diff --git a/CHANGELOG.md b/CHANGELOG.md
index f86982c0..852a6eb0 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,5 +1,35 @@
 # Changelog
 
+## 0.1.43 - 2026-09-04
+
+MendCode v0.1.43 adds the current OpenAI GPT-5.6 family and GPT-6 Astra to
+the model and Codex OAuth surfaces while keeping the terminal session view
+compact during concurrent work.
+
+### Changed
+
+- Add GPT-6 Astra plus GPT-5.6 Sol, Terra, and Luna API and ChatGPT OAuth
+  presets with documented limits, pricing, reasoning levels, and fast/pro
+  variants where the provider exposes them.
+- Keep the OpenAI model picker populated from an in-process fallback when a
+  cached or bundled `models.dev` catalog predates the current model family;
+  live provider metadata still takes precedence.
+- Keep Commands history in the Ctrl+T horizontal widget tray with a direct
+  action to open the originating session.
+- Record the Codex CLI `0.153.3` comparison and the compatibility decisions in
+  the release audit.
+
+### Fixed
+
+- Allow GPT-6 Astra through the ChatGPT OAuth model filter with its documented
+  context and input limits while retaining the existing Responses Lite scope
+  for GPT-5.6 tiers.
+
+### Tests
+
+- Add model-catalog fallback, model-preset, Astra OAuth filtering, mode
+  normalization, and context-limit regressions.
+
 ## 0.1.42 - 2026-09-03
 
 MendCode v0.1.42 makes queued prompts, cancellation, reconnection, and communication between sessions reliable under concurrent work.
diff --git a/docs/README.md b/docs/README.md
index e6f9882b..92fd8383 100644
--- a/docs/README.md
+++ b/docs/README.md
@@ -40,6 +40,7 @@ Use this index by intent. **Use MendCode** explains the public workflow, **Featu
 - [Package index](package-index.md): workspace packages versus runtime `.mendcode` packages.
 - [MendCode Source Map](source-map.md): one maintained map from public features to implementation and test paths.
 - [Releasing](releasing.md): installer contract, checksums, release notes, and public installer smoke tests.
+- [Codex CLI and model audit](audits/codex-cli-0.153.3-gpt56-2026-09-04.md): upstream comparison and the GPT-5.6/GPT-6 Astra compatibility decisions for 0.1.43.
 - [Changelog](../CHANGELOG.md): versioned behavior changes and regression coverage for recent releases.
 - [Supply chain security](supply-chain-security.md): provenance, SBOM, pinned actions, dependency review, and scanner policy.
 - [Public readiness audit](public-readiness-audit.md): branch, secret, dependency, legacy-reference, and public-surface checks.
diff --git a/docs/audits/codex-cli-0.153.3-gpt56-2026-09-04.md b/docs/audits/codex-cli-0.153.3-gpt56-2026-09-04.md
new file mode 100644
index 00000000..017a11b3
--- /dev/null
+++ b/docs/audits/codex-cli-0.153.3-gpt56-2026-09-04.md
@@ -0,0 +1,46 @@
+# Codex CLI and OpenAI model audit — 2026-09-04
+
+This audit records the upstream Codex CLI comparison used for the 0.1.43
+release. The reference CLI stable release was `0.153.3` (`rust-v0.153.3`).
+OpenAI's current model pages identify the flagship model as `gpt-6-astra` and
+the 5.6 family as `gpt-5.6`/Sol, Terra, and Luna.
+
+## Ported in 0.1.43
+
+- Add `gpt-6-astra`, `gpt-5.6`, `gpt-5.6-sol`, `gpt-5.6-terra`, and
+  `gpt-5.6-luna` to the OpenAI catalog fallback. A stale cache or bundled
+  `models.dev` snapshot no longer hides these models from the picker.
+- Preserve live `models.dev` entries when they exist, so the fallback cannot
+  replace newer provider metadata.
+- Add API and ChatGPT OAuth presets with documented context limits and API
+  prices. GPT-6 Astra keeps its 1.05M context/922K input limit; ChatGPT OAuth
+  Sol, Terra, and Luna keep the existing 256K effective limit imposed by the
+  Responses Lite adapter.
+- Recognize Astra and its documented `fast` variant in the ChatGPT OAuth
+  catalog. Astra stays on the standard Codex Responses transport; the current
+  MendCode Responses Lite compatibility contract is limited to the 5.6 tiers.
+- Keep the compact Commands history in the Ctrl+T horizontal widget tray,
+  including its direct session-open action.
+
+## Reviewed and intentionally not ported
+
+- Upstream's asynchronous model-picker refresh (`5fc7840cf6d0`) is a larger
+  TUI/server contract change. MendCode already refreshes `models.dev` in the
+  provider service; this release adds the static safety net without changing
+  picker lifecycle or selection state.
+- Luna Reserve usage fallback, app-server collaboration-mode discovery, live
+  context-compaction status, rate-limit banners, permission-profile discovery,
+  startup warning condensation, automatic thread naming, and submission-ID
+  steer acknowledgement were reviewed from the upstream history. Each needs a
+  separate MendCode contract and regression surface, so none is represented as
+  shipped behavior here.
+- The upstream CLI version is not copied into MendCode's `version` request
+  header. The existing `0.144.0` value belongs to MendCode's internal
+  Responses Lite compatibility contract, and the upstream source contains no
+  equivalent current header contract to justify changing it.
+
+## Sources
+
+- [OpenAI GPT-6 Astra model documentation](https://developers.openai.com/api/docs/models/gpt-6-astra)
+- [OpenAI GPT-5.6 Sol model documentation](https://developers.openai.com/api/docs/models/gpt-5.6-sol)
+- [OpenAI Codex CLI releases](https://github.com/openai/codex/releases)
diff --git a/docs/features.md b/docs/features.md
index 163507fa..88d3e822 100644
--- a/docs/features.md
+++ b/docs/features.md
@@ -433,7 +433,9 @@ MendCode normalizes supported model aliases, preserves session affinity while th
 effective instructions stay the same, and rotates that affinity when instructions
 change. GPT-5.6 OAuth models expose a 256K effective input/context limit and a
 provider-aware compaction threshold so long sessions compact before the provider
-rejects the request.
+rejects the request. GPT-6 Astra is also available when the OpenAI catalog or
+ChatGPT account exposes it, with its documented 1.05M context and 922K input
+limits.
 
 ## Model Roles
 
diff --git a/src/mendcode/packages/extensions/zed/extension.toml b/src/mendcode/packages/extensions/zed/extension.toml
index 4339d9dd..12195a9d 100644
--- a/src/mendcode/packages/extensions/zed/extension.toml
+++ b/src/mendcode/packages/extensions/zed/extension.toml
@@ -1,7 +1,7 @@
 id = "mendcode"
 name = "MendCode"
 description = "The open source coding agent."
-version = "0.1.42"
+version = "0.1.43"
 schema_version = 1
 authors = ["MendCode"]
 repository = "https://github.com/MendCode/MendCode"
@@ -11,26 +11,26 @@ name = "MendCode"
 icon = "./icons/opencode.svg"
 
 [agent_servers.mendcode.targets.darwin-aarch64]
-archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.42/mendcode-darwin-arm64.zip"
+archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.43/mendcode-darwin-arm64.zip"
 cmd = "./mendcode"
 args = ["acp"]
 
 [agent_servers.mendcode.targets.darwin-x86_64]
-archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.42/mendcode-darwin-x64.zip"
+archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.43/mendcode-darwin-x64.zip"
 cmd = "./mendcode"
 args = ["acp"]
 
 [agent_servers.mendcode.targets.linux-aarch64]
-archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.42/mendcode-linux-arm64.tar.gz"
+archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.43/mendcode-linux-arm64.tar.gz"
 cmd = "./mendcode"
 args = ["acp"]
 
 [agent_servers.mendcode.targets.linux-x86_64]
-archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.42/mendcode-linux-x64.tar.gz"
+archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.43/mendcode-linux-x64.tar.gz"
 cmd = "./mendcode"
 args = ["acp"]
 
 [agent_servers.mendcode.targets.windows-x86_64]
-archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.42/mendcode-windows-x64.zip"
+archive = "https://github.com/MendCode/MendCode/releases/download/v0.1.43/mendcode-windows-x64.zip"
 cmd = "./mendcode.exe"
 args = ["acp"]
diff --git a/src/mendcode/packages/opencode/package.json b/src/mendcode/packages/opencode/package.json
index cd992d84..4c70e975 100644
--- a/src/mendcode/packages/opencode/package.json
+++ b/src/mendcode/packages/opencode/package.json
@@ -1,6 +1,6 @@
 {
   "$schema": "https://json.schemastore.org/package.json",
-  "version": "0.1.42",
+  "version": "0.1.43",
   "name": "mendcode",
   "type": "module",
   "license": "MIT",
diff --git a/src/mendcode/packages/opencode/script/dev-rearm-smoke.ts b/src/mendcode/packages/opencode/script/dev-rearm-smoke.ts
index 543721d7..65142f79 100644
--- a/src/mendcode/packages/opencode/script/dev-rearm-smoke.ts
+++ b/src/mendcode/packages/opencode/script/dev-rearm-smoke.ts
@@ -2,7 +2,7 @@ import { createServer, type IncomingMessage, type ServerResponse } from "node:ht
 import { once } from "node:events"
 import { probeRuntime } from "./dev-rearm"
 
-const VERSION = "0.1.42"
+const VERSION = "0.1.43"
 const PROJECT = "/tmp/mendcode-dev-rearm-smoke-project"
 
 function assert(condition: unknown, message: string): asserts condition {
diff --git a/src/mendcode/packages/opencode/src/cli/cmd/tui/component/agent-command-panel.tsx b/src/mendcode/packages/opencode/src/cli/cmd/tui/component/agent-command-panel.tsx
new file mode 100644
index 00000000..9c59630e
--- /dev/null
+++ b/src/mendcode/packages/opencode/src/cli/cmd/tui/component/agent-command-panel.tsx
@@ -0,0 +1,141 @@
+import { createMemo, createSignal, For, Show } from "solid-js"
+import type { TuiThemeCurrent } from "@mendcode/plugin/tui"
+import { Locale } from "@/util/locale"
+import {
+  agentViewCommandStateRank,
+  formatAgentViewCommandSummary,
+  isAgentViewCommandActionable,
+  type AgentViewCommand,
+} from "../util/agent-view"
+
+export function AgentCommandPanel(props: {
+  commands: readonly AgentViewCommand[]
+  sessionID: string
+  width: number
+  height?: number
+  busyID?: string
+  theme: Pick<TuiThemeCurrent, "text" | "textMuted" | "accent" | "error">
+  onUpdate: (command: AgentViewCommand, state: "accepted" | "rejected") => void
+  onOpenSession?: (sessionID: string) => void
+}) {
+  const [expanded, setExpanded] = createSignal(false)
+  const active = createMemo(() =>
+    props.commands
+      .filter((command) => agentViewCommandStateRank(command) < 2)
+      .toSorted(
+        (a, b) =>
+          Number(b.targetSessionID === props.sessionID && isAgentViewCommandActionable(b)) -
+            Number(a.targetSessionID === props.sessionID && isAgentViewCommandActionable(a)) ||
+          agentViewCommandStateRank(a) - agentViewCommandStateRank(b) ||
+          b.time.updated - a.time.updated,
+      ),
+  )
+  const history = createMemo(() =>
+    props.commands
+      .filter((command) => agentViewCommandStateRank(command) >= 2)
+      .toSorted((a, b) => b.time.updated - a.time.updated),
+  )
+  const visible = createMemo(() => [...active().slice(0, 4), ...(expanded() ? history().slice(0, 4) : [])])
+  const rows = () => (
+    <>
+      <For each={visible()}>
+        {(command) => {
+          const incoming = () => command.targetSessionID === props.sessionID
+          const actionable = () => incoming() && isAgentViewCommandActionable(command)
+          const busy = () => props.busyID === command.id
+          const sender = () =>
+            incoming() ? command.payload?.sourceTitle || command.sourceSessionID : command.targetSessionID
+          const relatedSessionID = () => (incoming() ? command.sourceSessionID : command.targetSessionID)
+          const canOpen = () =>
+            Boolean(props.onOpenSession && relatedSessionID() && relatedSessionID() !== props.sessionID)
+          const trailingWidth = () => (canOpen() ? 7 : 0) + (actionable() ? 19 : 0)
+          return (
+            <box width="100%" flexDirection="column" flexShrink={0}>
+              <box width="100%" flexDirection="row" gap={1}>
+                <text fg={props.theme.text} wrapMode="none" flexGrow={1} minWidth={0}>
+                  {Locale.truncate(
+                    `${incoming() ? "←" : "→"} ${Locale.truncate(sender().replace(/\s+/g, " "), 24)} · ${formatAgentViewCommandSummary(command)}`,
+                    Math.max(1, props.width - trailingWidth()),
+                  )}
+                </text>
+                <Show when={canOpen()}>
+                  <text
+                    id={`agent-command-open-${command.id}`}
+                    fg={props.theme.textMuted}
+                    wrapMode="none"
+                    flexShrink={0}
+                    onMouseUp={() => {
+                      const sessionID = relatedSessionID()
+                      if (sessionID) props.onOpenSession?.(sessionID)
+                    }}
+                  >
+                    [open]
+                  </text>
+                </Show>
+                <Show when={actionable()}>
+                  <text
+                    id={`agent-command-accept-${command.id}`}
+                    fg={busy() ? props.theme.textMuted : props.theme.accent}
+                    wrapMode="none"
+                    flexShrink={0}
+                    onMouseUp={() => {
+                      if (!busy()) props.onUpdate(command, "accepted")
+                    }}
+                  >
+                    [accept]
+                  </text>
+                  <text
+                    id={`agent-command-reject-${command.id}`}
+                    fg={busy() ? props.theme.textMuted : props.theme.error}
+                    wrapMode="none"
+                    flexShrink={0}
+                    onMouseUp={() => {
+                      if (!busy()) props.onUpdate(command, "rejected")
+                    }}
+                  >
+                    [reject]
+                  </text>
+                </Show>
+              </box>
+              <Show when={expanded() && command.error}>
+                <text fg={props.theme.textMuted} wrapMode="none">
+                  {Locale.truncate(command.error!.replace(/\s+/g, " "), props.width)}
+                </text>
+              </Show>
+            </box>
+          )
+        }}
+      </For>
+      <Show when={active().length > 4 || (expanded() && history().length > 4)}>
+        <text fg={props.theme.textMuted} wrapMode="none">
+          More commands in Agent View
+        </text>
+      </Show>
+    </>
+  )
+
+  return (
+    <Show when={props.commands.length > 0}>
+      <box width={props.width} height={props.height} flexDirection="column" flexShrink={0}>
+        <text
+          id="agent-command-history-toggle"
+          fg={props.theme.textMuted}
+          wrapMode="none"
+          onMouseUp={() => setExpanded((value) => !value)}
+        >
+          {history().length > 0 ? (expanded() ? "▾ " : "▸ ") : ""}Commands
+          {active().length > 0 ? ` · ${active().length} active` : ""}
+          {history().length > 0 ? ` · ${history().length} past` : ""}
+        </text>
+        <Show
+          when={props.height !== undefined}
+          fallback={rows()}
+        >
+          <scrollbox height={Math.max(1, props.height! - 1)} width="100%">
+            {rows()}
+          </scrollbox>
+        </Show>
+      </box>
+    </Show>
+  )
+}
diff --git a/src/mendcode/packages/opencode/src/cli/cmd/tui/component/dialog-command.tsx b/src/mendcode/packages/opencode/src/cli/cmd/tui/component/dialog-command.tsx
index be135267..54855aee 100644
--- a/src/mendcode/packages/opencode/src/cli/cmd/tui/component/dialog-command.tsx
+++ b/src/mendcode/packages/opencode/src/cli/cmd/tui/component/dialog-command.tsx
@@ -231,7 +231,7 @@ const titleOverrides: Record<string, string> = {
   "session.context": "View Context Usage",
   "session.permission.status": "Change Approval Mode",
   "session.toggle.sticky_user_header": "Toggle Sticky User Header",
-  "session.toggle.todos": "Toggle Todos",
+  "session.toggle.todos": "Toggle Session Widgets",
   "session.toggle.thinking": "Toggle Thinking",
   "session.toggle.timestamps": "Toggle Timestamps",
   "session.toggle.actions": "Toggle Tool Details",
diff --git a/src/mendcode/packages/opencode/src/cli/cmd/tui/routes/session/index.tsx b/src/mendcode/packages/opencode/src/cli/cmd/tui/routes/session/index.tsx
index 15b7ec0a..3db17978 100644
--- a/src/mendcode/packages/opencode/src/cli/cmd/tui/routes/session/index.tsx
+++ b/src/mendcode/packages/opencode/src/cli/cmd/tui/routes/session/index.tsx
@@ -191,6 +191,7 @@ import {
   sessionBottomDockLayout,
   sessionTodoIcon,
   sessionTodoPanelWidth,
+  sessionWidgetTrayContentWidth,
   type SessionTodo,
 } from "../../util/session-bottom-dock"
 import { renderSessionExitSummary } from "../../util/session-exit-summary"
@@ -220,12 +221,10 @@ import {
   unavailableReasoningLabel,
 } from "@/mend/tui/presentation"
 import { blurCompactionArcade, CompactionPanel, isCompactionArcadeFocused } from "../../component/compaction-panel"
+import { AgentCommandPanel } from "../../component/agent-command-panel"
 import {
   agentViewCommandStateRank,
   agentViewCommandTouchesSession,
-  formatAgentViewCommandSummary,
-  formatAgentViewCommandState,
-  formatAgentViewCommandType,
   isAgentViewCommandActionable,
   type AgentViewCommand,
 } from "../../util/agent-view"
@@ -1006,7 +1005,9 @@ export function Session() {
   const [headerTitleOffset, setHeaderTitleOffset] = createSignal(0)
   const [showGenericToolOutput, setShowGenericToolOutput] = kv.signal("generic_tool_output_visibility", false)
   const [showCompactedToolCalls, setShowCompactedToolCalls] = kv.signal(COMPACTED_TOOL_CALLS_KV_KEY, false)
-  const [showTodos, setShowTodos] = kv.signal("session_todos_visible", false)
+  // Keep the existing KV key so users who had the dock visible keep that preference.
+  const [showSessionWidgets, setShowSessionWidgets] = kv.signal("session_todos_visible", false)
+  const [focusSessionWidgets, setFocusSessionWidgets] = createSignal(false)
   const [followSessionOutput, setFollowSessionOutput] = createSignal(true)
   let submitFollowSyncTimer: ReturnType<typeof setTimeout> | undefined
   let submitFollowSyncToken = 0
@@ -1355,7 +1356,7 @@ export function Session() {
     const name = workflow?.name ? `${workflow.name} · ` : ""
     return Locale.truncate(`${name}${state}${progress ? ` · ${progress}` : ""}`, Math.max(12, contentWidth() - 16))
   })
-  const showSessionBottomDock = createMemo(() => showTodos() && !disabled())
+  const showSessionBottomDock = createMemo(() => showSessionWidgets() && !disabled())
   const promptDisabled = createMemo(() => disabled())
 
   const agentCommandURL = (path: string, query?: Record<string, string | undefined>) => {
@@ -1428,19 +1429,6 @@ export function Session() {
     }
   }
 
-  const incomingAgentCommands = createMemo(() =>
-    agentCommands().filter((command) => command.targetSessionID === route.sessionID),
-  )
-  const sentAgentCommands = createMemo(() =>
-    agentCommands().filter(
-      (command) => command.sourceSessionID === route.sessionID && command.targetSessionID !== route.sessionID,
-    ),
-  )
-  const visibleAgentCommands = createMemo(() => [
-    ...incomingAgentCommands().slice(0, 4),
-    ...sentAgentCommands().slice(0, 2),
-  ])
-
   createEffect(
     on(
       () => route.sessionID,
@@ -3617,6 +3605,16 @@ export function Session() {
 
   const command = useCommandDialog()
 
+  function toggleSessionWidgets(dialog?: ReturnType<typeof useDialog>) {
+    setShowSessionWidgets((visible) => {
+      const next = !visible
+      setFocusSessionWidgets(next)
+      if (!next) setTimeout(() => prompt?.focus(), 0)
+      return next
+    })
+    dialog?.clear()
+  }
+
   // Esc must cancel the active turn even when a transcript, widget, or an
   // autocomplete renderable owns focus. Pending approval/question UIs retain
   // their own Esc semantics (reject/close) and are deliberately excluded.
@@ -3700,16 +3698,15 @@ export function Session() {
       },
     },
     {
-      title: showTodos() ? "Hide todos" : "Show todos",
+      title: showSessionWidgets() ? "Hide session widgets" : "Show session widgets",
       value: "session.toggle.todos",
       keybind: "todo_toggle",
       category: "Session",
       description: todos().length
-        ? `${todos().filter((todo) => todo.status !== "completed").length} open todos`
-        : "No todos in this session.",
+        ? `${todos().filter((todo) => todo.status !== "completed").length} open todos · horizontal widget tray`
+        : "Open the horizontal widget tray for this session.",
       onSelect: (dialog) => {
-        setShowTodos((prev) => !prev)
-        dialog.clear()
+        toggleSessionWidgets(dialog)
       },
     },
     {
@@ -4395,92 +4392,6 @@ export function Session() {
     }
   })
 
-  const agentCommandPanel = () => (
-    <Show when={visibleAgentCommands().length > 0}>
-      <box
-        paddingLeft={contentInset()}
-        paddingRight={contentInset()}
-        width={insetRowWidth()}
-        flexDirection="column"
-        gap={0}
-        flexShrink={0}
-      >
-        <text fg={theme.accent} wrapMode="none">
-          ⟡ Commands
-        </text>
-        <For each={visibleAgentCommands()}>
-          {(command) => {
-            const incoming = command.targetSessionID === route.sessionID
-            const actionable = incoming && isAgentViewCommandActionable(command)
-            const busy = agentCommandBusyID() === command.id
-            const compact = insetRowWidth() < 76
-            const statusLine = () =>
-              actionable ? (
-                <box flexDirection="row" flexShrink={0}>
-                  <text fg={theme.textMuted} wrapMode="none">
-                    mouse:{" "}
-                  </text>
-                  <text
-                    fg={busy ? theme.textMuted : theme.accent}
-                    wrapMode="none"
-                    onMouseUp={() => void updateAgentCommand(command, "accepted")}
-                  >
-                    [accept]
-                  </text>
-                  <text fg={theme.textMuted} wrapMode="none">
-                    {" "}
-                  </text>
-                  <text
-                    fg={busy ? theme.textMuted : theme.error}
-                    wrapMode="none"
-                    onMouseUp={() => void updateAgentCommand(command, "rejected")}
-                  >
-                    [reject]
-                  </text>
-                </box>
-              ) : (
-                <text fg={theme.textMuted} wrapMode="none">
-                  {formatAgentViewCommandState(command)}
-                </text>
-              )
-            return (
-              <box
-                width="100%"
-                flexDirection="column"
-                border={["left"]}
-                customBorderChars={SplitBorder.customBorderChars}
-                borderColor={actionable ? theme.accent : theme.backgroundPanel}
-                paddingLeft={1}
-                paddingTop={0}
-                paddingBottom={0}
-              >
-                <box width="100%" flexDirection="row" overflow="hidden">
-                  <text fg={incoming ? theme.text : theme.textMuted} wrapMode="none">
-                    {incoming ? "in" : "sent"} · {formatAgentViewCommandSummary(command)}
-                  </text>
-                </box>
-                <box width="100%" flexDirection={compact ? "column" : "row"} overflow="hidden">
-                  <box width={compact ? "100%" : undefined} overflow="hidden">
-                    <text fg={theme.textMuted} wrapMode="none">
-                      {formatAgentViewCommandType(command)} ·{" "}
-                      {incoming
-                        ? `from ${Locale.truncate(command.sourceSessionID, compact ? 14 : 18)}`
-                        : `to ${Locale.truncate(command.targetSessionID, compact ? 14 : 18)}`}
-                    </text>
-                  </box>
-                  <Show when={!compact}>
-                    <box flexGrow={1} minWidth={1} />
-                  </Show>
-                  {statusLine()}
-                </box>
-              </box>
-            )
-          }}
-        </For>
-      </box>
-    </Show>
-  )
-
   return (
     <context.Provider
       value={{
@@ -4590,7 +4501,6 @@ export function Session() {
                 </box>
               )}
             </Show>
-            {agentCommandPanel()}
             <box position="relative" flexGrow={1} width="100%">
               <For each={[transcriptRenderKey()]}>
                 {() => (
@@ -4850,8 +4760,14 @@ export function Session() {
                 <SessionBottomDock
                   todos={todos()}
                   subagents={subagents()}
+                  commands={agentCommands()}
+                  commandBusyID={agentCommandBusyID()}
                   width={contentWidth()}
                   sessionID={route.sessionID}
+                  autoFocus={focusSessionWidgets()}
+                  onAutoFocus={() => setFocusSessionWidgets(false)}
+                  onUpdateCommand={(command, state) => void updateAgentCommand(command, state)}
+                  onOpenSession={(sessionID) => navigate({ type: "session", sessionID })}
                   onOpenSubagent={(sessionID) => navigate({ type: "session", sessionID })}
                   info={{
                     branch: topBranchLabel(),
@@ -5339,6 +5255,10 @@ function dockWidgetWidth(item: MendWidgetEntry) {
   return undefined
 }
 
+function dockWidgetMinWidth(item: MendWidgetEntry) {
+  return Math.max(1, typeof item.width === "number" ? item.width : (item.minWidth ?? 18))
+}
+
 function dockWidgetHeight(item: MendWidgetEntry, fallback: number) {
   if (typeof item.height === "number") return Math.max(1, Math.min(fallback, item.height))
   return fallback
@@ -5347,8 +5267,7 @@ function dockWidgetHeight(item: MendWidgetEntry, fallback: number) {
 function dockWidgetsMinWidth(items: MendWidgetEntry[]) {
   if (!items.length) return 0
   return items.reduce((total, item, index) => {
-    const width = typeof item.width === "number" ? item.width : (item.minWidth ?? 18)
-    return total + Math.max(1, width) + (index > 0 ? 1 : 0)
+    return total + dockWidgetMinWidth(item) + (index > 0 ? 1 : 0)
   }, 0)
 }
 
@@ -5483,14 +5402,23 @@ function RenderMendWidget(props: { item: MendWidgetEntry }) {
 function SessionBottomDock(props: {
   todos: SessionTodo[]
   subagents: SessionSubagentInfo[]
+  commands: readonly AgentViewCommand[]
+  commandBusyID?: string
   width: number
   sessionID: string
   info: SessionBottomInfo
+  autoFocus?: boolean
+  onAutoFocus?: () => void
+  onUpdateCommand: (command: AgentViewCommand, state: "accepted" | "rejected") => void
+  onOpenSession: (sessionID: string) => void
   onOpenSubagent: (sessionID: string) => void
 }) {
   const { theme } = useTheme()
   const mend = useMendTuiProfile()
   const dockWidgets = createMemo(() => listMendWidgets("sessionBottomDock"))
+  const commandWidth = createMemo(() =>
+    props.commands.length > 0 ? Math.max(42, Math.min(72, Math.floor(props.width * 0.55))) : 0,
+  )
   const builtinDockWidgetIDs = ["todo", "notes", "subagents", "info"]
   const profileControlsBuiltins = createMemo(() => {
     const widgets = mend.profile.widgets
@@ -5518,48 +5446,112 @@ function SessionBottomDock(props: {
       },
     }),
   )
+  const trayContentWidth = createMemo(() =>
+    sessionWidgetTrayContentWidth({
+      dockWidth: layout().dockWidth,
+      widgetWidths: [
+        commandWidth(),
+        layout().todoWidth,
+        layout().showNotes ? layout().notesWidth : 0,
+        layout().showSubagents ? layout().subagentsWidth : 0,
+        layout().showInfo ? layout().infoWidth : 0,
+        ...dockWidgets().map(dockWidgetMinWidth),
+      ],
+    }),
+  )
+  const trayOverflow = createMemo(() => trayContentWidth() > layout().dockWidth)
+  const trayHeight = createMemo(() => layout().dockHeight + (trayOverflow() ? 1 : 0))
+  let tray: ScrollBoxRenderable | undefined
+
+  onMount(() => {
+    if (!props.autoFocus) return
+    tray?.focus()
+    props.onAutoFocus?.()
+  })
 
   return (
     <box flexShrink={0} width="100%" paddingBottom={1}>
-      <box
+      <scrollbox
+        ref={(value: ScrollBoxRenderable) => {
+          tray = value
+        }}
         width={layout().dockWidth}
-        height={layout().dockHeight}
-        flexDirection="row"
-        gap={1}
-        alignItems="stretch"
-        overflow="hidden"
+        height={trayHeight()}
+        scrollX
+        scrollY={false}
+        contentOptions={{ width: trayContentWidth(), minWidth: trayContentWidth() }}
+        horizontalScrollbarOptions={{
+          visible: trayOverflow(),
+          trackOptions: {
+            backgroundColor: theme.backgroundElement,
+            foregroundColor: theme.border,
+          },
+        }}
+        verticalScrollbarOptions={{ visible: false }}
       >
-        <SessionTodoPanel todos={props.todos} width={layout().todoWidth} height={layout().dockHeight} />
-        <Show when={layout().showNotes}>
-          <SessionNotesWidget sessionID={props.sessionID} width={layout().notesWidth} height={layout().dockHeight} />
-        </Show>
-        <Show when={layout().showSubagents}>
-          <SessionSubagentsWidget
-            subagents={props.subagents}
-            width={layout().subagentsWidth}
-            height={layout().dockHeight}
-            onOpen={props.onOpenSubagent}
-          />
-        </Show>
-        <Show when={layout().showInfo}>
-          <SessionInfoWidget info={props.info} width={layout().infoWidth} height={layout().dockHeight} />
-        </Show>
-        <For each={dockWidgets()}>
-          {(item) => (
+        <box
+          width={trayContentWidth()}
+          minWidth={trayContentWidth()}
+          height={layout().dockHeight}
+          flexDirection="row"
+          gap={1}
+          alignItems="stretch"
+          overflow="visible"
+        >
+          <Show when={props.commands.length > 0}>
             <box
-              flexShrink={item.width === "auto" ? 1 : 0}
-              width={dockWidgetWidth(item)}
-              minWidth={item.minWidth ?? 18}
-              maxWidth={item.maxWidth}
-              height={dockWidgetHeight(item, layout().dockHeight)}
+              width={commandWidth()}
+              minWidth={commandWidth()}
+              height={layout().dockHeight}
+              flexShrink={0}
               overflow="hidden"
+              paddingLeft={1}
+              paddingRight={1}
               backgroundColor={theme.backgroundPanel}
             >
-              <RenderMendWidget item={item} />
+              <AgentCommandPanel
+                commands={props.commands}
+                sessionID={props.sessionID}
+                width={Math.max(1, commandWidth() - 2)}
+                busyID={props.commandBusyID}
+                theme={theme}
+                onUpdate={props.onUpdateCommand}
+                onOpenSession={props.onOpenSession}
+              />
             </box>
-          )}
-        </For>
-      </box>
+          </Show>
+          <SessionTodoPanel todos={props.todos} width={layout().todoWidth} height={layout().dockHeight} />
+          <Show when={layout().showNotes}>
+            <SessionNotesWidget sessionID={props.sessionID} width={layout().notesWidth} height={layout().dockHeight} />
+          </Show>
+          <Show when={layout().showSubagents}>
+            <SessionSubagentsWidget
+              subagents={props.subagents}
+              width={layout().subagentsWidth}
+              height={layout().dockHeight}
+              onOpen={props.onOpenSubagent}
+            />
+          </Show>
+          <Show when={layout().showInfo}>
+            <SessionInfoWidget info={props.info} width={layout().infoWidth} height={layout().dockHeight} />
+          </Show>
+          <For each={dockWidgets()}>
+            {(item) => (
+              <box
+                flexShrink={0}
+                width={dockWidgetWidth(item)}
+                minWidth={dockWidgetMinWidth(item)}
+                maxWidth={item.maxWidth}
+                height={dockWidgetHeight(item, layout().dockHeight)}
+                overflow="hidden"
+                backgroundColor={theme.backgroundPanel}
+              >
+                <RenderMendWidget item={item} />
+              </box>
+            )}
+          </For>
+        </box>
+      </scrollbox>
     </box>
   )
 }
@@ -5633,6 +5625,7 @@ function SessionNotesWidget(props: { sessionID: string; width: number; height: n
     <box
       width={props.width}
       height={props.height}
+      flexShrink={0}
       paddingLeft={2}
       paddingRight={2}
       paddingTop={1}
@@ -5707,6 +5700,7 @@ function SessionInfoWidget(props: { info: SessionBottomInfo; width: number; heig
     <box
       width={props.width}
       height={props.height}
+      flexShrink={0}
       paddingLeft={2}
       paddingRight={2}
       paddingTop={1}
@@ -5750,6 +5744,7 @@ function SessionSubagentsWidget(props: {
     <box
       width={props.width}
       height={props.height}
+      flexShrink={0}
       paddingLeft={2}
       paddingRight={2}
       paddingTop={1}
@@ -6050,6 +6045,7 @@ function UserMessage(props: {
       kind: metadata?.kind === "peer_response" ? ("response" as const) : ("message" as const),
       sourceTitle: typeof metadata?.sourceTitle === "string" ? metadata.sourceTitle : undefined,
       sourceSessionID: typeof metadata?.sourceSessionID === "string" ? metadata.sourceSessionID : undefined,
+      cancelled: typeof metadata?.cancelledAt === "number",
     }
   })
   const subagentInitialPrompt = createMemo(() => props.compactSubagentPrompt && props.index === 0)
@@ -6223,11 +6219,11 @@ function UserMessage(props: {
             <box flexDirection="row" justifyContent="space-between" width="100%" gap={2}>
               <box flexDirection="row" flexGrow={1} minWidth={0} overflow="hidden">
                 <text fg={theme.textMuted} wrapMode="none">
-                  <span style={{ fg: color() }}>●</span>{" "}
+                  <span style={{ fg: agentMessage() ? theme.accent : color() }}>{agentMessage() ? "◆" : "●"}</span>{" "}
                   {agentMessage()
-                    ? `${agentMessage()!.kind === "response" ? "Reply from" : "From"} · ${
+                    ? `${agentMessage()!.kind === "response" ? "Agent reply from" : "Agent message from"} · ${
                         agentMessage()!.sourceTitle ?? agentMessage()!.sourceSessionID ?? "session"
-                      }`
+                      }${agentMessage()!.cancelled ? " · cancelled" : ""}`
                     : subagentInitialPrompt()
                       ? "Subagent prompt"
                       : "You"}
diff --git a/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.test.ts b/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.test.ts
index f01d38eb..aca1bdd1 100644
--- a/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.test.ts
+++ b/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.test.ts
@@ -1,7 +1,15 @@
 import { describe, expect, test } from "bun:test"
-import { sessionBottomDockLayout, sessionTodoPanelWidth } from "./session-bottom-dock"
+import { sessionBottomDockLayout, sessionTodoPanelWidth, sessionWidgetTrayContentWidth } from "./session-bottom-dock"
 
 describe("session bottom dock layout", () => {
+  test("keeps the tray viewport width when widgets fit", () => {
+    expect(sessionWidgetTrayContentWidth({ dockWidth: 100, widgetWidths: [28, 32] })).toBe(100)
+  })
+
+  test("keeps every stacked widget reachable when the tray overflows", () => {
+    expect(sessionWidgetTrayContentWidth({ dockWidth: 60, widgetWidths: [42, 28, 32] })).toBe(104)
+  })
+
   test("keeps room for notes beside todos when the dock has side space", () => {
     const layout = sessionBottomDockLayout({
       width: 120,
diff --git a/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.ts b/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.ts
index 5afee8ab..15fd2877 100644
--- a/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.ts
+++ b/src/mendcode/packages/opencode/src/cli/cmd/tui/util/session-bottom-dock.ts
@@ -25,6 +25,18 @@ const MAX_SUBAGENTS_WIDGET_WIDTH = 56
 const MAX_INFO_WIDGET_WIDTH = 36
 const SIDE_WIDGET_GAP = 1
 
+export function sessionWidgetTrayContentWidth(input: {
+  dockWidth: number
+  widgetWidths: readonly number[]
+  gap?: number
+}) {
+  const gap = Math.max(0, input.gap ?? SIDE_WIDGET_GAP)
+  const contentWidth = input.widgetWidths
+    .filter((width) => Number.isFinite(width) && width > 0)
+    .reduce((total, width, index) => total + Math.max(1, width) + (index > 0 ? gap : 0), 0)
+  return Math.max(1, input.dockWidth, contentWidth)
+}
+
 function clampDockWidth(value: number, min: number, max: number) {
   return Math.max(min, Math.min(max, value))
 }
diff --git a/src/mendcode/packages/opencode/src/config/keybinds.ts b/src/mendcode/packages/opencode/src/config/keybinds.ts
index 4521929a..c25b74fe 100644
--- a/src/mendcode/packages/opencode/src/config/keybinds.ts
+++ b/src/mendcode/packages/opencode/src/config/keybinds.ts
@@ -72,7 +72,7 @@ const KeybindsSchema = Schema.Struct({
   agent_cycle_reverse: keybind("none", "Previous primary agent"),
   variant_cycle: keybind("f3", "Cycle model variants"),
   variant_list: keybind("shift+f3", "List model variants"),
-  todo_toggle: keybind("ctrl+t", "Toggle session todos"),
+  todo_toggle: keybind("ctrl+t", "Toggle session widgets"),
   input_clear: keybind("ctrl+c", "Clear input field"),
   input_paste: keybind("ctrl+v", "Paste from clipboard"),
   input_submit: keybind("return", "Submit input"),
diff --git a/src/mendcode/packages/opencode/src/mend/cli/control-plane.ts b/src/mendcode/packages/opencode/src/mend/cli/control-plane.ts
index 4be640cc..58432273 100644
--- a/src/mendcode/packages/opencode/src/mend/cli/control-plane.ts
+++ b/src/mendcode/packages/opencode/src/mend/cli/control-plane.ts
@@ -1621,7 +1621,7 @@ async function models(args: string[]) {
   if (sub === "presets") {
     console.log(
       JSON.stringify(
-        { presets: modelPresets, secretsIncluded: false, pricingSource: "official OpenAI docs captured 2026-05-06" },
+        { presets: modelPresets, secretsIncluded: false, pricingSource: "official OpenAI docs verified 2026-09-04" },
         null,
         2,
       ),
diff --git a/src/mendcode/packages/opencode/src/mend/config/models.ts b/src/mendcode/packages/opencode/src/mend/config/models.ts
index f72811c9..337a1b03 100644
--- a/src/mendcode/packages/opencode/src/mend/config/models.ts
+++ b/src/mendcode/packages/opencode/src/mend/config/models.ts
@@ -167,6 +167,15 @@ function withoutRemovedModelRoles(config: ModelsConfig): ModelsConfig {
 }
 
 export const modelPresets = {
+  "openai-codex-subscription-gpt-6-astra": {
+    providerID: "openai",
+    modelID: "gpt-6-astra",
+    authMode: "chatgpt-subscription-oauth",
+    env: [],
+    pricingPer1MTokens: null,
+    source: ".agents/vendor/opencode/packages/opencode/src/plugin/codex.ts",
+    note: "Uses the Codex-supported GPT-6 Astra model with ChatGPT Plus/Pro OAuth when the account rollout includes Astra; not OpenAI API billing.",
+  },
   "openai-codex-subscription-gpt-5.6-sol": {
     providerID: "openai",
     modelID: "gpt-5.6-sol",
@@ -176,6 +185,24 @@ export const modelPresets = {
     source: ".agents/vendor/opencode/packages/opencode/src/plugin/codex.ts",
     note: "Uses the Codex-supported GPT-5.6 Sol model with ChatGPT Plus/Pro OAuth; not OpenAI API billing.",
   },
+  "openai-codex-subscription-gpt-5.6-terra": {
+    providerID: "openai",
+    modelID: "gpt-5.6-terra",
+    authMode: "chatgpt-subscription-oauth",
+    env: [],
+    pricingPer1MTokens: null,
+    source: ".agents/vendor/opencode/packages/opencode/src/plugin/codex.ts",
+    note: "Uses the Codex-supported GPT-5.6 Terra model with ChatGPT Plus/Pro OAuth; not OpenAI API billing.",
+  },
+  "openai-codex-subscription-gpt-5.6-luna": {
+    providerID: "openai",
+    modelID: "gpt-5.6-luna",
+    authMode: "chatgpt-subscription-oauth",
+    env: [],
+    pricingPer1MTokens: null,
+    source: ".agents/vendor/opencode/packages/opencode/src/plugin/codex.ts",
+    note: "Uses the Codex-supported GPT-5.6 Luna model with ChatGPT Plus/Pro OAuth; not OpenAI API billing.",
+  },
   "openai-codex-subscription-gpt-5.2": {
     providerID: "openai",
     modelID: "gpt-5.2",
@@ -217,9 +244,36 @@ export const modelPresets = {
     modelID: "gpt-5.6",
     authMode: "api-key",
     env: ["OPENAI_API_KEY"],
-    pricingPer1MTokens: { inputUsd: 5, cachedInputUsd: 0.5, outputUsd: 30 },
-    source: "https://platform.openai.com/docs/pricing/",
-    note: "OpenAI API billing; newest GPT-5.6 default with max reasoning support.",
+    pricingPer1MTokens: { inputUsd: 4, cachedInputUsd: 0.4, outputUsd: 20 },
+    source: "https://developers.openai.com/api/docs/models/gpt-5.6-sol",
+    note: "OpenAI API billing; GPT-5.6 alias for the Sol tier with max reasoning support.",
+  },
+  "openai-api-gpt-6-astra": {
+    providerID: "openai",
+    modelID: "gpt-6-astra",
+    authMode: "api-key",
+    env: ["OPENAI_API_KEY"],
+    pricingPer1MTokens: { inputUsd: 10, cachedInputUsd: 1, outputUsd: 50 },
+    source: "https://developers.openai.com/api/docs/models/gpt-6-astra",
+    note: "OpenAI API billing; flagship GPT-6 Astra for the hardest reasoning and coding work.",
+  },
+  "openai-api-gpt-5.6-terra": {
+    providerID: "openai",
+    modelID: "gpt-5.6-terra",
+    authMode: "api-key",
+    env: ["OPENAI_API_KEY"],
+    pricingPer1MTokens: { inputUsd: 2, cachedInputUsd: 0.2, outputUsd: 12 },
+    source: "https://developers.openai.com/api/docs/models/gpt-5.6-terra",
+    note: "OpenAI API billing; GPT-5.6 Terra for balanced everyday work.",
+  },
+  "openai-api-gpt-5.6-luna": {
+    providerID: "openai",
+    modelID: "gpt-5.6-luna",
+    authMode: "api-key",
+    env: ["OPENAI_API_KEY"],
+    pricingPer1MTokens: { inputUsd: 0.2, cachedInputUsd: 0.02, outputUsd: 1.2 },
+    source: "https://developers.openai.com/api/docs/models/gpt-5.6-luna",
+    note: "OpenAI API billing; GPT-5.6 Luna for cost-efficient high-volume work.",
   },
   "openai-api-gpt-5-mini": {
     providerID: "openai",
diff --git a/src/mendcode/packages/opencode/src/mend/prompt/compose.ts b/src/mendcode/packages/opencode/src/mend/prompt/compose.ts
index ddb4ba78..de6e56a0 100644
--- a/src/mendcode/packages/opencode/src/mend/prompt/compose.ts
+++ b/src/mendcode/packages/opencode/src/mend/prompt/compose.ts
@@ -241,7 +241,7 @@ function marketplaceExtensionContract() {
     "- Custom Prompt Mode is one of `minimal`, `focus`, `full`, or `custom`; `custom` reads the bounded project prompt at `.mendcode/prompts/custom.md` while preserving the MendCode boundary. It is project instruction text, not an upstream hidden prompt.",
     "- Custom AI tools live in `.mendcode/tools/*.{ts,js}` and can invoke bounded scripts; expose only the smallest safe interface and keep secrets out of tools and packages.",
     "- Custom pages/routes render through the public plugin route/slot API. Custom widgets use `api.ui.runtime.setWidget` with `aboveEditor`, `belowEditor`, or `sessionBottomDock` placement and clean up with the plugin lifecycle.",
-    "- Ctrl+T toggles the current session's TODO view. Ctrl+P -> Customize TUI or `/customize` opens live customization; these controls do not replace the public plugin API.",
+    "- Ctrl+T toggles the current session's horizontal widget tray (todos included). Ctrl+P -> Customize TUI or `/customize` opens live customization; these controls do not replace the public plugin API.",
     "- Custom pages can build terminal-native ASCII/Solid UIs similar to built-in Usage, Memory Center, or Loop pages when the required state is available through the public API.",
     "- Shell-backed widgets use `api.shell.spawn()` for bounded stdout/stderr streams. This is not a PTY; do not implement full-screen terminal apps, cursor-addressing programs, alternate-screen apps, Doom, or real cava by piping stdout into the main TUI.",
     "- If a package needs private MendCode runtime data, add or request a public API first. Do not import private runtime internals from packages.",
diff --git a/src/mendcode/packages/opencode/src/mend/prompt/sources.ts b/src/mendcode/packages/opencode/src/mend/prompt/sources.ts
index a82218ea..1cbd354f 100644
--- a/src/mendcode/packages/opencode/src/mend/prompt/sources.ts
+++ b/src/mendcode/packages/opencode/src/mend/prompt/sources.ts
@@ -33,6 +33,17 @@ const modelBehaviorProfiles: Array<PromptBehaviorProfile & { match: RegExp }> =
       "Treat aliases, transport, reasoning settings, caching, and advanced features as runtime configuration; use only capabilities exposed by this session.",
     ],
   },
+  {
+    id: "gpt-6-astra",
+    focusID: "codex",
+    label: "GPT-6 Astra compatibility",
+    sourcePolicy: "mendcode-compatibility",
+    match: /(^|[^a-z0-9])gpt[-_.:/]?6[-_.:/]?astra([^a-z0-9]|$)/i,
+    behavior: [
+      "Use the actual GPT-6 Astra runtime contract and exposed tools instead of assuming that every Codex CLI feature is available through MendCode.",
+      "Treat reasoning modes, transport, caching, context limits, and advanced features as runtime configuration; verify capabilities before promising them.",
+    ],
+  },
   {
     id: "claude-sonnet-5",
     focusID: "claude",
diff --git a/src/mendcode/packages/opencode/src/plugin/codex.ts b/src/mendcode/packages/opencode/src/plugin/codex.ts
index 954a6398..748a0d55 100644
--- a/src/mendcode/packages/opencode/src/plugin/codex.ts
+++ b/src/mendcode/packages/opencode/src/plugin/codex.ts
@@ -27,6 +27,7 @@ const ALLOWED_MODELS = new Set([
   "gpt-5.3-codex-spark",
   "gpt-5.4",
   "gpt-5.4-mini",
+  "gpt-6-astra",
 ])
 
 const CODEX_CHATGPT_MODEL_ALIASES: Record<string, string> = {
@@ -40,6 +41,12 @@ const CODEX_CHATGPT_5_6_LIMIT = {
   output: 128_000,
 }
 
+const CODEX_CHATGPT_6_ASTRA_LIMIT = {
+  context: 1_050_000,
+  input: 922_000,
+  output: 128_000,
+}
+
 const CODEX_CHATGPT_FAST_MODE_MODELS = new Set([
   "gpt-5.4",
   "gpt-5.4-mini",
@@ -48,9 +55,15 @@ const CODEX_CHATGPT_FAST_MODE_MODELS = new Set([
   "gpt-5.6-sol",
   "gpt-5.6-terra",
   "gpt-5.6-luna",
+  "gpt-6-astra",
 ])
 
-const CODEX_CHATGPT_PRO_MODE_MODELS = new Set(["gpt-5.6", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"])
+const CODEX_CHATGPT_PRO_MODE_MODELS = new Set([
+  "gpt-5.6",
+  "gpt-5.6-sol",
+  "gpt-5.6-terra",
+  "gpt-5.6-luna",
+])
 
 export function normalizeCodexChatGPTModel(modelID: string) {
   const fastBase = modelID.endsWith("-fast") ? modelID.slice(0, -"-fast".length) : undefined
@@ -68,6 +81,20 @@ export function normalizeCodexChatGPTModel(modelID: string) {
   }
 }
 
+export function isCodexChatGPTModelSupported(modelID: string) {
+  const normalized = normalizeCodexChatGPTModel(modelID).modelID
+  if (ALLOWED_MODELS.has(normalized)) return true
+  const match = normalized.match(/^gpt-(\d+\.\d+)/)
+  return match ? parseFloat(match[1]) > 5.4 : false
+}
+
+function codexChatGPTLimit(modelID: string) {
+  const normalized = normalizeCodexChatGPTModel(modelID).modelID
+  if (normalized === "gpt-6-astra") return CODEX_CHATGPT_6_ASTRA_LIMIT
+  if (normalized.startsWith("gpt-5.6")) return CODEX_CHATGPT_5_6_LIMIT
+  return undefined
+}
+
 export function normalizeCodexChatGPTRequestBody(body: BodyInit | null | undefined) {
   if (typeof body !== "string") return body
   let parsed: unknown
@@ -569,12 +596,12 @@ export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPlug
         return Object.fromEntries(
           Object.entries(provider.models)
             .filter(([, model]) => {
-              if (ALLOWED_MODELS.has(model.api.id)) return true
-              const match = model.api.id.match(/^gpt-(\d+\.\d+)/)
-              return match ? parseFloat(match[1]) > 5.4 : false
+              return isCodexChatGPTModelSupported(model.api.id)
             })
             .map(([modelID, model]) => {
               const modelOptions = isRecord(model.options) ? model.options : {}
+              const limit = codexChatGPTLimit(model.api.id)
+              const usesCompactionThreshold = limit !== undefined
               return [
                 modelID,
                 {
@@ -584,8 +611,8 @@ export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPlug
                     output: 0,
                     cache: { read: 0, write: 0 },
                   },
-                  limit: model.api.id.startsWith("gpt-5.6")
-                    ? CODEX_CHATGPT_5_6_LIMIT
+                  limit: limit
+                    ? limit
                     : model.id.includes("gpt-5.5")
                       ? {
                           context: 400_000,
@@ -593,7 +620,7 @@ export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPlug
                           output: 128_000,
                         }
                       : model.limit,
-                  options: model.api.id.startsWith("gpt-5.6")
+                  options: usesCompactionThreshold
                     ? {
                         ...modelOptions,
                         compaction: {
diff --git a/src/mendcode/packages/opencode/src/provider/models.ts b/src/mendcode/packages/opencode/src/provider/models.ts
index 3b826f98..26ad753b 100644
--- a/src/mendcode/packages/opencode/src/provider/models.ts
+++ b/src/mendcode/packages/opencode/src/provider/models.ts
@@ -8,6 +8,7 @@ import { Flock } from "@mendcode/core/util/flock"
 import { Hash } from "@mendcode/core/util/hash"
 import { AppFileSystem } from "@mendcode/core/filesystem"
 import { withTransientReadRetry } from "@/util/effect-http-client"
+import { withOpenAIModelFallbacks } from "./openai-models"
 
 const Cost = Schema.Struct({
   input: Schema.Finite,
@@ -166,7 +167,7 @@ export const layer: Layer.Layer<Service, never, AppFileSystem.Service | HttpClie
         }),
       )
       return JSON.parse(text) as Record<string, Provider>
-    }).pipe(Effect.withSpan("ModelsDev.populate"), Effect.orDie)
+    }).pipe(Effect.map(withOpenAIModelFallbacks), Effect.withSpan("ModelsDev.populate"), Effect.orDie)
 
     const [cachedGet, invalidate] = yield* Effect.cachedInvalidateWithTTL(populate, Duration.infinity)
 
diff --git a/src/mendcode/packages/opencode/src/provider/openai-models.ts b/src/mendcode/packages/opencode/src/provider/openai-models.ts
new file mode 100644
index 00000000..f806aa57
--- /dev/null
+++ b/src/mendcode/packages/opencode/src/provider/openai-models.ts
@@ -0,0 +1,201 @@
+import type * as ModelsDev from "./models"
+
+const GPT_56_REASONING = ["none", "low", "medium", "high", "xhigh", "max"]
+const GPT_6_ASTRA_REASONING = ["low", "medium", "high", "xhigh", "max"]
+
+const GPT_56_LIMIT = {
+  context: 1_050_000,
+  input: 922_000,
+  output: 128_000,
+}
+
+const GPT_56_MODALITIES = {
+  input: ["text", "image", "pdf"] as Array<"text" | "image" | "pdf">,
+  output: ["text"] as Array<"text">,
+}
+
+const GPT_56_BASE = {
+  attachment: true,
+  reasoning: true,
+  reasoning_options: [{ type: "effort", values: GPT_56_REASONING }],
+  temperature: false,
+  tool_call: true,
+  limit: GPT_56_LIMIT,
+  modalities: GPT_56_MODALITIES,
+  release_date: "2026-07-09",
+}
+
+const GPT_56_PRO_MODE = {
+  provider: { body: { reasoning: { mode: "pro" } } },
+}
+
+function gpt56Modes(fastCost: ModelsDev.Model["cost"]) {
+  return {
+    fast: {
+      cost: fastCost,
+      provider: { body: { service_tier: "priority" } },
+    },
+    pro: GPT_56_PRO_MODE,
+  }
+}
+
+const GPT_56_COSTS = {
+  sol: {
+    input: 4,
+    output: 20,
+    cache_read: 0.4,
+    cache_write: 5,
+    context_over_200k: {
+      input: 8,
+      output: 30,
+      cache_read: 0.8,
+      cache_write: 10,
+    },
+  },
+  terra: {
+    input: 2,
+    output: 12,
+    cache_read: 0.2,
+    cache_write: 2.5,
+    context_over_200k: {
+      input: 4,
+      output: 18,
+      cache_read: 0.4,
+      cache_write: 5,
+    },
+  },
+  luna: {
+    input: 0.2,
+    output: 1.2,
+    cache_read: 0.02,
+    cache_write: 0.25,
+    context_over_200k: {
+      input: 0.4,
+      output: 1.8,
+      cache_read: 0.04,
+      cache_write: 0.5,
+    },
+  },
+}
+
+const GPT_56_FAST_COSTS = {
+  sol: { input: 8, output: 40, cache_read: 0.8, cache_write: 10 },
+  terra: { input: 4, output: 24, cache_read: 0.4, cache_write: 5 },
+  luna: { input: 0.4, output: 2.4, cache_read: 0.04, cache_write: 0.5 },
+}
+
+const GPT_6_ASTRA_COST = {
+  input: 10,
+  output: 50,
+  cache_read: 1,
+  cache_write: 12.5,
+  context_over_200k: {
+    input: 20,
+    output: 75,
+    cache_read: 2,
+    cache_write: 25,
+  },
+}
+
+const GPT_6_ASTRA_FAST_COST = {
+  input: 20,
+  output: 100,
+  cache_read: 2,
+  cache_write: 25,
+  context_over_200k: {
+    input: 40,
+    output: 150,
+    cache_read: 4,
+    cache_write: 50,
+  },
+}
+
+const GPT_6_ASTRA_MODES = {
+  fast: {
+    cost: GPT_6_ASTRA_FAST_COST,
+    provider: { body: { service_tier: "priority" } },
+  },
+}
+
+function gpt56Model(
+  id: string,
+  name: string,
+  family: string,
+  cost: ModelsDev.Model["cost"],
+  fastCost: ModelsDev.Model["cost"],
+): ModelsDev.Model {
+  return {
+    id,
+    name,
+    family,
+    ...GPT_56_BASE,
+    cost,
+    experimental: { modes: gpt56Modes(fastCost) },
+  }
+}
+
+/**
+ * Keep the picker usable when a cached or bundled models.dev catalog predates
+ * the current OpenAI model family. Live catalog entries always win.
+ */
+export const OPENAI_MODEL_FALLBACKS: Record<string, ModelsDev.Model> = {
+  "gpt-5.6": gpt56Model("gpt-5.6", "GPT-5.6", "gpt-sol", GPT_56_COSTS.sol, GPT_56_FAST_COSTS.sol),
+  "gpt-5.6-sol": gpt56Model(
+    "gpt-5.6-sol",
+    "GPT-5.6 Sol",
+    "gpt-sol",
+    GPT_56_COSTS.sol,
+    GPT_56_FAST_COSTS.sol,
+  ),
+  "gpt-5.6-terra": gpt56Model(
+    "gpt-5.6-terra",
+    "GPT-5.6 Terra",
+    "gpt-terra",
+    GPT_56_COSTS.terra,
+    GPT_56_FAST_COSTS.terra,
+  ),
+  "gpt-5.6-luna": gpt56Model(
+    "gpt-5.6-luna",
+    "GPT-5.6 Luna",
+    "gpt-luna",
+    GPT_56_COSTS.luna,
+    GPT_56_FAST_COSTS.luna,
+  ),
+  "gpt-6-astra": {
+    id: "gpt-6-astra",
+    name: "GPT-6 Astra",
+    family: "gpt-astra",
+    attachment: true,
+    reasoning: true,
+    reasoning_options: [{ type: "effort", values: GPT_6_ASTRA_REASONING }],
+    temperature: false,
+    tool_call: true,
+    limit: {
+      context: 1_050_000,
+      input: 922_000,
+      output: 128_000,
+    },
+    modalities: GPT_56_MODALITIES,
+    // OpenAI publishes a knowledge cutoff for Astra, but not a model release
+    // date on the model page. Keep this field empty until the catalog provides
+    // an authoritative value.
+    release_date: "",
+    cost: GPT_6_ASTRA_COST,
+    experimental: { modes: GPT_6_ASTRA_MODES },
+  },
+}
+
+export function withOpenAIModelFallbacks(catalog: Record<string, ModelsDev.Provider>) {
+  const provider = catalog.openai
+  if (!provider) return catalog
+  return {
+    ...catalog,
+    openai: {
+      ...provider,
+      models: {
+        ...OPENAI_MODEL_FALLBACKS,
+        ...provider.models,
+      },
+    },
+  }
+}
diff --git a/src/mendcode/packages/opencode/src/session/prompt.ts b/src/mendcode/packages/opencode/src/session/prompt.ts
index 9eb6b614..b7249a62 100644
--- a/src/mendcode/packages/opencode/src/session/prompt.ts
+++ b/src/mendcode/packages/opencode/src/session/prompt.ts
@@ -153,6 +153,7 @@ export function interruptedToolPromptText(messages: readonly MessageV2.WithParts
 }
 
 type RecoverablePromptMessage = {
+  parts?: readonly { type: string; metadata?: Record<string, unknown> }[]
   info: {
     id: string
     role: "user" | "assistant"
@@ -162,11 +163,24 @@ type RecoverablePromptMessage = {
   }
 }
 
+function isCancelledPeerPrompt(message: Pick<RecoverablePromptMessage, "parts">) {
+  return (
+    message.parts?.some(
+      (part) =>
+        part.type === "text" &&
+        (part.metadata?.kind === "peer_message" || part.metadata?.kind === "peer_response") &&
+        typeof part.metadata.cancelledAt === "number",
+    ) === true
+  )
+}
+
 export function findRecoverableQueuedPrompt<T extends RecoverablePromptMessage>(
   messages: readonly T[],
   options?: { includeLegacy?: boolean },
 ) {
-  const ordered = messages.toSorted((a, b) => a.info.time.created - b.info.time.created)
+  const ordered = messages
+    .filter((message) => !isCancelledPeerPrompt(message))
+    .toSorted((a, b) => a.info.time.created - b.info.time.created)
   const latestAssistant = ordered.findLast((message) => message.info.role === "assistant")
   if (latestAssistant && latestAssistant.info.time.completed === undefined) return
 
@@ -568,6 +582,44 @@ export const layer = Layer.effect(
       yield* backgroundTasks.dismissNotifications(sessionID)
     })
 
+    const cancelPeerExchanges = Effect.fn("SessionPrompt.cancelPeerExchanges")(function* (sessionID: SessionID) {
+      const messages = yield* sessions.messages({ sessionID, view: "full" })
+      const answered = new Set(
+        messages.flatMap((message) => (message.info.role === "assistant" ? [message.info.parentID] : [])),
+      )
+      for (const message of messages) {
+        if (message.info.role !== "user" || message.info.queued !== true || answered.has(message.info.id)) continue
+        const parts = message.parts.filter(
+          (part): part is MessageV2.TextPart =>
+            part.type === "text" &&
+            (part.metadata?.kind === "peer_message" || part.metadata?.kind === "peer_response"),
+        )
+        if (parts.length === 0) continue
+        for (const part of parts) {
+          yield* sessions.updatePart({ ...part, ignored: true, metadata: { ...part.metadata, cancelledAt: Date.now() } })
+        }
+        yield* sessions.updateMessage({ ...message.info, queued: false })
+        yield* state.cancelQueued(sessionID, message.info.id)
+      }
+      const commands = (yield* agentCommands.list()).filter(
+        (command) =>
+          command.type === "peer_message" &&
+          (command.targetSessionID === sessionID || command.sourceSessionID === sessionID) &&
+          (command.state === "pending" || command.state === "accepted" || command.state === "running"),
+      )
+      // Settle the durable exchange before idle is published, including replies
+      // still being produced elsewhere. Recovery must not replay a stopped run.
+      for (const command of commands) {
+        yield* agentCommands
+          .update({
+            id: command.id,
+            state: command.state === "pending" ? "rejected" : "failed",
+            error: "Session stopped by the user before the message exchange completed.",
+          })
+          .pipe(Effect.catchTag("AgentCommandInvalidStateTransitionError", () => Effect.void), Effect.orDie)
+      }
+    })
+
     const finishOrphanedAssistant = Effect.fn("SessionPrompt.finishOrphanedAssistant")(function* (
       sessionID: SessionID,
       messageID: MessageID,
@@ -622,6 +674,7 @@ export const layer = Layer.effect(
       // A background task can finish after the parent run is cancelled. Do not
       // let its late notification start a fresh assistant turn.
       yield* cancelOwnerWakeSession(sessionID)
+      yield* cancelPeerExchanges(sessionID)
       promptAbortReasons.set(sessionID, "user")
       const controller = promptAbortControllers.get(sessionID)?.controller
       let interruptedAssistantID: MessageID | undefined
@@ -654,6 +707,9 @@ export const layer = Layer.effect(
       // the cancel request reached the runner.
       yield* cancelOwnerWakeSession(input.sessionID)
       const active = promptAbortControllers.get(input.sessionID)
+      if (active?.targetMessageID === input.targetMessageID || (!active && !(yield* state.isBusy(input.sessionID)))) {
+        yield* cancelPeerExchanges(input.sessionID)
+      }
       let interruptedAssistantID: MessageID | undefined
       let markedUserAbort = false
       const result = yield* state
@@ -2439,6 +2495,9 @@ NOTE: At any point in time through this workflow you should feel free to ask the
         yield* slog.info("loop", { step })
 
         let msgs = yield* MessageV2.filterCompactedEffect(sessionID)
+        const cancelled = msgs.find((message) => message.info.id === targetMessageID && isCancelledPeerPrompt(message))
+        if (cancelled) return cancelled
+        msgs = msgs.filter((message) => !isCancelledPeerPrompt(message))
 
         if (!initialMessageIDs) {
           const targetIndex = targetMessageID ? msgs.findIndex((message) => message.info.id === targetMessageID) : -1
@@ -3159,6 +3218,8 @@ NOTE: At any point in time through this workflow you should feel free to ask the
               return metadata?.kind === "peer_response" && metadata.deliveryID === command.id
             }),
           )
+          const current = yield* agentCommands.get(command.id)
+          if (current.state !== "accepted" && current.state !== "running") return
           const responsePrompt = Option.isSome(existingResponse)
             ? existingResponse.value
             : yield* promptAsync({
@@ -3260,6 +3321,8 @@ NOTE: At any point in time through this workflow you should feel free to ask the
             return
           }
 
+          // A queued delivery may have been cancelled while its marker was read.
+          if ((yield* agentCommands.get(info.id)).state !== "running") return
           const message = yield* promptAsync({
             sessionID: info.targetSessionID,
             parts: [
diff --git a/src/mendcode/packages/opencode/test/cli/tui/agent-command-panel.test.tsx b/src/mendcode/packages/opencode/test/cli/tui/agent-command-panel.test.tsx
new file mode 100644
index 00000000..d9dd0b28
--- /dev/null
+++ b/src/mendcode/packages/opencode/test/cli/tui/agent-command-panel.test.tsx
@@ -0,0 +1,140 @@
+/** @jsxImportSource @opentui/solid */
+import { expect, test } from "bun:test"
+import { RGBA } from "@opentui/core"
+import { testRender } from "@opentui/solid"
+import { createSignal } from "solid-js"
+import { AgentCommandPanel } from "@/cli/cmd/tui/component/agent-command-panel"
+import type { AgentViewCommand } from "@/cli/cmd/tui/util/agent-view"
+
+const theme = {
+  text: RGBA.fromHex("#f0f0f0"),
+  textMuted: RGBA.fromHex("#999999"),
+  accent: RGBA.fromHex("#66ccff"),
+  error: RGBA.fromHex("#ff6666"),
+}
+
+function command(id: string, state: AgentViewCommand["state"]): AgentViewCommand {
+  return {
+    id,
+    state,
+    type: "peer_message",
+    sourceSessionID: "source",
+    targetSessionID: "target",
+    payload: { sourceTitle: "Cache worker", text: "Delayed progress" },
+    time: { created: 1, updated: 1 },
+  }
+}
+
+async function settle(app: Awaited<ReturnType<typeof testRender>>) {
+  await app.renderOnce()
+  await app.renderOnce()
+}
+
+test("collapses historical commands to one row and exposes failure details on demand", async () => {
+  const app = await testRender(
+    () => (
+      <AgentCommandPanel
+        sessionID="target"
+        commands={[
+          { ...command("failed", "failed"), error: "Session stopped by the user." },
+          command("expired", "expired"),
+          command("completed", "completed"),
+        ]}
+        width={90}
+        theme={theme}
+        onUpdate={() => {
+          throw new Error("Historical command must not be actionable")
+        }}
+      />
+    ),
+    { width: 90, height: 12 },
+  )
+  try {
+    await settle(app)
+    expect(app.captureCharFrame().trim()).toBe("▸ Commands · 3 past")
+    const toggle = app.renderer.root.findDescendantById("agent-command-history-toggle")!
+    await app.mockMouse.click(toggle.x + 1, toggle.y)
+    await settle(app)
+    const frame = app.captureCharFrame()
+    expect(frame).toContain("▾ Commands · 3 past")
+    expect(frame).toContain("Cache worker")
+    expect(frame).toContain("Session stopped by the user.")
+    expect(frame.match(/failed/g)).toHaveLength(1)
+    expect(frame).not.toContain("[accept]")
+    await app.mockMouse.click(toggle.x + 1, toggle.y)
+    await settle(app)
+    expect(app.captureCharFrame().trim()).toBe("▸ Commands · 3 past")
+  } finally {
+    app.renderer.destroy()
+  }
+})
+
+test("offers a direct button to open the related session", async () => {
+  const opened: string[] = []
+  const updates: string[] = []
+  const app = await testRender(
+    () => (
+      <AgentCommandPanel
+        sessionID="target"
+        commands={[command("pending", "pending")]}
+        width={60}
+        height={7}
+        theme={theme}
+        onUpdate={(item, state) => updates.push(`${item.id}:${state}`)}
+        onOpenSession={(sessionID) => opened.push(sessionID)}
+      />
+    ),
+    { width: 60, height: 8 },
+  )
+  try {
+    await settle(app)
+    const open = app.renderer.root.findDescendantById("agent-command-open-pending")!
+    const accept = app.renderer.root.findDescendantById("agent-command-accept-pending")!
+    const reject = app.renderer.root.findDescendantById("agent-command-reject-pending")!
+    expect(open.x + open.width).toBeLessThanOrEqual(60)
+    expect(accept.x + accept.width).toBeLessThanOrEqual(60)
+    expect(reject.x + reject.width).toBeLessThanOrEqual(60)
+    await app.mockMouse.click(open.x + 1, open.y)
+    expect(opened).toEqual(["source"])
+    await app.mockMouse.click(accept.x + 1, accept.y)
+    expect(updates).toEqual(["pending:accepted"])
+  } finally {
+    app.renderer.destroy()
+  }
+})
+
+for (const width of [42, 100]) {
+  test(`keeps pending actions visible at ${width} columns and reacts to completion`, async () => {
+    const [commands, setCommands] = createSignal([command("pending", "pending"), command("old", "failed")])
+    const updates: string[] = []
+    const app = await testRender(
+      () => (
+        <AgentCommandPanel
+          commands={commands()}
+          sessionID="target"
+          width={width}
+          theme={theme}
+          onUpdate={(item, state) => {
+            updates.push(`${item.id}:${state}`)
+            setCommands((items) =>
+              items.map((entry) => (entry.id === item.id ? { ...entry, state: "completed" } : entry)),
+            )
+          }}
+        />
+      ),
+      { width, height: 8 },
+    )
+    try {
+      await settle(app)
+      expect(app.captureCharFrame()).toContain("[accept] [reject]")
+      const accept = app.renderer.root.findDescendantById("agent-command-accept-pending")!
+      expect(accept.x + accept.width).toBeLessThanOrEqual(width)
+      await app.mockMouse.click(accept.x + 1, accept.y)
+      await settle(app)
+      expect(updates).toEqual(["pending:accepted"])
+      expect(app.captureCharFrame().trim()).toBe("▸ Commands · 2 past")
+    } finally {
+      app.renderer.destroy()
+    }
+  })
+}
diff --git a/src/mendcode/packages/opencode/test/mend/models-config.test.ts b/src/mendcode/packages/opencode/test/mend/models-config.test.ts
index 5c8ddacd..562f7a1d 100644
--- a/src/mendcode/packages/opencode/test/mend/models-config.test.ts
+++ b/src/mendcode/packages/opencode/test/mend/models-config.test.ts
@@ -5,6 +5,7 @@ import { tmpdir } from "../fixture/fixture"
 import { initProject } from "../../src/mend/config/project"
 import {
   modelRoleProjection,
+  modelPresets,
   parsePromptModel,
   refreshGeneratedRuntimeModelConfig,
   resolveEffectivePromptSelection,
@@ -17,6 +18,22 @@ async function writeText(file: string, value: string) {
 }
 
 describe("mend model roles", () => {
+  test("publishes the current OpenAI model family and pricing presets", () => {
+    expect(modelPresets["openai-codex-subscription-gpt-6-astra"].modelID).toBe("gpt-6-astra")
+    expect(modelPresets["openai-api-gpt-6-astra"].pricingPer1MTokens).toEqual({
+      inputUsd: 10,
+      cachedInputUsd: 1,
+      outputUsd: 50,
+    })
+    expect(modelPresets["openai-api-gpt-5.6"].pricingPer1MTokens).toEqual({
+      inputUsd: 4,
+      cachedInputUsd: 0.4,
+      outputUsd: 20,
+    })
+    expect(modelPresets["openai-api-gpt-5.6-terra"].modelID).toBe("gpt-5.6-terra")
+    expect(modelPresets["openai-api-gpt-5.6-luna"].modelID).toBe("gpt-5.6-luna")
+  })
+
   test("does not seed command packs that duplicate native TUI model/provider surfaces", async () => {
     await using dir = await tmpdir()
 
diff --git a/src/mendcode/packages/opencode/test/mend/tui-prompt-chrome.test.ts b/src/mendcode/packages/opencode/test/mend/tui-prompt-chrome.test.ts
index 2e82c256..8a8e4c9a 100644
--- a/src/mendcode/packages/opencode/test/mend/tui-prompt-chrome.test.ts
+++ b/src/mendcode/packages/opencode/test/mend/tui-prompt-chrome.test.ts
@@ -355,7 +355,7 @@ describe("mend tui prompt chrome", () => {
     expect(promptChromeUsesFullSessionWidth("ascii-box")).toBe(true)
   })
 
-  test("ctrl-t is reserved for session todos and f3 cycles model variants", () => {
+  test("ctrl-t toggles session widgets and f3 cycles model variants", () => {
     const keybinds = ConfigKeybinds.Keybinds.parse({})
 
     expect(keybinds.todo_toggle).toBe("ctrl+t")
diff --git a/src/mendcode/packages/opencode/test/plugin/codex.test.ts b/src/mendcode/packages/opencode/test/plugin/codex.test.ts
index 35f33128..9a51490b 100644
--- a/src/mendcode/packages/opencode/test/plugin/codex.test.ts
+++ b/src/mendcode/packages/opencode/test/plugin/codex.test.ts
@@ -5,6 +5,7 @@ import {
   extractAccountIdFromClaims,
   extractAccountId,
   normalizeCodexChatGPTRequestBody,
+  isCodexChatGPTModelSupported,
   prepareCodexChatGPTOAuthRequest,
   type IdTokenClaims,
 } from "../../src/plugin/codex"
@@ -73,6 +74,19 @@ describe("plugin.codex", () => {
           ) as string,
         ),
       ).toEqual({ model: "gpt-5.6-terra", reasoning: { effort: "high", mode: "pro" } })
+      expect(
+        JSON.parse(
+          normalizeCodexChatGPTRequestBody(
+            JSON.stringify({ model: "gpt-5.6-terra-pro", reasoning: { effort: "high" } }),
+          ) as string,
+        ),
+      ).toEqual({ model: "gpt-5.6-terra", reasoning: { effort: "high", mode: "pro" } })
+    })
+
+    test("routes Astra fast mode to the priority service tier", () => {
+      expect(
+        JSON.parse(normalizeCodexChatGPTRequestBody(JSON.stringify({ model: "gpt-6-astra-fast" })) as string),
+      ).toEqual({ model: "gpt-6-astra", service_tier: "priority" })
     })
 
     test("preserves explicit GPT-5.6 tiers and non-JSON bodies", () => {
@@ -89,8 +103,10 @@ describe("plugin.codex", () => {
     })
   })
 
-  test("keeps Sol, Terra, and Luna modes in the ChatGPT OAuth catalog", async () => {
+  test("keeps Astra, Sol, Terra, and Luna modes in the ChatGPT OAuth catalog", async () => {
     const ids = [
+      "gpt-6-astra",
+      "gpt-6-astra-fast",
       "gpt-5.6-sol",
       "gpt-5.6-sol-fast",
       "gpt-5.6-terra",
@@ -104,7 +120,7 @@ describe("plugin.codex", () => {
           id,
           {
             id,
-            api: { id: id.replace(/-fast$/, "") },
+            api: { id: id.replace(/-(fast|pro)$/, "") },
             cost: { input: 1, output: 1, cache: { read: 1, write: 1 } },
             limit: { context: 1, input: 1, output: 1 },
           },
@@ -116,11 +132,22 @@ describe("plugin.codex", () => {
 
     expect(Object.keys(models)).toEqual(ids)
     for (const id of ids) {
-      expect(models[id]?.limit).toEqual({ context: 256_000, input: 256_000, output: 128_000 })
+      expect(models[id]?.limit).toEqual(
+        id.startsWith("gpt-6-astra")
+          ? { context: 1_050_000, input: 922_000, output: 128_000 }
+          : { context: 256_000, input: 256_000, output: 128_000 },
+      )
       expect(models[id]?.options).toMatchObject({ compaction: { threshold: 90 } })
     }
   })
 
+  test("accepts the official Astra model and rejects unknown GPT-6 IDs", () => {
+    expect(isCodexChatGPTModelSupported("gpt-6-astra")).toBe(true)
+    expect(isCodexChatGPTModelSupported("gpt-6-astra-fast")).toBe(true)
+    expect(isCodexChatGPTModelSupported("gpt-6-astra-pro")).toBe(false)
+    expect(isCodexChatGPTModelSupported("gpt-6-unknown")).toBe(false)
+  })
+
   test("rewrites GPT-5.6 OAuth requests for Responses Lite with honest client identity", async () => {
     const requests: Array<{ headers: Headers; body: Record<string, unknown> }> = []
     using server = Bun.serve({
diff --git a/src/mendcode/packages/opencode/test/provider/provider.test.ts b/src/mendcode/packages/opencode/test/provider/provider.test.ts
index a6c9a05e..391a5811 100644
--- a/src/mendcode/packages/opencode/test/provider/provider.test.ts
+++ b/src/mendcode/packages/opencode/test/provider/provider.test.ts
@@ -8,6 +8,7 @@ import { Instance } from "../../src/project/instance"
 import { WithInstance } from "../../src/project/with-instance"
 import { Plugin } from "../../src/plugin/index"
 import { ModelsDev } from "@/provider/models"
+import { OPENAI_MODEL_FALLBACKS, withOpenAIModelFallbacks } from "@/provider/openai-models"
 import { Provider } from "@/provider/provider"
 import { ProviderID, ModelID } from "../../src/provider/schema"
 import { Filesystem } from "@/util/filesystem"
@@ -1984,6 +1985,34 @@ test("mode cost preserves over-200k pricing from base model", () => {
   })
 })
 
+test("OpenAI catalog fallback keeps current GPT-5.6 and Astra models available", () => {
+  const catalog = withOpenAIModelFallbacks({
+    openai: {
+      id: "openai",
+      name: "OpenAI",
+      env: ["OPENAI_API_KEY"],
+      models: {
+        "gpt-5.6": {
+          ...OPENAI_MODEL_FALLBACKS["gpt-5.6"],
+          name: "cached GPT-5.6",
+        },
+      },
+    },
+  } as unknown as Record<string, ModelsDev.Provider>)
+
+  expect(Object.keys(catalog.openai!.models)).toEqual(
+    expect.arrayContaining(["gpt-6-astra", "gpt-5.6", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"]),
+  )
+  expect(catalog.openai!.models["gpt-6-astra"]?.limit).toEqual({
+    context: 1_050_000,
+    input: 922_000,
+    output: 128_000,
+  })
+  expect(catalog.openai!.models["gpt-5.6"]?.name).toBe("cached GPT-5.6")
+  expect(catalog.openai!.models["gpt-5.6-luna"]?.experimental?.modes?.fast?.cost?.output).toBe(2.4)
+  expect(catalog.openai!.models["gpt-6-astra"]?.experimental?.modes?.fast?.cost?.output).toBe(100)
+})
+
 test("models.dev normalization fills required response fields", () => {
   const provider = {
     id: "gateway",
diff --git a/src/mendcode/packages/opencode/test/session/prompt.test.ts b/src/mendcode/packages/opencode/test/session/prompt.test.ts
index bfbbacef..6fbd65a8 100644
--- a/src/mendcode/packages/opencode/test/session/prompt.test.ts
+++ b/src/mendcode/packages/opencode/test/session/prompt.test.ts
@@ -2487,6 +2487,145 @@ it.live("automatically delivers a same-workspace agent message and returns its r
   ),
 )
 
+for (const cancellation of ["session", "turn"] as const) {
+  it.live(`cancels queued peer messages when stopping a ${cancellation}`, () =>
+    provideTmpdirServer(
+      Effect.fnUntraced(function* ({ llm }) {
+        const prompt = yield* SessionPrompt.Service
+        const sessions = yield* Session.Service
+        const status = yield* SessionStatus.Service
+        const commands = yield* AgentCommand.Service
+        const source = yield* sessions.create({ title: "Late subagent" })
+        const target = yield* sessions.create({ title: "Stopped parent" })
+        yield* llm.text("ready")
+        const finished = yield* prompt.prompt({
+          sessionID: target.id,
+          agent: "build",
+          model: ref,
+          parts: [{ type: "text", text: "start" }],
+        })
+        if (finished.info.role !== "assistant") throw new Error("assistant missing")
+        yield* status.set(target.id, { type: "busy", message: "active run" })
+        const command = yield* commands.create({
+          sourceSessionID: source.id,
+          targetSessionID: target.id,
+          type: "peer_message",
+          payload: { text: "delayed progress" },
+        })
+        if (cancellation === "session") yield* prompt.cancel(target.id)
+        else yield* prompt.cancelTurn({ sessionID: target.id, targetMessageID: finished.info.parentID })
+        expect((yield* commands.get(command.id)).state).toBe("failed")
+        yield* status.set(target.id, { type: "idle" })
+        yield* prompt.wakePeerDelivery(target.id)
+        yield* Effect.sleep("600 millis")
+        expect(yield* llm.calls).toBe(1)
+        expect(
+          (yield* sessions.messages({ sessionID: target.id, view: "full" })).some((message) =>
+            message.parts.some((part) => part.type === "text" && part.metadata?.deliveryID === command.id),
+          ),
+        ).toBe(false)
+      }),
+      { git: true, config: (url) => ({ ...providerCfg(url), agent: { build: { model: "test/test-model" } } }) },
+    ),
+  )
+}
+
+it.live("does not wake a stopped sender when a peer response arrives", () =>
+  provideTmpdirServer(
+    Effect.fnUntraced(function* ({ llm }) {
+      const prompt = yield* SessionPrompt.Service
+      const sessions = yield* Session.Service
+      const commands = yield* AgentCommand.Service
+      const source = yield* sessions.create({ title: "Stopped sender" })
+      const target = yield* sessions.create({ title: "Responding peer" })
+      yield* prompt.wakePeerDelivery(source.id)
+      const gate = defer<void>()
+      yield* llm.hold("late response", gate.promise)
+      const command = yield* commands.create({
+        sourceSessionID: source.id,
+        targetSessionID: target.id,
+        type: "peer_message",
+        payload: { text: "question" },
+      })
+      try {
+        yield* Effect.gen(function* () {
+          while ((yield* llm.calls) === 0) yield* Effect.sleep("1 millis")
+        }).pipe(Effect.timeout("2 seconds"))
+        yield* prompt.cancel(source.id)
+        expect((yield* commands.get(command.id)).state).toBe("failed")
+      } finally {
+        gate.resolve()
+      }
+      yield* Effect.gen(function* () {
+        while (
+          !(yield* sessions.messages({ sessionID: target.id })).some(
+            (message) => message.info.role === "assistant" && message.info.time.completed !== undefined,
+          )
+        ) yield* Effect.sleep("1 millis")
+      }).pipe(Effect.timeout("2 seconds"))
+      yield* Effect.sleep("100 millis")
+      expect(yield* llm.calls).toBe(1)
+      expect(yield* sessions.messages({ sessionID: source.id, view: "full" })).toEqual([])
+    }),
+    { git: true, config: (url) => ({ ...providerCfg(url), agent: { build: { model: "test/test-model" } } }) },
+  ),
+)
+
+it.live("stopping drops an already queued peer reply while preserving the queued human prompt", () =>
+  provideTmpdirServer(
+    Effect.fnUntraced(function* ({ llm }) {
+      const prompt = yield* SessionPrompt.Service
+      const sessions = yield* Session.Service
+      const status = yield* SessionStatus.Service
+      const chat = yield* sessions.create({ title: "Mixed prompt queue" })
+      yield* llm.hang
+      const current = yield* prompt.promptAsync({
+        sessionID: chat.id,
+        agent: "build",
+        model: ref,
+        parts: [{ type: "text", text: "active work" }],
+      })
+      yield* llm.wait(1)
+      const peer = yield* prompt.promptAsync({
+        sessionID: chat.id,
+        agent: "build",
+        model: ref,
+        parts: [
+          { type: "text", text: "late peer reply", metadata: { kind: "peer_response", sourceSessionID: "peer" } },
+        ],
+      })
+      const human = yield* prompt.promptAsync({
+        sessionID: chat.id,
+        agent: "build",
+        model: ref,
+        parts: [{ type: "text", text: "keep my real message" }],
+      })
+      yield* llm.text("human response")
+      expect(yield* prompt.cancelTurn({ sessionID: chat.id, targetMessageID: current.info.id })).toBe("cancelled")
+      yield* Effect.gen(function* () {
+        while (
+          !(yield* sessions.messages({ sessionID: chat.id })).some(
+            (message) =>
+              message.info.role === "assistant" &&
+              message.info.parentID === human.info.id &&
+              message.info.time.completed,
+          )
+        ) yield* Effect.sleep("1 millis")
+      }).pipe(Effect.timeout("2 seconds"))
+      yield* Effect.sleep("50 millis")
+      const messages = yield* sessions.messages({ sessionID: chat.id, view: "full" })
+      expect(
+        messages.some((message) => message.info.role === "assistant" && message.info.parentID === peer.info.id),
+      ).toBe(false)
+      expect(findRecoverableQueuedPrompt(messages, { includeLegacy: true })).toBeUndefined()
+      expect(yield* llm.calls).toBe(2)
+      expect(JSON.stringify((yield* llm.inputs).at(-1))).not.toContain("late peer reply")
+      expect((yield* status.get(chat.id)).type).toBe("idle")
+    }),
+    { git: true, config: (url) => ({ ...providerCfg(url), agent: { build: { model: "test/test-model" } } }) },
+  ),
+)
+
 it.live("startup recovery finalizes stale foreground work without replaying it", () =>
   provideTmpdirServer(
     Effect.fnUntraced(function* ({ llm }) {