Why

Fork commit 1dff62b ("consolidate sidebar, i18n, stale-build, session fixes") raised sidebar session rows from 26px to 38px min-height while fixing the timestamp overlap. Every single-line row now carries ~12px of dead air above and below its title, so the session list reads sparse and unpolished — the operator flagged it against Claude Desktop's compact list (~27pt row pitch). Upstream NousResearch/hermes-agent is still at 26px, so this is a fork-only regression; no upstream PR is needed.

What changed

• apps/desktop/src/app/chat/sidebar/session-row.tsx — row container min-h-[2.375rem] → min-h-[1.625rem] (38px → 26px, upstream's value); title button py-1 → py-0.5 (also upstream's value).
• apps/desktop/src/app/chat/sidebar/virtual-session-list.tsx — ROW_ESTIMATE_PX 40 → 28 so the virtualizer's pre-measure estimate matches real row height again.

The fork-specific sidebar features from 1dff62b (always-visible age column, presence dot, hover slide-in actions menu) are untouched; this restores only the density tokens.

How to review

1. Read the session-row.tsx diff: two className token changes, no structural edits.
2. Confirm the trailing age/menu slot still centers vertically — it uses h-full/self-stretch driven by the same min-h, so it tracks the new height automatically.
3. In virtual-session-list.tsx, ROW_ESTIMATE_PX only seeds the virtualizer; measureElement re-measures real rows, so the constant just needs to be close (28 = 26px row + ~2px).
4. Cross-check upstream/main:apps/desktop/src/app/chat/sidebar/session-row.tsx — these are exactly upstream's density values.

Evidence

• Upstream session-row.tsx uses min-h-[1.625rem] + py-0.5, and upstream virtual-session-list.tsx uses ROW_ESTIMATE_PX = 28; the fork diverged in commit 1dff62b (2026-06-09).
• Pixel math from the operator's 2x screenshot: current fork row pitch is 39pt (38px row + 1px grid gap) vs ~27pt in Claude Desktop's session list; this PR lands at 27pt (26px + 1px gap).
• Skeleton placeholder rows (min-h-7, 28px) already match the restored density — same pairing upstream ships.

Verification

• npm run type-check (tsc -b): pass.
• npx eslint on both changed files: pass, no warnings.
• npx vitest run --environment jsdom: 456 passed / 9 failed — the identical 9 failures reproduce at baseline origin/main (ab33859) with this change stashed; baseline-red, none sidebar-related.
• npm run build (production vite build): pass.
• Live visual check (rebuild + swap /Applications/Hermes.app on ko-mac) runs immediately after merge as the deploy step; result will be posted on this PR.

Risks / gaps

• Rows and their hover/selected backgrounds get visually tighter at 26px; this matches upstream's shipped density and is exactly the operator's request — low risk.
• Click-target height shrinks from 38px to 26px; upstream ships the same 26px rows on all desktop platforms — accepted tradeoff.
• Renderer vitest suite has 9 pre-existing failures on main unrelated to this diff — already tracked as baseline-red under the hermes-agent lane.

Collaborators

• @OmarB97 (operator)
• Claude Fable 5 (ko-mac.claude)