Release v1.7.1 — QoL bug-fix batch

[mrangelmarino]

Umbrella PR for the v1.7.1 point release: 13 QoL bug fixes and small enhancements from the 2026-08-27 triage run, each landed as an individually reviewed squash commit (green CI + claude review per PR, review-fix rounds included on #836/#837).

Draft until the HITL QA loop passes. Version bump + release notes land here before undrafting.

Closes #801
Closes #827
Closes #790
Closes #832
Closes #806
Closes #830
Closes #826
Closes #829
Closes #804
Closes #799
Closes #798
Closes #797
Closes #792

Also folded in: dependabot #819 was closed in favor of a dedicated dependency pass (45 open alerts, 22 high — separate follow-up).

🤖 Generated with Claude Code

https://claude.ai/code/session_01WqmQeUBBLbtmveh54SMwza

[mrangelmarino]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.

[mrangelmarino]

/review

Auto-triggered after E2E tests passed.

[claude]

Code Review

PR #855 — Release v1.7.1 QoL batch. Covers 13 fixes across comments, footnotes, file ops, session restore, and editor UX. Reviewing for correctness, security, and performance.

Security

• src/shared/fileOperationResult.ts — The new sanitizeFileErrorMessage replacing raw OS error strings is a positive security hardening. POSIX error codes are mapped to stable, path-free messages, preventing accidental filesystem metadata leakage through IPC error surfaces. ✓

• src/main/google/sync.ts — gray-matter removed in favor of a local parseFrontmatter using the already-present js-yaml v4. The yamlLoad() call uses the default safe schema in v4 (no !!js/function type), so YAML deserialization is safe. The regex FRONTMATTER_FENCE = /^---\r?\n([\s\S]*?)\r?\n---\r?\n?/ makes the trailing newline after the closing --- optional (\r?\n?). This is slightly more permissive than gray-matter (which requires the closing delimiter to be on its own line with a trailing newline), but the behavioral difference only appears on malformed input that gray-matter would reject anyway. Acceptable. ✓

• No Electron privilege-boundary files touched (contextIsolation, nodeIntegration, preload, BrowserWindow configs unchanged). Standard security checklist: ✓

Correctness

findFullyCoveredCommentIds + setComment guard (comment clobber fix, #830)

The logic in extension.ts is correct: a comment ID is "fully covered" iff every ProseMirror text node carrying that mark lies entirely within [from, to]. The intersecting/survives set tracking handles split-able marks (partial overlap) cleanly. The dual-check (tool executor in document.ts + extension in extension.ts) is intentional: the executor returns a structured COMMENT_EXISTS error to the Claude model; the extension fires onCommentBlocked for the UI notification path. Not redundant — they serve different call sites.

Session snapshot staleness check (#829)

isSnapshotStale calls window.api?.fileStat, which is correctly defined in the preload (src/preload/index.ts:433) and main (ipc.ts file:stat handler). session.savedAt is already a field on SessionState (persistence.ts:54). The graceful fallback (return false on missing API or error) is correct: an undetectable-stale snapshot falls back to existing behavior. ✓

parseFrontmatter round-trip

stringifyFrontmatter produces ---\n[yaml]\n---\n[content]. The regex matches this and recovers the yaml block correctly. Verified round-trip is lossless for well-formed input. ✓

isSnapshotStale — clean tab fast path

if (!tabDraft.isDirty) {
  const disk = await readDocumentFromDisk(tabDraft.path)
  if (disk) { restoredContent = disk.content; restoredFrontmatter = disk.frontmatter }
}

When the disk read fails, this silently falls through to the snapshot. That is the correct behavior per the comment ("Falls back to the snapshot on read failure"). ✓

Footnote selector fix (#799)

sup a.footnote-ref → sup.footnote-ref > a with anchor.classList.add(footnote-ref) stamped onto the anchor. This matches what markdown-it-footnote actually renders and what FootnoteReference.parseHTML() expects (a.footnote-ref:first-child). The test suite directly probes ProseMirror node types post-setContent, which is the right isolation level for this bug. ✓

New FootnoteReference input rule (#798)

The override uses find: /\[\^([^\]]+)\]$/ (anchored to $) and tr.replaceWith(range.from, range.to, type.create({ data-id: crypto.randomUUID() })). The UUID is for the node identity, not the displayed footnote number — numbers are assigned dynamically by the extension view. The atomic replaceWith avoids the deleteRange + addFootnote cursor mis-placement of the base package. ✓

executeCreateAndOpenFile annotation fix

Using generateIdFromPath(fullPath) instead of reading editorStore before its async update settles is the right fix. Eliminates the race where the annotation attached to the wrong documentId. ✓

ContentEditable guard in review panels (#826)

if (e.target instanceof HTMLElement && e.target.isContentEditable) return

Added to both QuickReviewPanel and CommentReviewPanel. Correctly gates keyboard shortcuts so typing in the TipTap editor body does not trigger Accept/Reject/Next actions. ✓

Minor Observations

1. opeChange and "Not now" button — The "Not now" inline onClick and the dialog onOpenChange both attempt cursor restoration. Because React batches the state update from setPendingPasteAnnotation(null) before onOpenChange re-renders, the onOpenChange closure sees pendingPasteAnnotation === null for button-triggered closes (so toPos = null, no double-restoration). Escape-key closes correctly see the old non-null value. The ordering is safe, though it reads like a subtle trap.

2. onFileDuplicate?.(item.path) in FileTree.tsx:860 — The ?. optional call is redundant since the enclosing && already guards truthiness. Harmless style nit.

3. node!.id in the E2E clobber test (electron.comment-threading.spec.ts:31) — If node is undefined (document has no text node longer than 5 chars), the test throws a JS error rather than a clean Playwright assertion failure. Acceptable risk for a controlled fixture file, but a expect(node).toBeDefined() guard would produce a clearer failure message.

Summary

The implementation is solid. All the bug fixes are targeted and the rollout strategy (dual-check, graceful fallbacks, store-as-source-of-truth for comment counts) reflects good defensive engineering. No blocking issues found.

[github-actions]

✅ Ready to Merge

Code review found no significant issues. This PR is ready for human approval and merge.