@oh-my-pi/pi-coding-agent 16.1.19 → 16.1.20

package/dist/types/cli/gallery-cli.d.ts

@@ -2,6 +2,12 @@ import { type GalleryFixture } from "./gallery-fixtures";
 /** Lifecycle states the gallery renders, in display order. */
 export declare const GALLERY_STATES: readonly ["streaming", "progress", "success", "error"];
 export type GalleryState = (typeof GALLERY_STATES)[number];
+/** User-facing labels printed above each rendered lifecycle state. */
+export declare const GALLERY_STATE_LABELS: Record<GalleryState, string>;
+/** Accepted `--state` tokens, including legacy lifecycle names and displayed labels. */
+export declare const GALLERY_STATE_TOKENS: string[];
+/** Normalize user-provided `--state` tokens to the internal gallery lifecycle states. */
+export declare function parseGalleryStates(states: readonly string[] | undefined): GalleryState[] | undefined;
 export interface GalleryCommandArgs {
     /** Render width in columns (defaults to terminal width, clamped). */
     width?: number;

package/dist/types/commands/gallery.d.ts

@@ -12,7 +12,7 @@ export default class Gallery extends Command {
         state: import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"string"> & {
             char: string;
             description: string;
-            options: ("error" | "progress" | "streaming" | "success")[];
+            options: string[];
             multiple: true;
         };
         width: import("@oh-my-pi/pi-utils/cli").FlagDescriptor<"integer"> & {

package/dist/types/config/service-tier.d.ts

@@ -0,0 +1,34 @@
+import type { ServiceTier } from "@oh-my-pi/pi-ai";
+import type { SubmenuOption } from "./settings-schema";
+/**
+ * Service-tier setting values shared by every "Service Tier" setting. `"none"`
+ * is the omit-the-parameter sentinel; the remaining values mirror
+ * {@link ServiceTier}.
+ */
+export declare const SERVICE_TIER_SETTING_VALUES: readonly ["none", "auto", "default", "flex", "scale", "priority", "openai-only", "claude-only"];
+export type ServiceTierSettingValue = (typeof SERVICE_TIER_SETTING_VALUES)[number];
+/** Variant value set for scoped service-tier settings (subagent/advisor) that can defer to the main agent. */
+export declare const SERVICE_TIER_INHERIT_SETTING_VALUES: readonly ["inherit", "none", "auto", "default", "flex", "scale", "priority", "openai-only", "claude-only"];
+export type ServiceTierInheritSettingValue = (typeof SERVICE_TIER_INHERIT_SETTING_VALUES)[number];
+/** Submenu descriptions shared by the base `serviceTier` setting. */
+export declare const SERVICE_TIER_OPTIONS: ReadonlyArray<SubmenuOption<ServiceTierSettingValue>>;
+/** Submenu descriptions for inherit-capable service-tier settings. */
+export declare const SERVICE_TIER_INHERIT_OPTIONS: ReadonlyArray<SubmenuOption<ServiceTierInheritSettingValue>>;
+/**
+ * Resolve a service-tier setting value to the wire {@link ServiceTier} (or
+ * `undefined` to omit). `"inherit"` defers to `inherited`; `"none"` omits.
+ */
+export declare function resolveServiceTierSetting(value: string, inherited: ServiceTier | undefined): ServiceTier | undefined;
+/**
+ * Resolve the `serviceTier` *setting value* to stamp onto a subagent's settings
+ * snapshot.
+ *
+ * - A concrete `subagentSetting` (`"none"` or a tier) wins outright.
+ * - `"inherit"` defers to the parent's live effective tier when the caller has a
+ *   live session (`inherited` passed as `ServiceTier | null`, where `null` means
+ *   the parent explicitly has no tier — e.g. `/fast off`). When no live session
+ *   is available (`inherited === undefined`, e.g. cold subagent revive) it falls
+ *   back to the parent's configured `serviceTier` setting so behavior matches a
+ *   plain settings snapshot.
+ */
+export declare function resolveSubagentServiceTier(subagentSetting: string, configuredTier: ServiceTierSettingValue, inherited: ServiceTier | null | undefined): ServiceTierSettingValue;

package/dist/types/config/settings-schema.d.ts

@@ -1153,39 +1153,32 @@ export declare const SETTINGS_SCHEMA: {
             readonly group: "Sampling";
             readonly label: "Service Tier";
             readonly description: 'Processing priority hint (none = omit). OpenAI accepts the tier values directly; Anthropic realizes `priority` as `speed: "fast"` on supported Opus models. Scoped values target one family.';
-            readonly options: readonly [{
-                readonly value: "none";
-                readonly label: "None";
-                readonly description: "Omit service_tier parameter";
-            }, {
-                readonly value: "auto";
-                readonly label: "Auto";
-                readonly description: "Use provider default tier selection (OpenAI)";
-            }, {
-                readonly value: "default";
-                readonly label: "Default";
-                readonly description: "Standard priority processing (OpenAI)";
-            }, {
-                readonly value: "flex";
-                readonly label: "Flex";
-                readonly description: "Flexible capacity tier when available (OpenAI)";
-            }, {
-                readonly value: "scale";
-                readonly label: "Scale";
-                readonly description: "Scale Tier credits when available (OpenAI)";
-            }, {
-                readonly value: "priority";
-                readonly label: "Priority";
-                readonly description: "Priority on every supported provider (OpenAI `service_tier`, Anthropic fast mode)";
-            }, {
-                readonly value: "openai-only";
-                readonly label: "Priority (OpenAI only)";
-                readonly description: "Priority on OpenAI/OpenAI-Codex requests; ignored elsewhere";
-            }, {
-                readonly value: "claude-only";
-                readonly label: "Priority (Claude only)";
-                readonly description: "Anthropic fast mode on direct Claude requests; ignored elsewhere (incl. Bedrock/Vertex)";
-            }];
+            readonly options: readonly SubmenuOption<"auto" | "claude-only" | "default" | "flex" | "none" | "openai-only" | "priority" | "scale">[];
+        };
+    };
+    readonly serviceTierSubagent: {
+        readonly type: "enum";
+        readonly values: readonly ["inherit", "none", "auto", "default", "flex", "scale", "priority", "openai-only", "claude-only"];
+        readonly default: "inherit";
+        readonly ui: {
+            readonly tab: "model";
+            readonly group: "Sampling";
+            readonly label: "Service Tier - Subagent";
+            readonly description: "Service Tier for spawned task/eval subagents. Inherit = match the main agent's live tier (tracks /fast); pick a value to scope subagents independently.";
+            readonly options: readonly SubmenuOption<"auto" | "claude-only" | "default" | "flex" | "inherit" | "none" | "openai-only" | "priority" | "scale">[];
+        };
+    };
+    readonly serviceTierAdvisor: {
+        readonly type: "enum";
+        readonly values: readonly ["inherit", "none", "auto", "default", "flex", "scale", "priority", "openai-only", "claude-only"];
+        readonly default: "none";
+        readonly ui: {
+            readonly tab: "model";
+            readonly group: "Sampling";
+            readonly label: "Service Tier - Advisor";
+            readonly description: "Service Tier for the advisor model. None = standard processing; Inherit = match the main agent's live tier; pick a value (e.g. Priority) to run the advisor on a faster serving path.";
+            readonly options: readonly SubmenuOption<"auto" | "claude-only" | "default" | "flex" | "inherit" | "none" | "openai-only" | "priority" | "scale">[];
+            readonly condition: "advisorEnabled";
         };
     };
     readonly fastModeScope: {
@@ -4243,6 +4236,16 @@ export declare const SETTINGS_SCHEMA: {
             readonly description: "Obfuscate secrets before sending to AI providers";
         };
     };
+    readonly "providers.ollama-cloud.maxConcurrency": {
+        readonly type: "number";
+        readonly default: 3;
+        readonly ui: {
+            readonly tab: "providers";
+            readonly group: "Services";
+            readonly label: "Ollama Cloud Max Concurrency";
+            readonly description: "Maximum concurrent Ollama Cloud subagent runs per process; 0 disables the provider-specific limit";
+        };
+    };
     readonly "providers.webSearch": {
         readonly type: "enum";
         readonly values: readonly ["auto", ...SearchProviderId[]];

package/dist/types/session/agent-session.d.ts

@@ -760,6 +760,8 @@ export declare class AgentSession {
     abort(options?: {
         goalReason?: "interrupted" | "internal";
         reason?: string;
+        /** Internal `/compact` startup keeps the manual-compaction marker alive while aborting the active turn. */
+        preserveCompaction?: boolean;
     }): Promise<void>;
     /**
      * Start a new session, optionally with initial messages and parent tracking.

package/dist/types/task/executor.d.ts

@@ -4,6 +4,7 @@
  * Runs each subagent on the main thread and forwards AgentEvents for progress tracking.
  */
 import type { AgentTelemetryConfig, ThinkingLevel } from "@oh-my-pi/pi-agent-core";
+import type { ServiceTier } from "@oh-my-pi/pi-ai";
 import type { Rule } from "../capability/rule";
 import { ModelRegistry } from "../config/model-registry";
 import type { PromptTemplate } from "../config/prompt-templates";
@@ -119,6 +120,13 @@ export interface ExecutorOptions {
     authStorage?: AuthStorage;
     modelRegistry?: ModelRegistry;
     settings?: Settings;
+    /**
+     * Parent session's live effective service tier, the source of truth for a
+     * subagent whose `serviceTierSubagent` is `"inherit"`. `null` = the parent
+     * explicitly has no tier (e.g. `/fast off`); omitted = no live session, so
+     * inherit falls back to the configured `serviceTier` setting.
+     */
+    parentServiceTier?: ServiceTier | null;
     /** Override local:// protocol options so subagent shares parent's local:// root */
     localProtocolOptions?: LocalProtocolOptions;
     /**
@@ -188,7 +196,7 @@ export declare function finalizeSubprocessOutput(args: FinalizeSubprocessOutputA
  * Create proxy tools that reuse the parent's MCP connections.
  */
 export declare function createMCPProxyTools(mcpManager: MCPManager): CustomTool[];
-export declare function createSubagentSettings(baseSettings: Settings, overrides?: Partial<Record<SettingPath, unknown>>): Settings;
+export declare function createSubagentSettings(baseSettings: Settings, overrides?: Partial<Record<SettingPath, unknown>>, inheritedServiceTier?: ServiceTier | null): Settings;
 /**
  * Run a single agent in-process.
  */

package/dist/types/task/parallel.d.ts

@@ -33,6 +33,22 @@ export declare function mapWithConcurrencyLimit<T, R>(items: T[], concurrency: n
 export declare class Semaphore {
     #private;
     constructor(max: number);
-    acquire(): Promise<void>;
+    /**
+     * Resolves when a slot is available. Pass an `AbortSignal` so callers that
+     * stop waiting (parent task cancelled, wall-clock budget elapsed) also stop
+     * occupying a queue slot — otherwise a later `release()` would resolve the
+     * abandoned waiter, permanently shrinking effective concurrency for the
+     * remaining lifetime of the process (issue #3464 review feedback).
+     */
+    acquire(signal?: AbortSignal): Promise<void>;
     release(): void;
+    /**
+     * Adjust the maximum concurrency in place. Raising the ceiling immediately
+     * admits queued waiters that now fit; lowering it lets in-flight holders
+     * drain naturally (new acquires keep blocking until `#current` falls below
+     * the new max). Resizing the single shared instance — instead of replacing
+     * it — keeps in-flight slots counted, so a runtime or mixed limit change can
+     * never push concurrency past the cap (issue #3464 review feedback).
+     */
+    resize(max: number): void;
 }

package/dist/types/tools/index.d.ts

@@ -1,6 +1,6 @@
 import type { InMemorySnapshotStore } from "@oh-my-pi/hashline";
 import type { AgentTelemetryConfig, AgentTool } from "@oh-my-pi/pi-agent-core";
-import type { FetchImpl, ImageContent, Model, ToolChoice } from "@oh-my-pi/pi-ai";
+import type { FetchImpl, ImageContent, Model, ServiceTier, ToolChoice } from "@oh-my-pi/pi-ai";
 import type { AsyncJobManager } from "../async/job-manager";
 import type { Rule } from "../capability/rule";
 import type { PromptTemplate } from "../config/prompt-templates";
@@ -191,6 +191,8 @@ export interface ToolSession {
     getActiveModelString?: () => string | undefined;
     /** Get the current session model object (provider/api capabilities), regardless of how it was chosen. */
     getActiveModel?: () => Model | undefined;
+    /** Get the session's live effective service tier (undefined = none). Source of truth for subagent `serviceTierSubagent: inherit`. */
+    getServiceTier?: () => ServiceTier | undefined;
     /** Auth storage for passing to subagents (avoids re-discovery) */
     authStorage?: import("../session/auth-storage").AuthStorage;
     /** Model registry for passing to subagents (avoids re-discovery) */

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-coding-agent",
-	"version": "16.1.19",
+	"version": "16.1.20",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -35,7 +35,7 @@
 		"check": "biome check . && bun run check:types",
 		"check:types": "tsgo -p tsconfig.json --noEmit",
 		"lint": "biome lint .",
-		"test": "bun test --parallel=4 test src",
+		"test": "bun ../../scripts/ci-test-ts.ts coding-agent-heavy --full",
 		"fix": "biome check --write --unsafe . && bun run format-prompts",
 		"fmt": "biome format --write . && bun run format-prompts",
 		"format-prompts": "bun scripts/format-prompts.ts",
@@ -48,17 +48,17 @@
 		"@agentclientprotocol/sdk": "0.25.0",
 		"@babel/parser": "^7.29.7",
 		"@mozilla/readability": "^0.6.0",
-		"@oh-my-pi/hashline": "16.1.19",
-		"@oh-my-pi/omp-stats": "16.1.19",
-		"@oh-my-pi/pi-agent-core": "16.1.19",
-		"@oh-my-pi/pi-ai": "16.1.19",
-		"@oh-my-pi/pi-catalog": "16.1.19",
-		"@oh-my-pi/pi-mnemopi": "16.1.19",
-		"@oh-my-pi/pi-natives": "16.1.19",
-		"@oh-my-pi/pi-tui": "16.1.19",
-		"@oh-my-pi/pi-utils": "16.1.19",
-		"@oh-my-pi/pi-wire": "16.1.19",
-		"@oh-my-pi/snapcompact": "16.1.19",
+		"@oh-my-pi/hashline": "16.1.20",
+		"@oh-my-pi/omp-stats": "16.1.20",
+		"@oh-my-pi/pi-agent-core": "16.1.20",
+		"@oh-my-pi/pi-ai": "16.1.20",
+		"@oh-my-pi/pi-catalog": "16.1.20",
+		"@oh-my-pi/pi-mnemopi": "16.1.20",
+		"@oh-my-pi/pi-natives": "16.1.20",
+		"@oh-my-pi/pi-tui": "16.1.20",
+		"@oh-my-pi/pi-utils": "16.1.20",
+		"@oh-my-pi/pi-wire": "16.1.20",
+		"@oh-my-pi/snapcompact": "16.1.20",
 		"@opentelemetry/api": "^1.9.1",
 		"@opentelemetry/context-async-hooks": "^2.7.1",
 		"@opentelemetry/exporter-trace-otlp-proto": "^0.218.0",

package/src/cli/gallery-cli.ts

@@ -21,13 +21,42 @@ import { captureGalleryScreenshots } from "./gallery-screenshot";
 export const GALLERY_STATES = ["streaming", "progress", "success", "error"] as const;
 export type GalleryState = (typeof GALLERY_STATES)[number];
 
-const STATE_LABELS: Record<GalleryState, string> = {
+/** User-facing labels printed above each rendered lifecycle state. */
+export const GALLERY_STATE_LABELS: Record<GalleryState, string> = {
 	streaming: "streaming args",
 	progress: "in progress",
 	success: "done",
 	error: "failed",
 };
 
+const GALLERY_STATE_ALIASES: Record<string, GalleryState> = {
+	streaming: "streaming",
+	"streaming args": "streaming",
+	progress: "progress",
+	"in progress": "progress",
+	success: "success",
+	done: "success",
+	error: "error",
+	failed: "error",
+};
+
+/** Accepted `--state` tokens, including legacy lifecycle names and displayed labels. */
+export const GALLERY_STATE_TOKENS = Object.keys(GALLERY_STATE_ALIASES);
+
+/** Normalize user-provided `--state` tokens to the internal gallery lifecycle states. */
+export function parseGalleryStates(states: readonly string[] | undefined): GalleryState[] | undefined {
+	if (!states || states.length === 0) return undefined;
+	const parsed: GalleryState[] = [];
+	for (const raw of states) {
+		const state = GALLERY_STATE_ALIASES[raw.trim().toLowerCase()];
+		if (!state) {
+			throw new Error(`Invalid --state '${raw}'. Valid values: ${GALLERY_STATE_TOKENS.join(", ")}`);
+		}
+		if (!parsed.includes(state)) parsed.push(state);
+	}
+	return parsed;
+}
+
 export interface GalleryCommandArgs {
 	/** Render width in columns (defaults to terminal width, clamped). */
 	width?: number;
@@ -166,7 +195,7 @@ async function renderGallerySections(
 		const heading = fixture.label && fixture.label !== name ? `${name} — ${fixture.label}` : name;
 		const lines: string[] = ["", sectionRule(heading, width)];
 		for (const state of states) {
-			lines.push("", theme.fg("dim", `  · ${STATE_LABELS[state]}`));
+			lines.push("", theme.fg("dim", `  · ${GALLERY_STATE_LABELS[state]}`));
 			try {
 				for (const line of await renderGalleryState(name, fixture, state, width, expanded)) lines.push(line);
 			} catch (err) {

package/src/cli/gallery-fixtures/agentic.ts

@@ -263,6 +263,11 @@ export const agenticFixtures: Record<string, GalleryFixture> = {
 				],
 			} satisfies IrcDetails,
 		},
+		errorResult: {
+			isError: true,
+			content: [{ type: "text", text: "IRC inbox failed: message store unavailable." }],
+			details: { op: "inbox" } satisfies IrcDetails,
+		},
 	},
 
 	irc_list: {
@@ -309,6 +314,11 @@ export const agenticFixtures: Record<string, GalleryFixture> = {
 				],
 			} satisfies IrcDetails,
 		},
+		errorResult: {
+			isError: true,
+			content: [{ type: "text", text: "IRC list failed: agent hub is unavailable." }],
+			details: { op: "list" } satisfies IrcDetails,
+		},
 	},
 
 	goal: {
@@ -388,19 +398,18 @@ export const agenticFixtures: Record<string, GalleryFixture> = {
 		},
 		errorResult: {
 			isError: true,
-			content: [{ type: "text", text: "Job cancelled by user." }],
+			content: [{ type: "text", text: "1 job failed." }],
 			details: {
 				jobs: [
 					{
 						id: "job_d4",
 						type: "task",
-						status: "cancelled",
+						status: "failed",
 						label: "Refactor the session store to Redis",
 						durationMs: 52_300,
-						errorText: "Aborted: superseded by goal re-scope.",
+						errorText: "Subagent exited 1: Redis connection string is missing.",
 					},
 				],
-				cancelled: [{ id: "job_d4", status: "cancelled" }],
 			},
 		},
 	},

package/src/commands/gallery.ts

@@ -2,7 +2,7 @@
  * Render every built-in tool's renderer across its lifecycle states.
  */
 import { Command, Flags } from "@oh-my-pi/pi-utils/cli";
-import { GALLERY_STATES, type GalleryState, runGalleryCommand } from "../cli/gallery-cli";
+import { GALLERY_STATE_TOKENS, type GalleryState, parseGalleryStates, runGalleryCommand } from "../cli/gallery-cli";
 
 export default class Gallery extends Command {
 	static description = "Preview tool renderers across streaming, in-progress, success, and failure states";
@@ -12,7 +12,7 @@ export default class Gallery extends Command {
 		state: Flags.string({
 			char: "s",
 			description: "Render only the given lifecycle state(s)",
-			options: [...GALLERY_STATES],
+			options: GALLERY_STATE_TOKENS,
 			multiple: true,
 		}),
 		width: Flags.integer({ char: "w", description: "Render width in columns" }),
@@ -37,9 +37,17 @@ export default class Gallery extends Command {
 
 	async run(): Promise<void> {
 		const { flags } = await this.parse(Gallery);
+		let states: GalleryState[] | undefined;
+		try {
+			states = parseGalleryStates(flags.state);
+		} catch (err) {
+			process.stderr.write(`${err instanceof Error ? err.message : String(err)}\n`);
+			process.exitCode = 1;
+			return;
+		}
 		await runGalleryCommand({
 			tool: flags.tool,
-			states: flags.state as GalleryState[] | undefined,
+			states,
 			width: flags.width,
 			expanded: flags.expanded,
 			plain: flags.plain,

package/src/config/service-tier.ts

@@ -0,0 +1,87 @@
+import type { ServiceTier } from "@oh-my-pi/pi-ai";
+import type { SubmenuOption } from "./settings-schema";
+
+/**
+ * Service-tier setting values shared by every "Service Tier" setting. `"none"`
+ * is the omit-the-parameter sentinel; the remaining values mirror
+ * {@link ServiceTier}.
+ */
+export const SERVICE_TIER_SETTING_VALUES = [
+	"none",
+	"auto",
+	"default",
+	"flex",
+	"scale",
+	"priority",
+	"openai-only",
+	"claude-only",
+] as const;
+
+export type ServiceTierSettingValue = (typeof SERVICE_TIER_SETTING_VALUES)[number];
+
+/** Variant value set for scoped service-tier settings (subagent/advisor) that can defer to the main agent. */
+export const SERVICE_TIER_INHERIT_SETTING_VALUES = ["inherit", ...SERVICE_TIER_SETTING_VALUES] as const;
+
+export type ServiceTierInheritSettingValue = (typeof SERVICE_TIER_INHERIT_SETTING_VALUES)[number];
+
+/** Submenu descriptions shared by the base `serviceTier` setting. */
+export const SERVICE_TIER_OPTIONS: ReadonlyArray<SubmenuOption<ServiceTierSettingValue>> = [
+	{ value: "none", label: "None", description: "Omit service_tier parameter" },
+	{ value: "auto", label: "Auto", description: "Use provider default tier selection (OpenAI)" },
+	{ value: "default", label: "Default", description: "Standard priority processing (OpenAI)" },
+	{ value: "flex", label: "Flex", description: "Flexible capacity tier when available (OpenAI)" },
+	{ value: "scale", label: "Scale", description: "Scale Tier credits when available (OpenAI)" },
+	{
+		value: "priority",
+		label: "Priority",
+		description: "Priority on every supported provider (OpenAI `service_tier`, Anthropic fast mode)",
+	},
+	{
+		value: "openai-only",
+		label: "Priority (OpenAI only)",
+		description: "Priority on OpenAI/OpenAI-Codex requests; ignored elsewhere",
+	},
+	{
+		value: "claude-only",
+		label: "Priority (Claude only)",
+		description: "Anthropic fast mode on direct Claude requests; ignored elsewhere (incl. Bedrock/Vertex)",
+	},
+];
+
+/** Submenu descriptions for inherit-capable service-tier settings. */
+export const SERVICE_TIER_INHERIT_OPTIONS: ReadonlyArray<SubmenuOption<ServiceTierInheritSettingValue>> = [
+	{ value: "inherit", label: "Inherit", description: "Use the main agent's Service Tier" },
+	...SERVICE_TIER_OPTIONS,
+];
+
+/**
+ * Resolve a service-tier setting value to the wire {@link ServiceTier} (or
+ * `undefined` to omit). `"inherit"` defers to `inherited`; `"none"` omits.
+ */
+export function resolveServiceTierSetting(value: string, inherited: ServiceTier | undefined): ServiceTier | undefined {
+	if (value === "inherit") return inherited;
+	if (value === "none" || value === "") return undefined;
+	return value as ServiceTier;
+}
+
+/**
+ * Resolve the `serviceTier` *setting value* to stamp onto a subagent's settings
+ * snapshot.
+ *
+ * - A concrete `subagentSetting` (`"none"` or a tier) wins outright.
+ * - `"inherit"` defers to the parent's live effective tier when the caller has a
+ *   live session (`inherited` passed as `ServiceTier | null`, where `null` means
+ *   the parent explicitly has no tier — e.g. `/fast off`). When no live session
+ *   is available (`inherited === undefined`, e.g. cold subagent revive) it falls
+ *   back to the parent's configured `serviceTier` setting so behavior matches a
+ *   plain settings snapshot.
+ */
+export function resolveSubagentServiceTier(
+	subagentSetting: string,
+	configuredTier: ServiceTierSettingValue,
+	inherited: ServiceTier | null | undefined,
+): ServiceTierSettingValue {
+	if (subagentSetting !== "inherit") return subagentSetting as ServiceTierSettingValue;
+	if (inherited === undefined) return configuredTier;
+	return inherited ?? "none";
+}

package/src/config/settings-schema.ts

@@ -35,6 +35,12 @@ import {
 } from "../tts/models";
 import { EDIT_MODES } from "../utils/edit-mode";
 import { SEARCH_PROVIDER_OPTIONS, SEARCH_PROVIDER_PREFERENCES, type SearchProviderId } from "../web/search/types";
+import {
+	SERVICE_TIER_INHERIT_OPTIONS,
+	SERVICE_TIER_INHERIT_SETTING_VALUES,
+	SERVICE_TIER_OPTIONS,
+	SERVICE_TIER_SETTING_VALUES,
+} from "./service-tier";
 
 /** Unified settings schema - single source of truth for all settings.
  *
@@ -1122,7 +1128,7 @@ export const SETTINGS_SCHEMA = {
 
 	serviceTier: {
 		type: "enum",
-		values: ["none", "auto", "default", "flex", "scale", "priority", "openai-only", "claude-only"] as const,
+		values: SERVICE_TIER_SETTING_VALUES,
 		default: "none",
 		ui: {
 			tab: "model",
@@ -1130,28 +1136,36 @@ export const SETTINGS_SCHEMA = {
 			label: "Service Tier",
 			description:
 				'Processing priority hint (none = omit). OpenAI accepts the tier values directly; Anthropic realizes `priority` as `speed: "fast"` on supported Opus models. Scoped values target one family.',
-			options: [
-				{ value: "none", label: "None", description: "Omit service_tier parameter" },
-				{ value: "auto", label: "Auto", description: "Use provider default tier selection (OpenAI)" },
-				{ value: "default", label: "Default", description: "Standard priority processing (OpenAI)" },
-				{ value: "flex", label: "Flex", description: "Flexible capacity tier when available (OpenAI)" },
-				{ value: "scale", label: "Scale", description: "Scale Tier credits when available (OpenAI)" },
-				{
-					value: "priority",
-					label: "Priority",
-					description: "Priority on every supported provider (OpenAI `service_tier`, Anthropic fast mode)",
-				},
-				{
-					value: "openai-only",
-					label: "Priority (OpenAI only)",
-					description: "Priority on OpenAI/OpenAI-Codex requests; ignored elsewhere",
-				},
-				{
-					value: "claude-only",
-					label: "Priority (Claude only)",
-					description: "Anthropic fast mode on direct Claude requests; ignored elsewhere (incl. Bedrock/Vertex)",
-				},
-			],
+			options: SERVICE_TIER_OPTIONS,
+		},
+	},
+
+	serviceTierSubagent: {
+		type: "enum",
+		values: SERVICE_TIER_INHERIT_SETTING_VALUES,
+		default: "inherit",
+		ui: {
+			tab: "model",
+			group: "Sampling",
+			label: "Service Tier - Subagent",
+			description:
+				"Service Tier for spawned task/eval subagents. Inherit = match the main agent's live tier (tracks /fast); pick a value to scope subagents independently.",
+			options: SERVICE_TIER_INHERIT_OPTIONS,
+		},
+	},
+
+	serviceTierAdvisor: {
+		type: "enum",
+		values: SERVICE_TIER_INHERIT_SETTING_VALUES,
+		default: "none",
+		ui: {
+			tab: "model",
+			group: "Sampling",
+			label: "Service Tier - Advisor",
+			description:
+				"Service Tier for the advisor model. None = standard processing; Inherit = match the main agent's live tier; pick a value (e.g. Priority) to run the advisor on a faster serving path.",
+			options: SERVICE_TIER_INHERIT_OPTIONS,
+			condition: "advisorEnabled",
 		},
 	},
 
@@ -4041,6 +4055,17 @@ export const SETTINGS_SCHEMA = {
 	},
 
 	// Provider selection
+	"providers.ollama-cloud.maxConcurrency": {
+		type: "number",
+		default: 3,
+		ui: {
+			tab: "providers",
+			group: "Services",
+			label: "Ollama Cloud Max Concurrency",
+			description:
+				"Maximum concurrent Ollama Cloud subagent runs per process; 0 disables the provider-specific limit",
+		},
+	},
 	"providers.webSearch": {
 		type: "enum",
 		values: SEARCH_PROVIDER_PREFERENCES,

package/src/eval/agent-bridge.ts

@@ -397,6 +397,8 @@ export async function runEvalAgent(args: unknown, options: EvalAgentBridgeOption
 		parentMnemopiSessionState: options.session.getMnemopiSessionState?.(),
 		parentTelemetry: options.session.getTelemetry?.(),
 		parentAgentId: options.session.getAgentId?.() ?? MAIN_AGENT_ID,
+		// Live source of truth for `serviceTierSubagent: inherit` (null = explicit none).
+		parentServiceTier: options.session.getServiceTier ? (options.session.getServiceTier() ?? null) : undefined,
 		// Deliberately omit parentEvalSessionId: the parent's Python kernel is
 		// blocked on this bridge call, so sharing the eval session would deadlock
 		// (subagent queues behind the parent's in-flight execution, parent waits

package/src/main.ts

@@ -140,6 +140,7 @@ const HOST_DEFAULTED_SETTING_PATHS: SettingPath[] = [
 	"advisor.subagents",
 	"advisor.syncBacklog",
 	"advisor.immuneTurns",
+	"serviceTierAdvisor",
 ];
 
 const RPC_BACKGROUND_DEFAULTED_SETTING_PATHS: SettingPath[] = [

package/src/mcp/transports/stdio.ts

@@ -340,6 +340,10 @@ export class StdioTransport implements MCPTransport {
 			platform: process.platform,
 		});
 
+		// Spawn in a new session (detached → setsid) so the MCP process tree has
+		// no controlling terminal. Otherwise terminal job-control signals (Ctrl+Z
+		// SIGTSTP, background-read SIGTTIN) can stop stdio servers such as
+		// chrome-devtools-mcp and leave our read loop blocked on silent pipes.
 		this.#process = spawn({
 			cmd: spawnCommand.cmd,
 			cwd,
@@ -348,6 +352,7 @@ export class StdioTransport implements MCPTransport {
 			stdout: "pipe",
 			stderr: "pipe",
 			windowsHide: spawnCommand.windowsHide,
+			detached: true,
 		});
 
 		this.#connected = true;