@oh-my-pi/pi-coding-agent 16.1.7 → 16.1.8

package/dist/types/cli/session-picker.d.ts

@@ -6,5 +6,4 @@ import type { SessionInfo } from "../session/session-listing";
  */
 export declare function selectSession(sessions: SessionInfo[], options?: {
     allSessions?: SessionInfo[];
-    startInAllScope?: boolean;
 }): Promise<SessionInfo | null>;

package/dist/types/cli/tiny-models-cli.d.ts

@@ -1,3 +1,4 @@
+import { type TinyLocalModelKey } from "../tiny/models";
 export type TinyModelsAction = "download" | "list";
 export interface TinyModelsCommandArgs {
     action: TinyModelsAction;
@@ -6,4 +7,5 @@ export interface TinyModelsCommandArgs {
         json?: boolean;
     };
 }
+export declare function resolveModels(model: string | undefined): TinyLocalModelKey[];
 export declare function runTinyModelsCommand(command: TinyModelsCommandArgs): Promise<void>;

package/dist/types/collab/protocol.d.ts

@@ -42,11 +42,30 @@ export type CollabFrame = Exclude<GuestFrame, {
     t: "welcome";
     proto: number;
     header: SessionHeader;
-    entries: SessionEntry[];
     state: CollabSessionState;
     agents: AgentSnapshot[];
+    /**
+     * Total number of `SessionEntry` items the host will deliver in the
+     * `snapshot-chunk` frames that follow. The guest stays in the
+     * snapshot-loading phase until it has accumulated that many entries
+     * (or a chunk arrives with `final: true`).
+     */
+    entryCount: number;
     /** True when this peer joined through a read-only (view) link. */
     readOnly?: boolean;
+}
+/**
+ * Targeted snapshot fragment delivered after `welcome`. Splits a large
+ * transcript across many small frames so the guest's per-chunk progress
+ * timeout resets each time the relay delivers another batch; without
+ * chunking, a multi-MB session has to fit one giant frame inside the
+ * 30 s first-welcome budget. The last chunk carries `final: true` so the
+ * guest can finalize the replica session.
+ */
+ | {
+    t: "snapshot-chunk";
+    entries: SessionEntry[];
+    final: boolean;
 } | {
     t: "entry";
     entry: SessionEntry;

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

@@ -201,7 +201,7 @@ export declare const SETTINGS_SCHEMA: {
     };
     readonly "advisor.immuneTurns": {
         readonly type: "number";
-        readonly default: 1;
+        readonly default: 3;
         readonly ui: {
             readonly tab: "model";
             readonly group: "Advisor";
@@ -214,13 +214,13 @@ export declare const SETTINGS_SCHEMA: {
             }, {
                 readonly value: "1";
                 readonly label: "1 turn";
-                readonly description: "Default.";
             }, {
                 readonly value: "2";
                 readonly label: "2 turns";
             }, {
                 readonly value: "3";
                 readonly label: "3 turns";
+                readonly description: "Default.";
             }, {
                 readonly value: "4";
                 readonly label: "4 turns";
@@ -852,6 +852,26 @@ export declare const SETTINGS_SCHEMA: {
             readonly description: "Hide thinking blocks in assistant responses";
         };
     };
+    readonly proseOnlyThinking: {
+        readonly type: "boolean";
+        readonly default: true;
+        readonly ui: {
+            readonly tab: "model";
+            readonly group: "Thinking";
+            readonly label: "Prose Only Thinking";
+            readonly description: "Omit code blocks from thinking summaries and replace them with an ellipsis";
+        };
+    };
+    readonly omitThinking: {
+        readonly type: "boolean";
+        readonly default: false;
+        readonly ui: {
+            readonly tab: "model";
+            readonly group: "Thinking";
+            readonly label: "Omit Thinking summaries";
+            readonly description: "Instruct upstream providers to completely omit thinking summaries from responses (where supported)";
+        };
+    };
     readonly "model.loopGuard.enabled": {
         readonly type: "boolean";
         readonly default: true;
@@ -1669,7 +1689,7 @@ export declare const SETTINGS_SCHEMA: {
     readonly "compaction.strategy": {
         readonly type: "enum";
         readonly values: readonly ["context-full", "handoff", "shake", "snapcompact", "off"];
-        readonly default: "context-full";
+        readonly default: "snapcompact";
         readonly ui: {
             readonly tab: "context";
             readonly group: "Compaction";
@@ -2362,6 +2382,17 @@ export declare const SETTINGS_SCHEMA: {
             readonly condition: "mnemopiActive";
         };
     };
+    readonly "mnemopi.proactiveLinking": {
+        readonly type: "boolean";
+        readonly default: false;
+        readonly ui: {
+            readonly tab: "memory";
+            readonly group: "Mnemopi";
+            readonly label: "Mnemopi Proactive Linking";
+            readonly description: "Ingest new memories into the episodic graph as they are stored, linking them to related entities and memories";
+            readonly condition: "mnemopiActive";
+        };
+    };
     readonly "mnemopi.noEmbeddings": {
         readonly type: "boolean";
         readonly default: false;
@@ -4294,6 +4325,26 @@ export declare const SETTINGS_SCHEMA: {
             }];
         };
     };
+    readonly "providers.fireworksTier": {
+        readonly type: "enum";
+        readonly values: readonly ["standard", "priority"];
+        readonly default: "standard";
+        readonly ui: {
+            readonly tab: "providers";
+            readonly group: "Fireworks";
+            readonly label: "Fireworks Tier";
+            readonly description: 'Serving path for Fireworks requests. Priority sends `service_tier: "priority"` for higher reliability during peak traffic at a higher price; Standard omits it. Fast (`-fast`) models ignore this — Fast is its own serving path.';
+            readonly options: readonly [{
+                readonly value: "standard";
+                readonly label: "Standard";
+                readonly description: "Default serving path (no service_tier)";
+            }, {
+                readonly value: "priority";
+                readonly label: "Priority";
+                readonly description: "Priority serving path: higher reliability, premium per-token pricing";
+            }];
+        };
+    };
     readonly "providers.tts": {
         readonly type: "enum";
         readonly values: readonly ["auto", "local", "xai"];
@@ -4568,7 +4619,7 @@ export declare const SETTINGS_SCHEMA: {
             } | {
                 value: "gemma-3-1b" | "lfm2-1.2b" | "qwen2.5-1.5b" | "qwen3-1.7b";
                 label: "Gemma 3 1B" | "LFM2 1.2B" | "Qwen2.5 1.5B" | "Qwen3 1.7B";
-                description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.";
+                description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
             })[];
         };
     };
@@ -4589,7 +4640,7 @@ export declare const SETTINGS_SCHEMA: {
             } | {
                 value: "gemma-3-1b" | "lfm2-1.2b" | "qwen2.5-1.5b" | "qwen3-1.7b";
                 label: "Gemma 3 1B" | "LFM2 1.2B" | "Qwen2.5 1.5B" | "Qwen3 1.7B";
-                description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.";
+                description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
             })[];
         };
     };
@@ -4620,7 +4671,7 @@ export declare const SETTINGS_SCHEMA: {
             } | {
                 value: "gemma-3-1b" | "lfm2-1.2b" | "qwen2.5-1.5b" | "qwen3-1.7b";
                 label: "Gemma 3 1B" | "LFM2 1.2B" | "Qwen2.5 1.5B" | "Qwen3 1.7B";
-                description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.";
+                description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
             })[];
         };
     };

package/dist/types/extensibility/legacy-pi-ai-shim.d.ts

@@ -19,6 +19,7 @@
  * `types.ts` via the `export *` below — pi-ai still exports both as types,
  * only the runtime `Type` builder and `StringEnum()` helper were removed.
  */
+import { getBundledModel, getBundledModels } from "@oh-my-pi/pi-catalog/models";
 import { type TSchema, Type } from "./typebox";
 export interface StringEnumOptions<T extends string> {
     description?: string;
@@ -26,6 +27,9 @@ export interface StringEnumOptions<T extends string> {
     examples?: T[];
     [key: string]: unknown;
 }
-export declare function StringEnum<T extends string>(values: readonly T[], options?: StringEnumOptions<T>): TSchema;
+export declare function StringEnum<T extends string | number>(values: readonly T[] | Record<string, T>, options?: StringEnumOptions<any>): TSchema;
 export * from "@oh-my-pi/pi-ai";
 export { Type };
+/** Compatibility aliases for renamed catalog functions */
+export declare const getModel: typeof getBundledModel;
+export declare const getModels: typeof getBundledModels;

package/dist/types/internal-urls/filesystem-resource.d.ts

@@ -0,0 +1,9 @@
+import type { InternalResource } from "./types";
+/**
+ * Builds a text resource for a filesystem directory resolved by an internal URL handler.
+ *
+ * The resource is flagged immutable so the read tool never mints hashline edit
+ * anchors against a directory listing — only file resources from the same
+ * handler stay editable.
+ */
+export declare function buildDirectoryResource(url: string, directoryPath: string, notes?: string[]): Promise<InternalResource>;

package/dist/types/mcp/loader.d.ts

@@ -2,6 +2,7 @@ import type { LoadedCustomTool } from "../extensibility/custom-tools/types";
 import { AgentStorage } from "../session/agent-storage";
 import type { AuthStorage } from "../session/auth-storage";
 import { MCPManager } from "./manager";
+import type { McpConnectionStatusEvent } from "./startup-events";
 /** Result from loading MCP tools */
 export interface MCPToolsLoadResult {
     /** MCP manager (for lifecycle management) */
@@ -20,8 +21,8 @@ export interface MCPToolsLoadResult {
 }
 /** Options for loading MCP tools */
 export interface MCPToolsLoadOptions {
-    /** Called when starting to connect to servers */
-    onConnecting?: (serverNames: string[]) => void;
+    /** Called when MCP server connection state changes. */
+    onStatus?: (event: McpConnectionStatusEvent) => void;
     /** Whether to load project-level config (default: true) */
     enableProjectConfig?: boolean;
     /** Whether to filter out Exa MCP servers (default: true) */

package/dist/types/mcp/manager.d.ts

@@ -2,6 +2,7 @@ import { type TSchema } from "@oh-my-pi/pi-ai";
 import type { SourceMeta } from "../capability/types";
 import type { CustomTool } from "../extensibility/custom-tools/types";
 import type { AuthStorage } from "../session/auth-storage";
+import type { McpConnectionStatusEvent } from "./startup-events";
 import type { MCPToolDetails } from "./tool-bridge";
 import type { MCPToolCache } from "./tool-cache";
 import type { MCPGetPromptResult, MCPPrompt, MCPRequestOptions, MCPResource, MCPResourceReadResult, MCPResourceTemplate, MCPServerConfig, MCPServerConnection } from "./types";
@@ -38,8 +39,8 @@ export interface MCPDiscoverOptions {
     filterExa?: boolean;
     /** Whether to filter out browser MCP servers when builtin browser tool is enabled (default: false) */
     filterBrowser?: boolean;
-    /** Called when starting to connect to servers */
-    onConnecting?: (serverNames: string[]) => void;
+    /** Called when MCP server connection state changes. */
+    onStatus?: (event: McpConnectionStatusEvent) => void;
 }
 /**
  * MCP Server Manager.
@@ -87,7 +88,7 @@ export declare class MCPManager {
      * Connect to specific MCP servers.
      * Connections are made in parallel for faster startup.
      */
-    connectServers(configs: Record<string, MCPServerConfig>, sources: Record<string, SourceMeta>, onConnecting?: (serverNames: string[]) => void): Promise<MCPLoadResult>;
+    connectServers(configs: Record<string, MCPServerConfig>, sources: Record<string, SourceMeta>, onStatus?: (event: McpConnectionStatusEvent) => void): Promise<MCPLoadResult>;
     /**
      * Get all loaded tools.
      */

package/dist/types/mcp/startup-events.d.ts

@@ -1,11 +1,28 @@
-export declare const MCP_CONNECTING_EVENT_CHANNEL = "mcp:connecting";
-export type McpConnectingEvent = {
+export declare const MCP_CONNECTION_STATUS_EVENT_CHANNEL = "mcp:connection-status";
+export type McpConnectionStatusEvent = {
+    type: "connecting";
     serverNames: string[];
+} | {
+    type: "connected";
+    serverName: string;
+} | {
+    type: "failed";
+    serverName: string;
+    error: string;
 };
-export declare function formatMCPConnectingMessage(serverNames: string[]): string;
+export type McpConnectionStatusSnapshot = {
+    pendingServers: readonly string[];
+    connectedServers: readonly string[];
+    failedServers: readonly {
+        serverName: string;
+        error: string;
+    }[];
+};
+export declare function formatMCPConnectingMessage(serverNames: readonly string[]): string;
+export declare function formatMCPConnectionStatusMessage(snapshot: McpConnectionStatusSnapshot): string;
 /**
  * Runtime validator for the cross-module event payload. The event bus is
  * untyped at runtime, so the subscriber verifies the shape before formatting
  * rather than trusting a cast — a malformed emit is ignored instead of throwing.
  */
-export declare function isMcpConnectingEvent(data: unknown): data is McpConnectingEvent;
+export declare function isMcpConnectionStatusEvent(data: unknown): data is McpConnectionStatusEvent;

package/dist/types/mnemopi/config.d.ts

@@ -15,6 +15,7 @@ export interface MnemopiBackendConfig {
     autoRetain: boolean;
     polyphonicRecall: boolean;
     enhancedRecall: boolean;
+    proactiveLinking: boolean;
     retainEveryNTurns: number;
     recallLimit: number;
     recallContextTurns: number;

package/dist/types/modes/components/agent-hub.d.ts

@@ -40,6 +40,7 @@ export interface AgentHubDeps {
     cwd?: string;
     /** Mirrors the main transcript's thinking-block visibility. */
     hideThinkingBlock?: () => boolean;
+    proseOnlyThinking?: () => boolean;
     /** Keys toggling tool output expansion (app.tools.expand). */
     expandKeys?: KeyId[];
     /** Focus the main view on this agent's live session (ctx.focusAgentSession). When absent (collab guest, tests), Enter opens the in-hub chat view instead. */

package/dist/types/modes/components/agent-transcript-viewer.d.ts

@@ -20,6 +20,7 @@ export interface AgentTranscriptViewerDeps {
     getMessageRenderer?: (customType: string) => MessageRenderer | undefined;
     cwd: string;
     hideThinkingBlock?: () => boolean;
+    proseOnlyThinking?: () => boolean;
     expandKeys: KeyId[];
     /** Keys that toggle the whole hub closed (app.agents.hub + app.session.observe). */
     hubKeys: KeyId[];

package/dist/types/modes/components/assistant-message.d.ts

@@ -2,6 +2,8 @@ import type { AssistantMessage, ImageContent } from "@oh-my-pi/pi-ai";
 import { Container, type ImageBudget } from "@oh-my-pi/pi-tui";
 import type { AssistantThinkingRenderer } from "../../extensibility/extensions/types";
 import { type CacheInvalidation } from "./cache-invalidation-marker";
+/** Test-only: clear the shared gauge so observations don't leak across cases. */
+export declare function resetThinkingSpeedTracker(): void;
 /**
  * Component that renders a complete assistant message
  */
@@ -11,7 +13,8 @@ export declare class AssistantMessageComponent extends Container {
     private readonly onImageUpdate?;
     private readonly thinkingRenderers;
     private readonly imageBudget?;
-    constructor(message?: AssistantMessage, hideThinkingBlock?: boolean, onImageUpdate?: (() => void) | undefined, thinkingRenderers?: readonly AssistantThinkingRenderer[], imageBudget?: ImageBudget | undefined);
+    private proseOnlyThinking;
+    constructor(message?: AssistantMessage, hideThinkingBlock?: boolean, onImageUpdate?: (() => void) | undefined, thinkingRenderers?: readonly AssistantThinkingRenderer[], imageBudget?: ImageBudget | undefined, proseOnlyThinking?: boolean);
     /**
      * Show or clear the slim cache-invalidation divider above this turn. Set at
      * `message_end` (live) or during rebuild, once the turn's usage is known and
@@ -21,6 +24,7 @@ export declare class AssistantMessageComponent extends Container {
     setCacheInvalidation(info: CacheInvalidation | undefined): void;
     invalidate(): void;
     setHideThinkingBlock(hide: boolean): void;
+    setProseOnlyThinking(proseOnly: boolean): void;
     dispose(): void;
     /**
      * Toggle suppression of the inline `Error: …` line while the same error is

package/dist/types/modes/components/btw-panel.d.ts

@@ -12,6 +12,8 @@ export declare class BtwPanelComponent extends Container {
     markAborted(): void;
     markError(message: string): void;
     isBranchable(): boolean;
+    isCopyable(): boolean;
+    getCopyText(): string | undefined;
     close(): void;
 }
 export {};

package/dist/types/modes/components/chat-transcript-builder.d.ts

@@ -22,6 +22,7 @@ export interface ChatTranscriptBuilderDeps {
     getMessageRenderer?: (customType: string) => MessageRenderer | undefined;
     cwd: string;
     hideThinkingBlock?: () => boolean;
+    proseOnlyThinking?: () => boolean;
     requestRender: () => void;
 }
 export declare class ChatTranscriptBuilder {

package/dist/types/modes/components/session-selector.d.ts

@@ -50,8 +50,6 @@ export interface SessionSelectorOptions {
     loadAllSessions?: () => Promise<SessionInfo[]>;
     /** Preloaded all-projects list; cached so the first Tab toggle is instant. */
     allSessions?: SessionInfo[];
-    /** Open directly in all-projects scope (e.g. the current folder has no sessions). */
-    startInAllScope?: boolean;
     /**
      * Reads the live terminal height so the visible window fits the viewport.
      * Omitted only in tests; defaults to a conservative 24 rows.

package/dist/types/modes/controllers/btw-controller.d.ts

@@ -5,6 +5,8 @@ export declare class BtwController {
     constructor(ctx: InteractiveModeContext);
     hasActiveRequest(): boolean;
     canBranch(): boolean;
+    canCopy(): boolean;
+    handleCopy(): Promise<boolean>;
     handleBranch(): Promise<boolean>;
     handleEscape(): boolean;
     dispose(): void;

package/dist/types/modes/controllers/selector-controller.d.ts

@@ -1,6 +1,7 @@
 import type { Component } from "@oh-my-pi/pi-tui";
 import type { InteractiveModeContext } from "../../modes/types";
 import type { SessionObserverRegistry } from "../session-observer-registry";
+export declare function formatTemporaryModelStatus(modelLabel: string, roleSelectorHint?: string): string;
 export declare class SelectorController {
     #private;
     private ctx;

package/dist/types/modes/controllers/streaming-reveal.d.ts

@@ -7,10 +7,11 @@ type StreamingRevealComponent = Pick<AssistantMessageComponent, "updateContent">
 type StreamingRevealControllerOptions = {
     getSmoothStreaming(): boolean;
     getHideThinkingBlock(): boolean;
+    getProseOnlyThinking(): boolean;
     requestRender(): void;
 };
-export declare function visibleUnits(message: AssistantMessage, hideThinking: boolean): number;
-export declare function buildDisplayMessage(target: AssistantMessage, revealed: number, hideThinking: boolean, countOf?: (index: number, text: string) => number): AssistantMessage;
+export declare function visibleUnits(message: AssistantMessage, hideThinking: boolean, proseOnly?: boolean): number;
+export declare function buildDisplayMessage(target: AssistantMessage, revealed: number, hideThinking: boolean, proseOnly?: boolean, countOf?: (index: number, text: string) => number): AssistantMessage;
 export declare function nextStep(backlog: number): number;
 export declare class StreamingRevealController {
     #private;

package/dist/types/modes/interactive-mode.d.ts

@@ -108,6 +108,7 @@ export declare class InteractiveMode implements InteractiveModeContext {
     loopLimit: LoopLimitRuntime | undefined;
     todoPhases: TodoPhase[];
     hideThinkingBlock: boolean;
+    proseOnlyThinking: boolean;
     pendingImages: ImageContent[];
     pendingImageLinks: (string | undefined)[];
     compactionQueuedMessages: CompactionQueuedMessage[];
@@ -334,6 +335,8 @@ export declare class InteractiveMode implements InteractiveModeContext {
     handleBtwEscape(): boolean;
     canBranchBtw(): boolean;
     handleBtwBranchKey(): Promise<boolean>;
+    canCopyBtw(): boolean;
+    handleBtwCopyKey(): Promise<boolean>;
     handleBtwBranch(question: string, assistantMessage: AssistantMessage): Promise<void>;
     handleOmfgCommand(complaint: string): Promise<void>;
     hasActiveOmfg(): boolean;

package/dist/types/modes/types.d.ts

@@ -129,6 +129,7 @@ export interface InteractiveModeContext {
     loopLimit?: LoopLimitRuntime;
     planModePlanFilePath?: string;
     hideThinkingBlock: boolean;
+    proseOnlyThinking: boolean;
     pendingImages: ImageContent[];
     pendingImageLinks: (string | undefined)[];
     compactionQueuedMessages: CompactionQueuedMessage[];
@@ -325,6 +326,8 @@ export interface InteractiveModeContext {
     handleBtwEscape(): boolean;
     handleBtwBranchKey(): Promise<boolean>;
     canBranchBtw(): boolean;
+    canCopyBtw(): boolean;
+    handleBtwCopyKey(): Promise<boolean>;
     handleBtwBranch(question: string, assistantMessage: AssistantMessage): Promise<void>;
     handleOmfgCommand(complaint: string): Promise<void>;
     hasActiveOmfg(): boolean;

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

@@ -1062,12 +1062,14 @@ export declare class AgentSession {
     emitIrcRelayObservation(record: CustomMessage): void;
     /**
      * Run a single ephemeral side-channel turn against this session's current
-     * model + system prompt + history.  No tools are used; the side request
-     * does not block on, or interfere with, any in-flight main turn.  The
+     * model + system prompt + history. The main turn's tool catalog is sent
+     * to preserve the prompt cache, but the model is reminded not to call
+     * tools and any tool calls are discarded. The side request
+     * does not block on, or interfere with, any in-flight main turn. The
      * session's history and persisted state are NOT modified by this call.
      *
      * Used by `BtwController` (`/btw`) and `OmfgController` (`/omfg`) to share
-     * the snapshot + stream pipeline.  The snapshot includes any in-flight
+     * the snapshot + stream pipeline. The snapshot includes any in-flight
      * streaming assistant text so the model sees the half-finished response
      * rather than missing context.
      */

package/dist/types/session/session-history-format.d.ts

@@ -7,7 +7,32 @@ export interface HistoryFormatOptions {
     includeToolIntent?: boolean;
     /** Render watched-session roles as inline `**agent**:` / `**user**:` labels (collapsing consecutive same-role messages) instead of `## ` headings, so a primary transcript embedded inside an advisor turn stays visually distinct. */
     watchedRoles?: boolean;
+    /**
+     * Expand the primary agent's injected constraint context — plan mode's rules
+     * (`plan-mode-context`) and the approved plan it implements
+     * (`plan-mode-reference`) — verbatim instead of as a truncated one-liner,
+     * wrapped in a `<primary-context>` tag so a reviewer reads it as the primary's
+     * instructions, not its own. The advisor sets this: a truncated rule (plan
+     * mode's "NEVER create files … except the plan file") makes it raise false
+     * blockers. See {@link PRIMARY_CONTEXT_CUSTOM_TYPES}. Other custom messages
+     * still collapse to a one-liner.
+     */
+    expandPrimaryContext?: boolean;
 }
+/**
+ * Hidden custom messages that inject the primary agent's operative *constraints*
+ * — plan mode's rules and the approved plan it implements. A reviewer (the
+ * advisor) must read these verbatim; truncating them hides load-bearing
+ * exceptions (e.g. plan mode permits exactly one plan file). Every other custom
+ * type stays a one-liner.
+ *
+ * Deliberately excludes `goal-mode-context`: its body carries live budget
+ * counters (tokens/seconds used) that change every turn, so it can neither be
+ * deduped against a prior copy nor expanded each turn without flooding the
+ * reviewer — and its constraints don't drive the file-write misreads this
+ * targets.
+ */
+export declare const PRIMARY_CONTEXT_CUSTOM_TYPES: ReadonlySet<string>;
 /**
  * Format a session's message array as a concise markdown transcript.
  *

package/dist/types/session/snapcompact-inline.d.ts

@@ -141,5 +141,5 @@ export declare class SnapcompactInlineTransformer {
     private readonly options;
     private readonly onToolResultSavings?;
     constructor(options: SnapcompactInlineOptions, onToolResultSavings?: SnapcompactSavingsSink | undefined);
-    transform(context: Context, model: Model): Context;
+    transform(context: Context, model: Model): Promise<Context>;
 }

package/dist/types/slash-commands/builtin-registry.d.ts

@@ -3,6 +3,11 @@ import type { BuiltinSlashCommand, ParsedSlashCommand, SlashCommandResult, Slash
 export type { BuiltinSlashCommand, SubcommandDef } from "./types";
 /** TUI-specific runtime accepted by `executeBuiltinSlashCommand`. */
 export type BuiltinSlashCommandRuntime = TuiSlashCommandRuntime;
+export interface TuiBuiltinSlashCommand extends BuiltinSlashCommand {
+    getArgumentCompletions?: (prefix: string) => AutocompleteItem[] | null;
+    getInlineHint?: (argumentText: string) => string | null;
+    getAutocompleteDescription?: () => string | undefined;
+}
 export declare const BUILTIN_SLASH_COMMAND_RESERVED_NAMES: ReadonlySet<string>;
 /** Builtin command metadata used for slash-command autocomplete and help text. */
 export declare const BUILTIN_SLASH_COMMAND_DEFS: ReadonlyArray<BuiltinSlashCommand>;
@@ -10,10 +15,8 @@ export declare const BUILTIN_SLASH_COMMAND_DEFS: ReadonlyArray<BuiltinSlashComma
  * Materialized builtin slash commands with completion functions derived from
  * declarative subcommand/hint definitions.
  */
-export declare const BUILTIN_SLASH_COMMANDS: ReadonlyArray<BuiltinSlashCommand & {
-    getArgumentCompletions?: (prefix: string) => AutocompleteItem[] | null;
-    getInlineHint?: (argumentText: string) => string | null;
-}>;
+export declare const BUILTIN_SLASH_COMMANDS: ReadonlyArray<TuiBuiltinSlashCommand>;
+export declare function buildTuiBuiltinSlashCommands(runtime: TuiSlashCommandRuntime): ReadonlyArray<TuiBuiltinSlashCommand>;
 /**
  * Unified registry exposed for cross-mode tooling. Each spec carries at least
  * one of `handle` / `handleTui`. The TUI dispatcher prefers `handleTui`; the

package/dist/types/slash-commands/types.d.ts

@@ -18,6 +18,8 @@ export interface BuiltinSlashCommand {
     subcommands?: SubcommandDef[];
     /** Static inline hint when command takes a simple argument (no subcommands). */
     inlineHint?: string;
+    /** TUI-only dynamic status text for command-name autocomplete. Static `description` remains canonical for ACP/help. */
+    getTuiAutocompleteDescription?: (runtime: TuiSlashCommandRuntime) => string | undefined;
 }
 /** Parsed slash-command text after stripping the leading "/". */
 export interface ParsedSlashCommand {

package/dist/types/system-prompt.d.ts

@@ -34,6 +34,7 @@ export declare function loadProjectContextFiles(options?: LoadContextFilesOption
  * Project-level SYSTEM.md overrides user-level SYSTEM.md.
  */
 export declare function loadSystemPromptFiles(options?: LoadContextFilesOptions): Promise<string | null>;
+export declare const DEFAULT_SYSTEM_PROMPT_TOOL_NAMES: readonly ["read", "bash", "eval", "edit", "write"];
 export interface SystemPromptToolMetadata {
     label: string;
     description: string;

package/dist/types/tiny/models.d.ts

@@ -11,6 +11,8 @@ export interface TinyTitleLocalModelSpec {
     contextNote: string;
     /** Model family emits hidden reasoning unless the chat template disables it. */
     reasoning?: boolean;
+    /** Reason this model is blocked before loading the ONNX runtime. */
+    unsupportedReason?: string;
 }
 export declare const TINY_TITLE_LOCAL_MODELS: readonly [{
     readonly key: "lfm2-350m";
@@ -66,7 +68,7 @@ export declare function getTinyTitleModelSpec(key: TinyTitleLocalModelKey): (typ
 /** Default memory model: the online path (the configured smol / remote LLM; no local download). */
 export declare const ONLINE_MEMORY_MODEL_KEY = "online";
 /** Recommended local model for memory tasks when none is named. */
-export declare const DEFAULT_MEMORY_LOCAL_MODEL_KEY = "qwen3-1.7b";
+export declare const DEFAULT_MEMORY_LOCAL_MODEL_KEY = "lfm2-1.2b";
 /**
  * Local models for Mnemopi memory tasks (fact extraction + consolidation).
  * These are larger (1B-1.7B) than the title models: structured extraction and
@@ -78,9 +80,10 @@ export declare const TINY_MEMORY_LOCAL_MODELS: readonly [{
     readonly repo: "onnx-community/Qwen3-1.7B-ONNX";
     readonly dtype: "q4";
     readonly label: "Qwen3 1.7B";
-    readonly description: "Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.";
-    readonly contextNote: "Best single-model pick for memory from the local experiment.";
+    readonly description: "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates.";
+    readonly contextNote: "Blocked before load to avoid the unsupported RotaryEmbedding runtime path.";
     readonly reasoning: true;
+    readonly unsupportedReason: "onnxruntime-node does not support Qwen3 RotaryEmbedding cache updates in onnx-community/Qwen3-1.7B-ONNX";
 }, {
     readonly key: "gemma-3-1b";
     readonly repo: "onnx-community/gemma-3-1b-it-ONNX";
@@ -113,7 +116,7 @@ export declare const TINY_MEMORY_MODEL_OPTIONS: ({
 } | {
     value: "gemma-3-1b" | "lfm2-1.2b" | "qwen2.5-1.5b" | "qwen3-1.7b";
     label: "Gemma 3 1B" | "LFM2 1.2B" | "Qwen2.5 1.5B" | "Qwen3 1.7B";
-    description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.";
+    description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
 })[];
 export declare function isTinyMemoryLocalModelKey(value: string): value is TinyMemoryLocalModelKey;
 export declare function getTinyMemoryModelSpec(key: TinyMemoryLocalModelKey): (typeof TINY_MEMORY_LOCAL_MODELS)[number];
@@ -166,9 +169,10 @@ export declare const TINY_LOCAL_MODELS: readonly [{
     readonly repo: "onnx-community/Qwen3-1.7B-ONNX";
     readonly dtype: "q4";
     readonly label: "Qwen3 1.7B";
-    readonly description: "Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.";
-    readonly contextNote: "Best single-model pick for memory from the local experiment.";
+    readonly description: "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates.";
+    readonly contextNote: "Blocked before load to avoid the unsupported RotaryEmbedding runtime path.";
     readonly reasoning: true;
+    readonly unsupportedReason: "onnxruntime-node does not support Qwen3 RotaryEmbedding cache updates in onnx-community/Qwen3-1.7B-ONNX";
 }, {
     readonly key: "gemma-3-1b";
     readonly repo: "onnx-community/gemma-3-1b-it-ONNX";
@@ -208,5 +212,5 @@ export declare const AUTO_THINKING_MODEL_OPTIONS: ({
 } | {
     value: "gemma-3-1b" | "lfm2-1.2b" | "qwen2.5-1.5b" | "qwen3-1.7b";
     label: "Gemma 3 1B" | "LFM2 1.2B" | "Qwen2.5 1.5B" | "Qwen3 1.7B";
-    description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Recommended; most disciplined extraction (ignores chit-chat), good consolidation, about 1.1 GB cached.";
+    description: "Best consolidation/dedup; lighter footprint, but leaks small talk during extraction." | "Best extraction granularity (atomic facts); weaker consolidation." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
 })[];

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

@@ -74,6 +74,7 @@ export declare function applyOpsToPhases(currentPhases: TodoPhase[], ops: TodoPa
     phases: TodoPhase[];
     errors: string[];
 };
+export declare function resolveTodoMarkdownPath(input: string, cwd: string): string;
 /** Render todo phases as a Markdown checklist suitable for editing/copying. */
 export declare function phasesToMarkdown(phases: TodoPhase[]): string;
 /** Parse a Markdown checklist back into todo phases. */