[morgmart]

What this does

Gives chartreuse a clear job in each place it shows up, so the brand color reads as a deliberate mark instead of an olive or beige stain. This follow-up to #1292 changes how links, mentions, unread cues and sidebar selection use color. It doesn't change the light-mode pale chartreuse steps themselves; that gets its own decision.

• Links: blue in light mode and bright chartreuse in dark mode from Hummin’ up. Snoozin’ stays blue. Each accent family names its own link color, so this is a per-family choice rather than a component exception.
• Mentions: a mention of someone else is a quiet neutral chip at every level, matching the composer. A mention of you joins the strong group: a brand-chartreuse pill with black text in light mode, and chartreuse text on a neutral chip in dark mode. Chips sit in their own space instead of touching the words around them. Hovering a chip now shows a hover fill.
• System notices: names in notices like “Cody added by you” are plain text, not chips.
• Unread rows: unread names keep semibold near-black (white in dark mode) on hover and selection.
• Light sidebar selection: at Hummin’ and Swarmin’, the one selected sidebar destination takes the full brand fill with black ink. Chips, dots and row actions on it switch to the selection's paired colors, so they stay readable on every accent. Hover and other selections (Inbox, members, Settings navigation) stay neutral gray until Buzzin’.
• Solo row status and actions: on a selected row, the working pill, typing and DM-count badges sit on a quiet paired surface: pale chartreuse with black dots in light mode, raised gray with chartreuse dots in dark mode, so working never looks like the row's … menu. The … menu adds no fill of its own; the row's hover and selection already show its state. The row owns one trailing position for the menu, unread dots and working status, so status can't drift or get clipped again. The working pill's right edge lines up with the … menu's, mirroring the left inset.
• Sidebar hover: only the row under the pointer changes ink. Before, hovering one row in a list (like the top pages group) inked every row beside it, which showed as all-chartreuse labels in dark mode on main.
• Dark sidebar and menus: fills stay neutral gray, and the selected row’s label carries bright chartreuse.
• Mentions in your own bubble: a slightly deeper chartreuse chip with the bubble's own near-black text, deeper again on hover, instead of a dark chip punched into the bubble.
• Vibe levels, rebalanced: Swarmin’ now adds only your own messages, on the full brand chartreuse with near-black ink in both modes, as Buzzin’ already did. The colored app background moves to Buzzin’, so each level adds one step of color. Pale and dark chartreuse bubbles read khaki or olive, so the bubble uses the bright brand fill instead.
• Color mode previews: one sidebar line in each Settings thumbnail takes that mode's selection color, so the previews reflect the accent and vibe level.
• Unread jump cue: strong only when a DM, a mention of you or a thread reply waits past the edge. Ordinary unread uses a light gray cue, so the bright pill means something at every vibe level.

Why it matters

Chartreuse only works as a bright color. Darkened, it turns olive; mixed toward white, it reads as cream next to gray. The old rules used it in exactly those failing forms: olive link text, beige selection, muddy dark rows, and a highlighter on every mention. The color now goes where it can glow (dark-mode text, strong fills) and to what deserves attention. A mention of you stands out because ordinary mentions no longer compete with it.

New rule: In dark mode, fills stay neutral and chartreuse is the ink. In light mode, chartreuse is a fill and never text, and the full brand fill is spent only where it is rare: the selected destination, strong actions, and things that need your attention.

How it works

Everything is expressed through semantic roles, so components pick no colors themselves. Links use the family’s existing link pair. Mentions get a role trio for ordinary mentions, one for mentions of you, and one for inside your own bubble, which goes tone-on-tone on the brand-filled bubble: a deeper step of chartreuse with the bubble's ink, deeper again on hover. Dark rows restate the neutral row group explicitly, because the light vibe-level rules also match a dark page. A new selected-row ink lets dark selection carry the accent, and a paired selected-row detail surface and ink covers small badges on a selection. IconButton's fill-free bare style now also covers actions revealed by a row. Ordinary row actions keep standard artwork sizing; small glyph-state toggles retain their larger artwork so their on/off state stays clear. The contrast guard checks every new pair for every accent, level and mode. Near-black ink on chartreuse counts as black, not gray.

The PR also repairs two WebKit browser tests that raced on CI: a drag released before its target rendered, and an incidental read save landed during a deliberate disconnect. Both now wait for the right moment, without retries, error allowances or weaker assertions.

Verification

Head 6099dd946, rebased on main at b2787b2a3.

• Independent design-system reviews ran on each batch. Each one caught a real defect, all fixed with regression coverage: mention hover never painted; and on accents other than chartreuse, chips on a selected group DM had white digits on near-white. A Buzzin’ hover regression was also restored.
• Unit tests pass: design system (152), plus messages, theme, channels and design-system owners (2,752). Token tests cover every new role pair across all six accents, four levels and both modes. They confirm no accent leaks into dark rows, and that light selection, hover and inks are correct at each level.
• 76 browser cases pass in Chromium and WebKit across composer links, message content, sidebar unread, sidebar polish, surfaces, appearance and Inbox. A follow-up run of 56 cases also passes after the final chip fix, and the 14 appearance cases pass with the updated previews.
• Solo sidebar browser journeys (14 cases, Chromium and WebKit) cover status alignment with the … menu at resized widths, larger text, both modes and touch, with and without the Shared action. They also check that status is never clipped, the selected badge's paired colors, and that the … menu has no fill on hover or while open. The no-fill assertion fails against the previous ghost button.
• Token tests confirm, for every accent in both modes, that Swarmin’ keeps the shell neutral and gives your bubble the brand fill with its ink. Theme, design-system and message tests (2,055) and 18 appearance, gallery and composer-link browser cases in both engines pass after the rebalance.
• A new browser test hovers one page row and confirms its sibling keeps resting ink; it fails against the previous rule. Hovering a selected Solo row's actions keeps the selection ink. 74 sidebar and Solo browser cases pass in both engines.
• After rebasing on main (which renamed Solo projects to workspaces and independently moved the row's trailing position the same way), TypeScript, 3,271 unit tests and 156 browser cases across Solo, sidebar, appearance, composer links, membership and message navigation pass in both engines.
• Inbox sizing follow-up: restored the unread-only toggle’s original artwork size without enlarging the Solo overflow glyph or restoring its fill. Design typecheck, static design checks, design build, 154 design tests, and all 30 full Inbox/Solo workspace browser cases pass in Chromium and WebKit.
• Latest rebase preserves Polish appearance controls, panel dividers, and message actions #1337’s gray accent, Accent level dropdown, slider removal, dividers and message actions, as well as Stop saving agent activity and metrics locally #1347’s saved-activity removal and fix(sidebar): stop the sidebar jumping and re-sorting when a channel is deleted #1327’s sidebar deletion fix. An independent preservation audit found no intentional incoming behavior reverted.
• Rebase validation: typecheck, design checks/build and 3,277 unit tests passed at a051b79c3 before the final clean rebase onto fix(sidebar): stop the sidebar jumping and re-sorting when a channel is deleted #1327. The 190-case browser run passed 183 cases; seven WebKit failures coincided with laptop sleep (trace stalls matched power-management logs). All five affected WebKit files then passed unchanged at 6099dd946: 55 cases, default worker count, no retries or timeout increases. This does not turn the initial interrupted run into a full pass. Mandatory push checks passed on the final head.
• Mandatory commit and push hooks passed, including TypeScript, related unit tests and design guards.
• Morgan reviewed each change in the running desktop app, in light and dark modes, as it was built.
• Not yet checked: narrow and enlarged-text layouts, and non-chartreuse accents in the running app. Hosted CI for this head is pending.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating independent Rocket (systems/lifecycle), Mantis (correctness/security), and Groot (product/accessibility) lanes.

Reviewed head da95f354171dea00723bf83467f6c0b29b31fb88 against base a3752f0.

No concrete correctness, security, or agreed-contract defect requiring changes found. Source assessment: 9/10 for minimalness, ownership and correctness. The shared link-role move preserves own-message pairing; removing the sidebar override preserves unread ink without changing selection, preparation or focus handlers. The near-black guard adjustment matches the stated product policy. One optional browser-coverage improvement is inline.

Public description and commit material inspected; no attached images and no disclosure finding in those surfaces.

Validation limits: source-only, immutable Git objects; no PR code, tests, builds or app executed. The hosted check snapshot at this head has CI required failed, with WebKit shards 2/9 and 6/9 failed; their causes were not investigated. Windows native validation was skipped. Rendered contrast, both themes/viewport sizes/enlarged text, keyboard/pointer behavior, native acceptance and human testing remain unverified here. The PR description also records deferred visual/human validation. This COMMENT is not approval or a merge-readiness claim.

[wesbillman]

🤖 Optional coverage improvement: this row is already selected (aria-current=page at line 214), so hovering it retains the selected fill and does not exercise the unselected-hover branch in ChannelSidebarRow.module.css:28–31. Consider asserting the unread color on a separate unselected unread row while hovered. The existing selected-state assertion is useful; this is a coverage suggestion, not a demonstrated product defect.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating independent Rocket (systems/lifecycle), Mantis (correctness/concurrency), and Groot (product/accessibility) lanes.

Reviewed head 73dd32a6a108b403ce16fec3c975ec056cd977ed against base a3752f0. This follow-up covers the two browser-test files changed since reviewed head da95f354171dea00723bf83467f6c0b29b31fb88; the production patch is unchanged.

No new actionable findings. Source assessment: 9/10. The reconnect test now settles the thread scroller, drains admitted read-state writes before the deliberate outage, recovers a reply withheld from live delivery, and waits for completed arrival accounting before checking focus, draft and reading position. The drag test waits for the actual insertion cue to clear before releasing the pointer. Existing keyboard/assistive activation assertions remain intact. No browser cases were added or removed.

The public PR description and both commit messages were inspected; no attached images or disclosure finding in those surfaces. The previous optional unread-hover coverage suggestion remains optional and is not duplicated here.

Validation limits: source-only; no PR code, tests, builds or apps executed. The one hosted-check snapshot still had four browser shards running; completed checks were successful except Windows native validation was skipped. Completed browser-job wall times were roughly 4–9 minutes, but check outputs supplied no test-count, summed-execution or slowest-test summaries; this review does not establish test-cost parity or eliminate flakes. The author reports 66 local browser cases passing in 2.2 minutes; that report was not independently reproduced. Rendered contrast, themes/viewports/enlarged text, native behavior and human acceptance remain unverified. This COMMENT is not approval or merge authorization.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating independent Rocket (systems/lifecycle), Mantis (correctness/security/concurrency), and Groot (product/accessibility) lanes.

Reviewed head ba1a37ffdc602d2285d6f2a3b32bbc1c6314cf55 against base 7d50425. This follow-up covers new commit 20ba36b58 and its integration with main since the previously reviewed 73dd32a6; unchanged feedback is not duplicated.

Two low-priority visual-state defects are inline. Source assessment: 8/10 pending those small state-mapping repairs. The implementation keeps shared semantic-role ownership, profile actions, unread styling and focus handlers. No security or concurrency defect found in the new changes.

Public description and new commit messages inspected; no attached images or disclosure finding in those surfaces.

Validation limits: source-only, immutable Git objects; no PR code, tests, builds or app executed. The single hosted-check snapshot had JavaScript, Rust, browser measurements/journeys and flake checks still running; completed non-skipped checks were successful, and Windows validation was skipped. No completed test-count, summed-execution or slowest-test evidence was available from that snapshot; no cost-parity or flake-elimination claim. Author-reported runtime tests were not independently reproduced. Rendered contrast, narrow/wide layouts, enlarged text, keyboard/pointer and error/retry focus behavior, native acceptance and human testing remain unverified here. This COMMENT is not approval or merge authorization.

[wesbillman]

🤖 [P3] Preserve the self-mention distinction in dark Snoozin’

At the neutral/Snoozin’ level, --affordance-prominent-mark resolves through --affordance-prominent to --neutral-12, which is also --text-standard. Both ordinary and self mentions therefore use neutral-12 ink on neutral-6, with the same neutral-5 hover fill. In a received message, a mention of the viewer looks exactly like a mention of someone else, despite the PR’s stated attention distinction and DESIGN.md’s “Link contrast” rule. Give the neutral dark self-mention pair a distinct, readable treatment while retaining the neutral palette; cover this level with an assertion that self and ordinary treatments differ.

[wesbillman]

🤖 [P3] Keep a distinct hover state for own-bubble mentions at Buzzin’

Here and in the dark full-level block at line 1397, the own-mention hover fill is identical to its resting fill. Messages.module.css passes these through --mention-fill/--mention-hover, so hovering a clickable profile mention inside an outgoing Buzzin’ bubble produces no fill change in either theme. That leaves this surface outside the PR’s promised chip-hover repair. Choose a distinct paired hover treatment in both full-level blocks and extend the browser assertion to an own-message bubble; retain the existing readable inverted resting pair and keyboard ring.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating independent Rocket (systems/lifecycle), Mantis (correctness/security), and Groot (product/accessibility) reviews.

Reviewed head db615c87ae38863d607f03c6a042b07b7e1d8b39 against base 7d50425. This follow-up checks the 11-file delta from previously reviewed ba1a37ffdc602d2285d6f2a3b32bbc1c6314cf55 in the full feature context.

One new P2 is inline: selected Solo overflow actions lose their glyph contrast on hover, press, and while their menu is open for light saturated accents. Overall 8/10 pending that paired-state fix and the existing mention repairs. The semantic token ownership and reuse of the unread-edge state are appropriately scoped.

The previous dark Snoozin’ self-mention distinction and Buzzin’ own-mention hover P3 findings remain unchanged. Those comments are not duplicated here.

Optional documentation consistency: DESIGN.md’s Rows entry still says selected unread labels remain white in dark mode, while ChannelSidebarRow.module.css:118 now intentionally uses the selected accent ink and retains semibold weight. Align that sentence with the intended selected-row rule.

Validation: immutable source and callers, CSS cascade, focus/activation and failure/recovery paths, tests as source, and public PR description/new commit were inspected. No attached images in the description and no actionable public-material disclosure found. No PR code, tests, builds, or apps were executed. One hosted-check snapshot showed JavaScript, Rust, browser journeys and several flake lanes still running; completed non-skipped checks were successful, Windows was skipped. This is not an all-green CI or runtime acceptance claim. Rendered contrast, keyboard/pointer behavior, narrow/wide layouts, enlarged text, and native acceptance remain unverified. The follow-up adds no browser cases; test cost and flake elimination were not measured. This is non-blocking COMMENT feedback, not approval or merge authorization.

[wesbillman]

🤖 [P2] Pair selected row-action ink with its interactive fills

In light Flame (also the other white-ink accents) at Hummin’/Swarmin’, this override makes the selected Solo chat’s overflow glyph white. The live caller is SoloProjects.tsx:817–827: its ghost IconButton is a sibling of the DraggableChannel wrapper containing the selected NavigationItem, so the new :has selector reaches the action. But the existing rules in this same stylesheet at lines 99–108 still paint ghost hover, active, and aria-expanded states with --affordance-panel-hover (#f5f5f6 at these levels). The icon therefore becomes white on near-white when hovered/pressed and stays that way while its menu is open. Buzzin’ likewise pairs the white glyph with a pale accent-3 fill. This violates DESIGN.md’s Contrast/State requirement to keep a control identifiable on its actual interactive fill.

Give selected ghost actions a paired hover/pressed/expanded treatment (fill and ink together), and cover a selected Solo overflow action using a white-ink accent in those states. Preserve the readable resting selected pair and focus ring.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating independent Rocket (systems/lifecycle), Mantis (correctness/security/concurrency), and Groot (product/accessibility) lanes.

Reviewed head 902d5a573415614e3642f6fc9ea0ef0973cd3543 against base 7d50425. The follow-up from db615c87ae38863d607f03c6a042b07b7e1d8b39 changes only ThemePicker.tsx and settings-controls.css: one decorative sidebar stroke now samples the selected-row color in each preview mode.

No new actionable defect in this delta. Theme scoping resolves the appropriate light fill/dark ink, and radio labels, keyboard selection, persistence and retry ownership remain intact. The delta is appropriately small. Overall source assessment remains 8/10 because the existing selected Solo overflow contrast P2, dark Snoozin’ self-mention distinction P3, and Buzzin’ own-mention hover P3 are unchanged. Their repair advice remains on the original comments; this review does not duplicate them.

Optional visual refinement: at Snoozin’, the new light-preview stroke samples neutral-3 (#e8e8e8) on the white panel, so it is much fainter than the neighboring metadata strokes. Consider a more legible neutral depiction if that detail should remain prominent. This is a decorative thumbnail sampling the real row fill; labelled radios and the selection outline remain available, so this is not an accessibility violation or required change.

Public PR description and new commit inspected; no attached images or actionable disclosure found in those surfaces. Source-only: no PR code, tests, builds or apps executed. The single hosted-check snapshot had 33 checks running, one queued, five successful, and Windows native validation skipped. Author-reported tests and desktop inspection were not independently reproduced. Rendered contrast, keyboard/pointer and error/retry focus behavior, narrow/wide layouts, enlarged text and native acceptance remain unverified. No tests changed in this follow-up; test cost and flake elimination were not measured. This is non-blocking COMMENT feedback, not approval or merge authorization.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating independent Rocket (systems/lifecycle), Mantis (correctness/security/concurrency), and Groot (product/accessibility) lanes.

Reviewed head 23a02e2c47f9fb913004c4455c770dd65b2e14d3 against base 7d50425. This follow-up checks the 13-file delta from previously reviewed 902d5a573415614e3642f6fc9ea0ef0973cd3543 in the feature and caller context.

No new actionable findings. The previous selected Solo overflow contrast and own-bubble mention hover issues have source-level repairs: bare actions retain the row’s paired fill/ink across hover and menu-open states, and own mentions now have distinct resting/hover fills. The selected-row badge pair, retained label clipping and touch trailing layout are consistent in source. The documented Swarmin’/Buzzin’ rebalance is an intentional product change.

The existing P3 dark Snoozin’ self-mention distinction remains: tokens.css:644–649, 662, 694 and 700 still resolve ordinary and self mentions to the same neutral fill, hover and ink in dark neutral mode. The original repair advice remains applicable; no duplicate inline comment is added. Overall source assessment: 8/10 pending that remaining distinction. The implementation otherwise keeps these changes with the existing semantic-token, shared-button and row-layout owners.

Validation: immutable source, callers, CSS cascade, menu dismissal/focus restoration and error/retry owners, and tests as source inspected; git diff --check passed. The public PR description and three new commit messages were inspected, with no attached images or actionable disclosure found. No PR code, tests, builds or apps executed. One hosted-check snapshot showed only Semgrep, zizmor and DCO successful; broader validation for this head was not established. Author-reported tests and desktop acceptance were not independently reproduced. Rendered contrast, narrow/wide and enlarged-text layouts, touch/keyboard and error/retry focus behavior, screen-reader/native behavior, performance and test cost remain unverified. The follow-up extends existing browser cases with geometry, clipping and menu-state checks; it adds no new browser test declarations. This is non-blocking COMMENT feedback, not approval or merge authorization.

[wesbillman]

🤖 Star Lord’s automated source review via Wes’s account, integrating independent Rocket (systems/lifecycle), Mantis (correctness/security), and Groot (product/accessibility) lanes.

Reviewed head ad656d0921263aafba89e4abffa0d7c7f909106c against base 7d50425. This follow-up covers the three-file delta from reviewed head 23a02e2c47f9fb913004c4455c770dd65b2e14d3 in its caller context.

No new actionable findings. The row-owner selectors stop hover and selected-text styling from spreading across a containing list. Selected rows retain their paired ink when their attached action is hovered. Existing NavigationItem, channel-row and Solo menu owners remain responsible for selection, activation, dismissal and focus recovery; the production delta adds no lifecycle machinery.

The existing P3 dark Snoozin’ self-mention distinction remains unchanged: ordinary and self mentions still resolve to the same neutral fill, hover and ink in dark neutral mode. Its original repair advice remains applicable; no duplicate inline comment is added. Overall source assessment remains 8/10 pending that distinction.

Validation: immutable source, callers, CSS cascade, error/retry and focus owners, and browser tests as source inspected; git diff --check passed. The public description and new commit message were inspected; no attached images or actionable disclosure found. The new browser case exercises real hover/computed colors and checks a sibling retains resting ink; the Solo case adds an assertion after action hover. One browser test declaration was added, none removed. Author-reported fail-before/pass-after evidence and desktop testing were not independently reproduced.

Source-only: no PR code, tests, builds or apps executed. The hosted-check snapshot showed only Semgrep, zizmor and DCO success; broader CI completion for this head was not established. Test-count/timing summaries were unavailable from those checks, so runtime cost and flake elimination remain unverified. Rendered contrast, keyboard/pointer behavior, narrow/enlarged layouts, screen-reader and native acceptance remain unverified here. This non-blocking COMMENT is not approval or merge authorization.