[brennanb2025]

	Files	Added	Deleted	Net
Test	1	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​5	0	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​5
Prod	3	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​12	$\color{#cf222e}{\Huge{\mathbf{−}}}$​6	$\color{#1a7f37}{\Huge{\mathbf{+}}}$​6

ELI5

The phone's agent history list now gets the same fast rows as the desktop. A long transcript is read only at its start and end, and its card shows no message count instead of a guessed one.

Stacked on #28927 (desktop) and #28925 (mechanism).

What Changed

Problem. After #28927 the desktop reads big transcripts by their ends, but the phone does not advertise that it can show such rows. A paired phone therefore makes its host read every one of those transcripts whole. That is slow for the phone, and it is the cost the desktop just removed.

Before / after on the phone. For a large transcript read fresh, the history card no longer shows "N messages". Title, agent, time, last message and the current-worktree badge are unchanged. Small transcripts still show their count.

Mechanism.

• The mobile app advertises aiVault.list.partial-rows.v1.
• The card view-model's messageCount becomes null for a row with contentCoverage, and the card hides the label when it is null.
• Older phone builds don't advertise the capability, so they keep getting whole rows from new hosts.

Why

This uses the same capability gate as the desktop: the host never sends a partial row to a client that would print its count as exact.

Linked Issue

N/A (internal performance lane).

Visual Proof

N/A for screenshots in this PR. The only visible change is the "N messages" label being absent on large rows, the same change as #28927's visual proof, on the mobile card.

Testing

• agent-history-session-card.test.ts: a partial row's card has no count (17/17 pass).
• Mobile oxlint is clean. These ran on a Linux machine. Mobile tsc there reports only errors from generated files missing on that machine; CI generates them.

[OrcaWin]

Closing the mobile companion after reviewing all three production changes and its test at unchanged head 6424f1f323c3cf41fc8f22e373031a5f1fb63b00.

Its complete functional scope is advertising aiVault.list.partial-rows.v1, turning a marked row's message count into null, and hiding that count. It supplies no independent transcript-reading mechanism or correctness repair. The parent #28927 has now been closed after final-head source review and production reader/parser measurements; the final implementation reads the whole file, returns the complete row/count, and adds selective tokenization rather than the description's bounded head/tail reads. On four synthetic approximately 16 MB Claude/Codex workloads it read the same bytes and produced equivalent rows while increasing reader/parser elapsed time from main's 16.7–23.5 ms to 50.5–56.8 ms. Those are reader/parser measurements, not phone or whole-app latency measurements; methodology and limits are documented in the parent closure comment.

Enabling that declined performance design and suppressing an available exact count has no remaining independent benefit under #27143's free-win/substantial-gain standard. The branch is preserved. A future demonstrated history optimization can introduce the mobile contract appropriate to its actual semantics.