping may temporarily construct larger text before rejection; neither check is a peak-memory guarantee. These are local signer display limits, not new hash/verifier validity rules. Both checks remain in place. Pagination is deferred.

Current-main integration / attribution

#112 is merged; this PR now targets main. GitHub already retargeted the base. Commit 3b9c056ae500be9d906041bff0684d73dd826fa4 merges current main 254fbce8732166fff8c4ad74f54953ae80fc7baa into the reviewed disclosure history without rewriting it.

Current head: b0c26bca04975e87bcc4ae7920e431c08315f858. The two original disclosure commits (8800b497d4ac, 3b23298edf8c) and their attribution are preserved. The reviewed disclosure production files and their tests are unchanged. The follow-up only refreshes README's full-view description and wires the existing approval-renderer suite into CI. The original core unit-test/build job is byte-identical to main; a small ubuntu-24.04 renderer job in the same workflow uses Playwright's standard --with-deps setup. Both jobs retain the fork/draft safeguards, read-only permissions and existing pinned actions. No shared runner image/settings were changed.

The branch now inherits the published JSR 0.1.0-rc.0 via npm:@jsr/dusk__typed-data@0.1.0-rc.0, with current main's genuine registry lock/URL/integrity, fixture-access adaptation and recent discovery/RPC regressions. Fresh standalone installs work. Native npm publication is not a prerequisite. There is no additional npm dependency, RPC, key, signing or lifecycle change relative to current main.

Based on ichbindas's #113 review, acknowledged in the original commit; his earlier integration authorship/history remain preserved. #113's limitations predated #112 and were not integration regressions. Identical bounded previews are not digest collisions, and the hypothetical line-joined example was not demonstrated injection in the boxed Wallet DOM.

Fresh validation of the combined result

• Registry-backed npm ci and full coverage: 679 tests /59 files. Chrome, Firefox and Tauri frontend builds pass.
• Existing approval-renderer suite with the real renderer and only pending-request transport stubbed: 11/11, including keyboard expansion/scrolling, full-value/schema/origin equality, Unicode/markup handling and disabled-Sign/reject controls. The same command now runs in CI.
• Fresh disposable production Chrome flow: 34/34, 14 actual signatures, no uncaught page errors. Real provider, permissions, approval UI, keys and signing; no signing mocks or transactions. Checks include original/full-view equality, NFC/NFD distinctions, accepted-but-undisclosable refusal, RC E_COMPLEXITY separately from E_POLICY_LIMIT, and profile/network/lock/disconnect cancellation.
• All 14 signatures verify through a separately installed current-main Connect SDK tarball with the registry-loaded RC, the native JSR RC, and a fresh isolated locked/offline Rust build, with changed-context/message and bare-digest rejection controls. Native JSR types/runtime require no local library mapping or sloppy imports. The SDK tarball is a tested current-main build, not a claim of a new SDK publication.
• Both installed library copies match all 76 official bridge tarball files. Frozen digest/tagged-message expectations and included BLS fixtures pass. The bridge's fixture/spec files are read as installed test data, not invented public exports.
• Real-hasher direct-helper check accepts exactly 2,097,152 final characters and refuses boundary+1. This is a helper boundary check above the RPC JSON floor, not a claim that RPC policy accepts a 2 MiB request.
• Four isolated source mutations independently detect missing digest binding (2 failures), normalization (1), missing escaping (1), and missing final-size enforcement (1); baseline/restored 41/41. Scoped author-side preflight found no remaining in-scope findings.

Executed current-head GitHub CI: 34990496437 passes on b0c26bca0497: 679 tests /59 files, 11 renderer tests, Chrome and Firefox builds. Both jobs and every step executed successfully. Both checked out 91bfd5278f30, GitHub's test merge of current main 254fbce8732166 and this head; its tree exactly equals the submitted candidate. The PR is non-draft and remains open/review-required, not merged.

Bootstrap history: 34987874463 passed 679 unit tests but could not launch Chromium without GLib; 34989895687 then exposed the core runner's lack of elevation permission. Only the renderer job moved to Ubuntu; the original core job is restored unchanged. No production/test/package files changed; the local renderer suite was rerun 11/11, and the other local evidence above applies to identical files. Failed and draft-skipped runs are not counted as passing.

The existing non-gating Codecov uploader reported that a token is required for protected-branch uploads; generated coverage/test success does not establish successful upload. Its authentication and fail_ci_if_error: false policy are unchanged.

The supplied re-review found no actionable disclosure regression; its standalone copied-function probes used stubbed hashing and are not counted as this integration evidence. Earlier 677-test/33-browser-case results are superseded by the fresh results above. Chrome signing coverage is not Firefox/Tauri signing or native-desktop E2E, and native BLS agreement is not independent encoder/security certification. Protocol v1 remains draft/unfrozen. Normal approving-review/conversation gates still apply; no merge or release is requested by this validation update.

[Copilot]

🔵 Needs a closer look

The construction budget must account for escaped representation size to prevent exceeding the stated 2 MiB bound.

Pull request overview

Adds a complete escaped-JSON typed-data signing disclosure with digest validation, Unicode safeguards, and fail-closed approval behavior.

Changes:

• Adds full request disclosure and enhanced preview handling.
• Enforces serialization, integrity, and disclosure-budget checks.
• Adds UI, test, documentation, and changelog coverage.

File summaries

File	Reviewed changes
tests/e2e/typed-data-preview.spec.js	End-to-end disclosure, interaction, Unicode, and failure coverage
src/ui/notification/app.js	Full disclosure UI and guarded approval controls
src/ui/notification.app.test.js	Notification integration tests
src/shared/typedDataDisplay.test.js	Serialization, budget, and Unicode tests
src/shared/typedDataDisplay.js	Verified disclosure serialization and display safeguards
docs/provider-api.md	Disclosure behavior and limits documentation
CHANGELOG.md	Feature and fix release notes

Review details

Suppressed comments (1)

src/shared/typedDataDisplay.js:46

• The construction budget is charged from the original string length here, but the serializer later expands every non-ASCII code unit to a six-character \\uXXXX escape (and JSON also expands some controls). A valid request containing several large non-ASCII strings can therefore build a final disclosure well over 2 MiB before the check on line 51 rejects it, defeating the stated construction bound and allowing unnecessary renderer allocation. Account for the escaped representation (or conservatively reserve its worst-case expansion) while serializing, retaining the final-size check as well.

    budget -= key.length + (typeof value === "string" ? value.length : 1) + 2 * parents.length;

• Files reviewed: 7/7 changed files
• Comments generated: 0
• Review effort level: Lite

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.