feat: show Conductor's brand mark on Conductor sessions
Merged
DA
2mo ago+97+97+01234567890123456789-4-4-0123456789
Summary
- Conductor sessions now carry Conductor's own letter mark instead of being identified by name alone. The 19 paths are taken verbatim from the published brand kit (https://www.conductor.build/brandkit), and the mark keeps the brand's dark/light pair — a
#282423tile with the mark in#eae8e6— rather than inheriting the muted gray the rest of the meta line uses. That is the same near-black-tile treatment as the Conductor app icon. - The mark sits inline at the start of a row's meta line rather than becoming a leading column. A leading column would leave a ragged gap on rows for providers Luke has no mark for, and it would break the meta line's ellipsis truncation; inline flow keeps both.
apps/desktop/src/renderer/provider-mark.tsxholds a registry keyed by the provider id the adapter already reports (CREDENTIAL_PROVIDER_ID.CONDUCTORfrom #16), so a mark cannot drift onto another provider's sessions. A provider with no mark in the registry renders nothing rather than a stand-in glyph, so no row claims a brand Luke has no asset for. The mark isaria-hidden: the provider name is printed right beside it, and announcing it twice only repeats what a reader already hears.- The smoke fixture's third row becomes a Conductor cloud session, so the fixture covers one session per provider and the mark appears in generated evidence. It replaces the second Codex row rather than adding a fourth: four rows overflow the 620×520 panel and the last one is clipped behind the footer.
- Nothing about the trust boundary moves. This is renderer-only presentation plus a fixture row; no adapter, credential, or IPC path is touched.
Evidence
- Platform-independent checks:
./scripts/check.shgreen — 69 desktop, 31 core, 5 harness tests. - macOS Electron verification (
./scripts/verify.sh): run on a physical Mac against this branch's tree. It packagedLuke.appand captured all three fixture PNGs (app-smoke-expanded,app-smoke-compact,app-smoke-speaking); every dimension and format assertion passed. - Visual inspection:
app-smoke-expanded.pngfrom that run, inspected and attached below. The mark is centered in its tile, correctly proportioned against the brand kit, and sits on the meta line's baseline without disturbing the two rows above it.
Expanded panel — artifacts/evidence/app-smoke-expanded.png, the packaged Luke.app rendering the smoke fixture on macOS. The third row is the new Conductor session.

Mark detail — the Conductor row from that same PNG, cropped and scaled 1.8×. Nothing is re-rendered here; it is the macOS capture enlarged so the mark is legible at reading size.

Physical-device evidence
- Screenshot or screen recording:
attachedabove — both images come from./scripts/verify.shon a physical Mac (Apple silicon, macOS 26.3), published to the sharedpr-assetsbranch underpr-17/per WORKFLOW.md step 7. - Physical-notch check:
not performed - Device/display configuration: macOS 26.3 on Apple silicon; the capture ran on a display the app reported as having a hardware notch, but the panel was not viewed against the physical notch.
Notes
- Rebased onto #16 after it merged. The reconciliation dropped a
PROVIDER_IDset this branch had added toshared/contracts.ts: #16'scredential-providers.tsalready establishes Conductor's id as the shared provider identity, and a second set naming"conductor"is exactly the drift that registry exists to prevent. The Codex and Claude Code adapters are left untouched as a result. - Follow-up: Codex and Claude Code rows still show a name with no mark. Giving them marks means sourcing each vendor's asset and licence terms, which is a separate call from using Conductor's published kit.
Automated visual evidence
Download the deterministic macOS evidence · workflow run
- Commit:
3823d33e854a30ed20174055e3af9690ff05f165 - Scenario:
smoke - Physical-notch check: not performed by CI