RigelBuild · rigel-mintaka · Aug 23, 2026
@@ -303,6 +303,11 @@ check enforces the mechanical half. Full rationale:
 | DL-232 | The Cmd/Ctrl+K command palette host anatomy is a bare fixed wrapper: a permanently-open Kobalte **Search** primitive (`open` pinned true; Control + Input + Listbox, with `Search.Portal`/`Search.Content` replaced by a plain div per Kobalte's published command-menu recipe, <https://kobalte.dev/docs/core/components/search/>, `@kobalte/core` v0.13.13) inside a hand-positioned `position:fixed` div that IS `.cx-palette`, with its OWN `.cx-palette-backdrop`, Escape/outside-click dismiss, hand-wired focus grant (`ref` + `input.focus()` on mount) and focus-restore on close, plus per-row `onMouseDown`-preventDefault so a result click never blurs the pinned-open input — NO Portal/Content, floating-ui out of the test path. Search over Combobox because Search is built for caller-managed filtering (the palette's own fuzzy match) and publishes this pinned-open/no-portal recipe verbatim, where the same shape on bare Combobox was an unsupported extrapolation. The decider: a DOM-only test path with Kobalte reserved for the Search primitive's ARIA + keyboard traversal (DL-150's a11y-hard carve-out). A Search/Combobox hosted inside a Kobalte `Dialog` (scrim/trap/restore free) was rejected for nested-primitive + Portal/z-layer reconciliation cost against the shipped `.cx-palette` CSS and the window-level `Mod+K` toggle. | Active (Matt, 2026-08-23) | [command palette §Decisions](compass-command-palette/design.md#decisions) |
 | DL-233 | The palette's navigation mode ships ALL six destination kinds, including issue AND pr. The issue provider rides the tracker's assigned-issue seam (`store.assignedIssues`); the pr provider requires a NEW store-level PR accessor — `prs: Accessor<PrRow[]>` as a `createMemo` over `issues()` via the existing pure `prRows()` (`board.ts:132-134`), mapping each `{issue, pr}` row to a `Destination{kind:"pr"}` that selects the owning issue and reveals the PR pane (`store.selectIssue` + `store.setActiveRightTab("pr")`) — which Matt explicitly accepted as added palette-impl scope (the store has no PR collection today; `IssueTab`'s `"pr"` is a detail-pane tab, not a collection). The fixture already carries PR shapes (`stub-data.ts:238`), so no new offline data source is needed. | Active (Matt, 2026-08-23) | [command palette §Decisions](compass-command-palette/design.md#decisions) |
 | DL-234 | Point-of-use shortcut chips ship the seam plus the view-button now-win: the `ShortcutChip` primitive + `shortcutFor(id, platform)` + the `PrimaryAction` contract note + populated `board.*` shortcut fields (chips uniformly via `shortcutFor`, no hand-authored shortcut strings), AND `aria-keyshortcuts` + `title` populated via `shortcutFor` on the LeftSidebar `.bridge-link` view buttons (`LeftSidebar.tsx:437-482`: Bridge/Backlog/Done/Settings — the surfaces that actually invoke the D6-seeded `show*` paths), plus the Bridge topbar tab (`App.tsx:68-94`) for parity. Actual `.cx-menu`/hover-tooltip chip adoption is out of scope, deferred to follow-up issue RIG-2530 (it belongs to the first surface that mounts one). Seeds `view.settings/backlog/done` registrations stay under the shared commands-as-inventory rule (DL-229, owned by RIG-2529): the palette's seeds are `scope:"global"` view commands, so the palette is correct on the shared rule. | Active (Matt, 2026-08-23) | [command palette §Decisions](compass-command-palette/design.md#decisions) |
+| DL-248 | Compass adopts Linear's leader/mnemonic keyboard-chord model, focus-context-gated (Matt-ratified fork): bare-letter and `G <key>` leader chords are modifier-less and inherit the EXISTING editable-target guard (`dispatch.ts:117-122`, extended to cover `<select>`/`combobox`/`listbox`/`menu`), so focus context — not a per-route switch — is the surface-awareness mechanism (Linear/Slack/Zulip converge here); "lighter on comms" is EMERGENT (a focused composer dormant-gates bare chords), never a per-route disable flag. Composes with the three-tier resolution (DL-222) unchanged. | Active (Matt, 2026-08-23) | [leader chords §A1](compass-leader-chords/design.md#a1--the-collision-guard-is-already-built-focus-gating-via-the-editable-target-guard) |
+| DL-249 | A leader sequence is authored as space-separated segments in the EXISTING `KeymapEntry.chord` string (`"G B"`), no new field — collision-free because the Space key is already normalized to the token `"Space"` (`dispatch.ts:45`), so no event-derived chord contains a space and sequence rows are inert in the `DEFAULT_KEYMAP.filter` (`dispatch.ts:112-114`); the leader-prefix set is DERIVED from the table via `leaderPrefixes`, never hard-coded; authoring rules (T1 invariant test): exactly two modifier-less segments, and a leader prefix never doubles as a complete single chord. A parallel `sequence?: string[]` field was rejected (forks every consumer for zero gain). | Active (Matt, 2026-08-23) | [leader chords §A2](compass-leader-chords/design.md#a2--sequence-authoring-space-separated-segments-in-keymapentrychord) |
+| DL-250 | The leader runtime is pending-leader closure state INSIDE the one `installKeymap` keydown handler (never a second listener, honoring DL-223): guard-FIRST ordering (a modifier-less key on an editable/interactive target types, never arms — the guard moves ahead of the `matching.length===0` early return at `dispatch.ts:115`), arm on a leader-set hit (`!event.repeat`) with a `LEADER_TIMEOUT_MS` disarm timeout, completion resolves the two-segment chord through the UNCHANGED three tiers (DL-222), dead sequences fall through RE-ENTERING arming (a re-pressed leader re-arms), Escape disarms, Mod-chords disarm-and-run, the uninstaller clears the timer. A separate leader-mode dispatcher was rejected (duplicate guard + tier logic). | Active (Matt, 2026-08-23) | [leader chords §A3](compass-leader-chords/design.md#a3--the-leader-runtime-pending-state-inside-the-one-keydown-handler) |
+| DL-251 | A leader sequence is NEVER written into `aria-keyshortcuts` (WAI-ARIA has no sequential-press grammar — `"G B"` reads as "G OR B"): the shipped scanners harden ASYMMETRICALLY (refining DL-234's `shortcutFor` contract) — `shortcutForAria` (`keymap.ts:70-76`) SKIPS sequence rows so a sequence-only command emits no attribute, while `shortcutFor` (`keymap.ts:57-63`) FORMATS them via one pure `formatChordForDisplay` (`"G then B"`) so point-of-use display (sidebar `title`, palette chip) is preserved. A dual-bound command keeps its modifier chord in both. Minting parallel replacement helpers was rejected (leaves the shipped ones emitting the bug). | Active (Matt, 2026-08-23) | [leader chords §A5](compass-leader-chords/design.md#a5--display-and-aria-keyshortcuts-for-sequences) |
+| DL-252 | Wave-1 leader allocation is `G B`/`G L`/`G D`/`G S` → view.bridge/backlog/done/settings — all four commands ALREADY registered by the shipped spine (`spine.ts:79-126`) and wired (`store.ts:1973-1980`), so wave 1 adds ONLY keymap rows (no spine/store change); `view.agentWorkspace` gets no `G A` (spine registers no such command). Zero bare letters in wave 1 (the board's verbs are group-relative; Linear's a/s/p act on issue fields Compass's board doesn't yet edit). Coexists additively with all existing `Mod+*` bindings (DL-152); deprecating overlapping modifier chords is a deferred later call (record OQ5). | Active (Matt, 2026-08-23) | [leader chords §A4](compass-leader-chords/design.md#a4--bare-letter-and-sequence-allocation-wave-1) |
 
 ## Bridge spawn control