[theogravity]

What

A pane that has both created helper subshells and reported harness subagents drew two tree rails. The helper rows hung off the section's ml-5 border-l with short connector ticks, while the roster rows carried their own ml-4 line, per-row rail spans, long elbows, and a different content indent. Mixed lists read as two trees bolted together (Theo's screenshot).

Fix

HarnessSubagentRows now renders with the exact geometry the helper-subshell section uses: ml-5 border-l pl-2 container plus a before:-left-2 before:w-2 tick per row. The row button pads pl-3 and the dot-to-name gap is gap-x-2, matching the child row's link, so status dots and names share one x across both kinds.

Measured on a live stack: before, rails sat at 28 vs 32 with 12px vs 8px elbows and dots 8px apart; after, rails, elbows, dots, and names are pixel-identical.

Verification

• bun run test:changed -- --unit: 3194 pass (server-web)
• bun run verify:changed: all affected builds/type checks, e2e 97 passed / 5 skipped
• bun run lint:design: clean
• Visual: seeded a parent pane with two helper children and a three-agent roster on a throwaway stack, screenshotted the mixed expansion before/after

[theogravity]

Code review round 1: zero MAJOR, zero MINOR. Two NITs left open on purpose: the 2px rail seam at the helpers-to-roster junction is in family with the existing breaks between tree sections, and the switch-error line riding the rail only exists on the failed-switch path.