feat(desktop): add session list density modes

[DavidMetcalfe]

Summary

• add Compact, Comfortable, and Detailed session-list density modes to Desktop Appearance settings
• surface deterministic branch, last-used model, message count, and total tool-call count metadata
• add the initial-request preview in Detailed mode while avoiding title/preview duplication
• persist the renderer-local preference and invalidate virtual-row measurements when density changes
• count labels are routed through i18n (en/ja/zh/zh-hant)

Test plan

• npx vitest run --project ui src/store/session-list-density.test.ts src/app/chat/sidebar/session-row-details.test.ts src/app/chat/sidebar/session-row-state.test.ts — 13 passed
• full --project ui suite — 398 files / 3499 tests passed
• changed-file ESLint — clean
• tsc --noEmit — clean
• npm run build — succeeded, including renderer and Electron bundles

Notes

• No backend schema, API, Dashboard, or model-call changes.
• Uses the existing SessionInfo metadata and deterministic session.preview value.
• Branch merged with current main; the density estimate is row-kind aware so date-divider rows keep their fixed height while session rows use the density-based estimate.

Closes #68119

[Bryntly]

Security review completed. No secrets leakage, XSS vectors, or injection vulnerabilities were found. The use of localStorage for the new density setting correctly sanitizes incoming values via the densityCodec, which ensures only safe, predefined literal strings are accepted. The changes are safe.

[Bryntly]

Overall, the implementation for session list density is very clean. The UI integration, virtualization cache invalidation, and density-aware layout look rock solid.

I have one minor suggestion for apps/desktop/src/app/chat/sidebar/session-row-details.ts:

Currently, countLabel will format zero counts (e.g., 0 messages, 0 tool calls). Since most sessions won't have tool calls and some may not have messages immediately, this can lead to unnecessary visual clutter in the sidebar.

Consider returning null when the count is falsy/zero to omit these labels entirely:

const countLabel = (count: number, singular: string, plural = `${singular}s`) =>
  count ? `${count} ${count === 1 ? singular : plural}` : null

Other than that, great use of the existing persistentAtom and correct fallback behavior when the title matches the preview. Code looks great!

[Bryntly]

Test Analyzer Review

I've reviewed the test coverage, quality, completeness, and correctness for PR #68124.

🔴 Critical Issues

• Missing i18n Support in Metadata Formatting: The test it('formats deterministic metadata without ambiguous call wording') in session-row-details.test.ts asserts hardcoded English strings like "26 messages · 8 tool calls". The underlying sessionRowDetails function hardcodes the words "message(s)" and "tool call(s)" without accepting localized translations. Since the PR updates zh.ts, ja.ts, etc. for density settings, these metadata strings must also be localized. The tests should be updated to verify correct translation formatting.

🟡 Important Issues

• Missing Edge Case Tests in session-row-details.test.ts:
  • The implementation uses session.title?.trim() and session.git_branch?.trim() to handle whitespace-only strings, but the tests only assert null values and not whitespace-only strings (e.g. title: " ").
  • The behavior of session.preview when it contains only whitespace isn't explicitly covered.

🟢 Strengths

• session-list-density.test.ts correctly isolates the persistentAtom initialization by utilizing vi.resetModules() and importing the module dynamically. This is a clean approach to test module-level state execution.
• Good assertions for fallback behavior (e.g., when localStorage contains an invalid value like "tiny", it correctly falls back to "comfortable").
• Density-aware virtual row estimation logic is concisely and effectively tested.

💡 Suggestions

• Consider adding tests to ensure that message_count: 0 or tool_call_count: 0 are handled appropriately (e.g. omitting the label entirely if they shouldn't be displayed).

[Bryntly]

Type Analyzer Review

I've reviewed PR #68124 for type design, invariants, and type safety, particularly around session-list-density.ts and types.ts.

🟡 Important Issues

• Duplicated Type Definition: In apps/desktop/src/app/chat/sidebar/session-row-details.ts, sessionRowEstimate re-declares the density type inline: (density: 'compact' | 'comfortable' | 'detailed'). This duplicates the single source of truth for the type. It should instead import and use SessionListDensity from @/store/session-list-density.

💡 Suggestions

• Const Assertion Strictness: In apps/desktop/src/app/settings/appearance-settings.tsx, sessionDensityOptions uses as const. While TypeScript successfully infers this and typechecks it properly via the generic SegmentedControl, consider using satisfies readonly { id: SessionListDensity; label: string }[] (and importing SessionListDensity). This enforces that all existing and future density options in SessionListDensity are explicitly defined, similar to how embedOptions works directly below it.

🟢 Strengths

• Type Safety in Local Storage Codec: The densityCodec correctly checks raw === 'compact' || raw === 'detailed' ? raw : 'comfortable'. Because the fallback is 'comfortable', the condition seamlessly handles both garbage data and the valid 'comfortable' literal without issue.
• No any Abuse: The PR successfully introduces typed state management with persistentAtom and UI options without a single use of any, ensuring proper end-to-end type safety for the new setting.

[DavidMetcalfe]

Feedback incorporated in b592a5dcd:

• sessionRowEstimate now uses the shared SessionListDensity type.
• Appearance density options now use a satisfies check against that type.
• Added coverage for whitespace-only title, branch, and preview values.
• Missing metadata is omitted from the row rather than rendering an empty subtitle.

I left zero-count labels (0 messages, 0 tool calls) intact because those are valid session facts, not missing metadata. Verified with 11 targeted UI tests, changed-file ESLint, and a filtered typecheck with no changed-file errors.

[teknium1]

Thanks for the focused renderer-local implementation. The feature is still absent on current main, but two items need attention during salvage.

Problems

• apps/desktop/src/app/chat/sidebar/session-row-details.ts:12 hardcodes English count labels. Desktop already localizes count grammar (apps/desktop/src/i18n/en.ts:870 versus apps/desktop/src/i18n/ja.ts:905), so Comfortable/Detailed metadata would remain English in Japanese and other locales.
• Current main now interleaves date-divider rows in apps/desktop/src/app/chat/sidebar/virtual-session-list.tsx:92-100; its list input is SidebarListRow[] at line 51. The density estimate must be merged without applying session-row sizing to divider rows or losing their measurement path.

Suggested changes

• Route message/tool-call labels through i18n and cover a non-English locale.
• Preserve the current divider branch and use row-kind-aware estimates when carrying the virtualizer change forward.

Automated hermes-sweeper review.

[teknium1]

This helper hardcodes English metadata labels, so Japanese/Chinese Desktop users would see English message(s) and tool call(s) strings. Please route these through the i18n layer (for example, formatter callbacks passed from the row) and add locale coverage.

[DavidMetcalfe]

@teknium1 thanks for the review. Both issues are addressed (force-pushed to PR branch).

Issue 1 — Hardcoded English labels: Routed through i18n. sessionRowDetails now accepts SessionRowFormatters with messageCount and toolCallCount callbacks. Added translations for all four locales (en/ja/zh/zh-hant). Verified with 8 targeted tests including Japanese locale coverage.

Issue 2 — Date-divider rows: Rebased onto current main (c92e2c0). estimateSize is now row-kind-aware — dividers return 28px regardless of density, session rows use sessionRowEstimate(density).

Cross-vendor review (Gemini 3.6 Flash + GPT-OSS 120B, full diff): both ACCEPTABLE. GPT-OSS caught unused imports from the rebase (fixed in a17dfb0).

Verification: ESLint clean, 11/11 targeted tests pass, MERGEABLE.

[GottZ]

This was generated by AI during triage.

Summary

One PR, #68124, addresses #68119 by implementing Desktop-local Compact, Comfortable, and Detailed session-row modes with deterministic metadata, initial-request previews, persisted presentation state, and density-aware virtualization.

Related pull requests

• feat(desktop): add session list density modes #68124 best fix — (+297/-8) — best existing fix (verify verdict: n/a): The diff implements the requested density modes, localized metadata, normalized non-duplicative previews, local persistence, and row-kind-aware virtualizer estimates. The contributor keep_open review identified hardcoded English labels and incorrect sizing for date-divider rows; the supplied diff explicitly addresses both with i18n formatter callbacks and locale coverage, plus a fixed 28px divider estimate alongside density-aware session-row sizing.

Suggested consolidation

Keep #68124 open with a salvage path: preserve its focused renderer-local preference, localized deterministic row details, preview behavior, and row-kind-aware virtualization. This follows the contributor keep_open review rather than overriding it; its two documented concerns are addressed in the supplied diff, so the updated branch should be validated against current main and any newly surfaced integration issues resolved.

Complex graph

flowchart LR
    classDef open fill:#dbeafe,stroke:#1d4ed8,color:#1e3a8a
    classDef merged fill:#dcfce7,stroke:#15803d,color:#14532d
    classDef closed fill:#e5e7eb,stroke:#6b7280,color:#1f2937
    classDef unverified fill:#f3f4f6,stroke:#9ca3af,color:#374151
    classDef best stroke-width:3px,stroke:#b45309
    classDef target stroke-width:3px,stroke:#4338ca
    I68119(["issue #68119 (open)"])
    P68124["PR #68124 (open)"]
    P68124 -->|best fix| I68119
    class I68119 open
    class P68124 open
    class P68124 best
    class P68124 target
    click I68119 "https://github.com/NousResearch/hermes-agent/issues/68119"
    click P68124 "https://github.com/NousResearch/hermes-agent/pull/68124"

Loading

Graph: solid arrow = fixes / best fix, dashed arrow = partial or unverified (see edge label); boxed group = PRs duplicating each other; amber border = best fix; indigo border = target; gray node = closed (state tag in the node label).

Cross-PR triage: Reviewed 1 pull request and 1 issue in this complex. Each diff was read against this issue; Assessment working set: 21 kB of PR diffs, 4 kB of issue/PR text, 7 kB of discussion (8 comments), 2 verify verdicts. verdicts reflect diff content, not PR titles. Part of an automated triage batch.

[teknium1]

Thanks for this fix! It was salvaged into #86771 (cherry-picked onto current main with your authorship preserved in the commit history) and is now merged. Closing since the work has landed.