@oh-my-pi/pi-coding-agent 16.2.8 → 16.2.11

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

@@ -1813,6 +1813,34 @@ export declare const SETTINGS_SCHEMA: {
             }[];
         };
     };
+    readonly "stt.submitTrigger": {
+        readonly type: "enum";
+        readonly values: readonly ["never", "release", "release-complete", "say-submit"];
+        readonly default: "never";
+        readonly ui: {
+            readonly tab: "interaction";
+            readonly group: "Speech";
+            readonly label: "Speech-to-Text Submit Trigger";
+            readonly description: "Choose when speech dictation automatically submits: Never, Release (2+ words), Release with complete sentence, or When I Say Submit.";
+            readonly options: ({
+                value: "never";
+                label: string;
+                description: string;
+            } | {
+                value: "release";
+                label: string;
+                description: string;
+            } | {
+                value: "release-complete";
+                label: string;
+                description: string;
+            } | {
+                value: "say-submit";
+                label: string;
+                description: string;
+            })[];
+        };
+    };
     readonly "contextPromotion.enabled": {
         readonly type: "boolean";
         readonly default: false;
@@ -4228,7 +4256,7 @@ export declare const SETTINGS_SCHEMA: {
             readonly tab: "tasks";
             readonly group: "Subagents";
             readonly label: "Soft Subagent Request Budget";
-            readonly description: "Soft per-subagent request budget (assistant requests per run). Crossing it injects one steering notice asking the subagent to wrap up; at 1.5x the budget the run is aborted gracefully, salvaging partial output. 0 disables the guard. Bundled explore/quick_task agents use a lower built-in budget.";
+            readonly description: "Soft per-subagent request budget (assistant requests per run). Crossing it injects one steering notice asking the subagent to wrap up; at 1.5x the budget the run is aborted gracefully, salvaging partial output. 0 disables the guard. Bundled explore/sonic agents use a lower built-in budget.";
             readonly options: readonly [{
                 readonly value: "0";
                 readonly label: "Disabled";
@@ -4847,7 +4875,7 @@ export declare const SETTINGS_SCHEMA: {
     };
     readonly "providers.memoryModel": {
         readonly type: "enum";
-        readonly values: readonly ["online", "qwen3-1.7b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
+        readonly values: readonly ["online", "qwen3-1.7b", "llama3.2:3b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
         readonly default: "online";
         readonly ui: {
             readonly tab: "memory";
@@ -4860,15 +4888,15 @@ export declare const SETTINGS_SCHEMA: {
                 label: string;
                 description: string;
             } | {
-                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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
+                value: "gemma-3-1b" | "lfm2-1.2b" | "llama3.2:3b" | "qwen2.5-1.5b" | "qwen3-1.7b";
+                label: "Gemma 3 1B" | "LFM2 1.2B" | "Llama 3.2 3B" | "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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Larger Llama 3.2 option for local memory/classifier tasks; higher quality potential at higher disk/RAM/latency cost.";
             })[];
         };
     };
     readonly "providers.autoThinkingModel": {
         readonly type: "enum";
-        readonly values: readonly ["online", "qwen3-1.7b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
+        readonly values: readonly ["online", "qwen3-1.7b", "llama3.2:3b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
         readonly default: "online";
         readonly ui: {
             readonly tab: "model";
@@ -4881,9 +4909,9 @@ export declare const SETTINGS_SCHEMA: {
                 label: string;
                 description: string;
             } | {
-                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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
+                value: "gemma-3-1b" | "lfm2-1.2b" | "llama3.2:3b" | "qwen2.5-1.5b" | "qwen3-1.7b";
+                label: "Gemma 3 1B" | "LFM2 1.2B" | "Llama 3.2 3B" | "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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Larger Llama 3.2 option for local memory/classifier tasks; higher quality potential at higher disk/RAM/latency cost.";
             })[];
         };
     };
@@ -4899,7 +4927,7 @@ export declare const SETTINGS_SCHEMA: {
     };
     readonly "providers.unexpectedStopModel": {
         readonly type: "enum";
-        readonly values: readonly ["online", "qwen3-1.7b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
+        readonly values: readonly ["online", "qwen3-1.7b", "llama3.2:3b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
         readonly default: "online";
         readonly ui: {
             readonly tab: "providers";
@@ -4912,9 +4940,9 @@ export declare const SETTINGS_SCHEMA: {
                 label: string;
                 description: string;
             } | {
-                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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
+                value: "gemma-3-1b" | "lfm2-1.2b" | "llama3.2:3b" | "qwen2.5-1.5b" | "qwen3-1.7b";
+                label: "Gemma 3 1B" | "LFM2 1.2B" | "Llama 3.2 3B" | "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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Larger Llama 3.2 option for local memory/classifier tasks; higher quality potential at higher disk/RAM/latency cost.";
             })[];
         };
     };

package/dist/types/extensibility/skills.d.ts

@@ -64,4 +64,33 @@ export interface BuiltSkillPromptMessage {
     details: SkillPromptDetails;
 }
 export declare function getSkillSlashCommandName(skill: Pick<Skill, "name">): string;
+/**
+ * Parsed `/skill:<name>` invocation: either at the start of the draft (the
+ * traditional slash-command position) or as a `/skill:<name>` token embedded
+ * mid-prompt. For the mid-prompt form the surrounding prose is threaded
+ * through as `args` so the skill sees the full user request.
+ */
+export interface ParsedSkillInvocation {
+    /** Bare skill name without the leading `skill:` prefix. */
+    name: string;
+    /** User-supplied arguments (everything outside the `/skill:<name>` token). */
+    args: string;
+}
+/**
+ * Detect a `/skill:<name>` invocation in a user draft.
+ *
+ * Returns `undefined` when the text contains no skill token. Otherwise:
+ *   - Leading form (`/skill:foo bar baz`): name=`foo`, args=`bar baz`.
+ *   - Mid-prompt form (`fix the bug /skill:foo focus on auth`): name=`foo`,
+ *     args=`fix the bug focus on auth` — the surrounding prose collapsed
+ *     into a single args string.
+ *
+ * Mid-prompt detection is disabled when the draft itself starts with a
+ * different slash command (e.g. `/compact /skill:foo`) or a local-execution
+ * sigil — `!cmd` / `!!cmd` for the bash tool and `$ cmd` / `$$ cmd` for the
+ * python tool. Those handlers run after the skill-command dispatcher and
+ * their bodies routinely contain `/skill:<name>` references that are not
+ * meant as skill invocations.
+ */
+export declare function parseSkillInvocation(text: string): ParsedSkillInvocation | undefined;
 export declare function buildSkillPromptMessage(skill: Pick<Skill, "name" | "filePath">, args: string): Promise<BuiltSkillPromptMessage>;

package/dist/types/modes/components/mcp-add-wizard.d.ts

@@ -21,6 +21,14 @@ export interface MCPAddWizardOAuthResult {
 interface MCPAddWizardOAuthOptions {
     serverUrl?: string;
     resource?: string;
+    /**
+     * External cancellation source. Aborting it tears down the in-flight OAuth
+     * flow and surfaces a neutral cancellation error. The wizard wires its own
+     * controller here so Esc cancels the OAuth wait instead of stepping back
+     * through the form (the wizard is focused, so the editor's Esc hook does
+     * not fire).
+     */
+    abortSignal?: AbortSignal;
 }
 export declare class MCPAddWizard extends Container {
     #private;

package/dist/types/modes/components/todo-reminder.d.ts

@@ -1,7 +1,9 @@
 import { Container } from "@oh-my-pi/pi-tui";
 import type { TodoItem } from "../../tools/todo";
 /**
- * Component that renders a todo completion reminder notification.
+ * Component that renders a todo completion reminder notification, committed into
+ * the transcript like a TTSR notification so it stays anchored in history rather
+ * than floating above the editor.
  * Shows when the agent stops with incomplete todos.
  */
 export declare class TodoReminderComponent extends Container {

package/dist/types/modes/controllers/mcp-command-controller.d.ts

@@ -7,6 +7,15 @@ export declare class MCPAuthorizationLinkPrompt implements Component {
     invalidate(): void;
     render(_width: number): readonly string[];
 }
+/**
+ * Thrown by {@link MCPCommandController}'s OAuth handler when the user (or a
+ * caller-supplied {@link AbortSignal}) cancels the in-flight flow. Distinct
+ * from network/timeout failures so callers can surface a neutral
+ * "cancelled" status instead of an error banner.
+ */
+export declare class MCPOAuthCancelledError extends Error {
+    constructor(message?: string);
+}
 export declare class MCPCommandController {
     #private;
     private ctx;

package/dist/types/modes/controllers/tool-args-reveal.d.ts

@@ -36,6 +36,11 @@ export declare class ToolArgsRevealController {
     setTarget(id: string, partialJson: string, target: ToolArgsRevealTarget): Record<string, unknown>;
     /** Attach the component future ticks push frames into. */
     bind(id: string, component: ToolArgsRevealComponent): void;
+    /** Migrate a live reveal entry from a placeholder key onto the real
+     *  tool-call id once the owned-dialect parser assigns it. No-op when no
+     *  entry exists under `from` (smoothing disabled, or the JSON already
+     *  closed and `finish` cleared it). */
+    rekey(from: string, to: string): void;
     /** Final arguments arrived (the JSON closed): drop the reveal so the
      *  caller's final-args render wins immediately, mirroring how assistant
      *  text snaps to the full message at message_end. */

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

@@ -83,7 +83,6 @@ export declare class InteractiveMode implements InteractiveModeContext {
     chatContainer: TranscriptContainer;
     pendingMessagesContainer: Container;
     statusContainer: Container;
-    todoReminderContainer: Container;
     todoContainer: Container;
     subagentContainer: Container;
     btwContainer: Container;

package/dist/types/modes/skill-command.d.ts

@@ -23,7 +23,7 @@ export interface BuiltSkillCommandPrompt {
     message: SkillPromptMessage;
     options: SkillPromptOptions;
 }
-/** Return true when `text` names a registered `/skill:<name>` command. */
+/** Return true when `text` invokes a registered `/skill:<name>` command. */
 export declare function isKnownSkillCommand(ctx: SkillCommandHost, text: string): boolean;
 /** Build the user-attributed custom message for a registered `/skill:<name>` command. */
 export declare function buildSkillCommandPrompt(ctx: SkillCommandHost, text: string, streamingBehavior: "steer" | "followUp", images?: ImageContent[]): Promise<BuiltSkillCommandPrompt | undefined>;

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

@@ -81,7 +81,6 @@ export interface InteractiveModeContext {
     chatContainer: TranscriptContainer;
     pendingMessagesContainer: Container;
     statusContainer: Container;
-    todoReminderContainer: Container;
     todoContainer: Container;
     subagentContainer: Container;
     btwContainer: Container;

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

@@ -871,7 +871,7 @@ export declare class AgentSession {
      * log but NOT to settings.
      * @throws Error if no API key available for the model
      */
-    setModelTemporary(model: Model, thinkingLevel?: ThinkingLevel, options?: {
+    setModelTemporary(model: Model, thinkingLevel?: ConfiguredThinkingLevel, options?: {
         ephemeral?: boolean;
     }): Promise<void>;
     /**

package/dist/types/stt/asr-client.d.ts

@@ -1,4 +1,4 @@
-import { type SpawnedSubprocess, type WorkerHandle } from "../subprocess/worker-client";
+import { type RefCountedWorkerHandle, type SpawnedSubprocess } from "../subprocess/worker-client";
 import type { SttProgressEvent, SttWorkerInbound, SttWorkerOutbound } from "./asr-protocol";
 import type { SttModelKey } from "./models";
 export interface SttTranscribeOptions {
@@ -9,6 +9,10 @@ export interface SttDownloadOptions {
     signal?: AbortSignal;
     onProgress?: (event: SttProgressEvent) => void;
 }
+export interface SttDownloadResult {
+    ok: boolean;
+    error?: string;
+}
 /** Live streaming session handle returned by {@link SttClient.startStream}. */
 export interface SttStreamHandle {
     /** Feed 16 kHz mono float samples as the recorder produces them. */
@@ -38,7 +42,7 @@ export declare const STT_WORKER_ARG = "__omp_worker_stt";
 export declare function createSttSubprocess(): SpawnedSubprocess<SttWorkerOutbound>;
 export declare class SttClient {
     #private;
-    constructor(spawnWorker?: () => WorkerHandle<SttWorkerInbound, SttWorkerOutbound>);
+    constructor(spawnWorker?: () => RefCountedWorkerHandle<SttWorkerInbound, SttWorkerOutbound>);
     onProgress(listener: (event: SttProgressEvent) => void): () => void;
     /**
      * Transcribe 16 kHz mono audio on the warm worker. Rejects with the worker
@@ -54,7 +58,7 @@ export declare class SttClient {
      * an aborted signal) tears the session down and resolves `stop()` with "".
      */
     startStream(modelKey: SttModelKey, options?: SttStreamOptions): SttStreamHandle;
-    downloadModel(modelKey: SttModelKey, options?: SttDownloadOptions): Promise<boolean>;
+    downloadModel(modelKey: SttModelKey, options?: SttDownloadOptions): Promise<SttDownloadResult>;
     terminate(): Promise<void>;
 }
 export declare const sttClient: SttClient;

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

@@ -3,5 +3,6 @@ export * from "./asr-protocol";
 export * from "./downloader";
 export * from "./models";
 export * from "./stt-controller";
+export * from "./submit-trigger";
 export * from "./transcriber";
 export * from "./wav";

package/dist/types/stt/stt-controller.d.ts

@@ -12,6 +12,8 @@ interface Editor {
     setVolatileText(text: string): void;
     clearVolatileText(): void;
     commitVolatileText(text: string): void;
+    submit(): void;
+    deleteBeforeCursor(count: number): void;
 }
 export declare class STTController {
     #private;

package/dist/types/stt/submit-trigger.d.ts

@@ -0,0 +1,30 @@
+/**
+ * TTS/STT Submit Trigger options and evaluation logic.
+ */
+export declare const STT_SUBMIT_TRIGGER_VALUES: readonly ["never", "release", "release-complete", "say-submit"];
+export type SttSubmitTrigger = (typeof STT_SUBMIT_TRIGGER_VALUES)[number];
+export declare const STT_SUBMIT_TRIGGER_OPTIONS: ({
+    value: "never";
+    label: string;
+    description: string;
+} | {
+    value: "release";
+    label: string;
+    description: string;
+} | {
+    value: "release-complete";
+    label: string;
+    description: string;
+} | {
+    value: "say-submit";
+    label: string;
+    description: string;
+})[];
+/**
+ * Evaluate the submit trigger against a transcribed utterance.
+ * Returns whether to submit, and the number of characters to trim from the end of the utterance.
+ */
+export declare function evaluateSubmitTrigger(utterance: string, trigger: SttSubmitTrigger): {
+    submit: boolean;
+    trimTrailing: number;
+};

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

@@ -22,7 +22,7 @@ export declare function formatResultOutputFallback(result: Pick<SingleResult, "o
  * Advisory — never a rejection — nudging the spawner toward tailored
  * specialists when it spawns generic role-less workers and still holds spawn
  * capacity (DepthCapacity: it currently has the `task` tool). Fires when a
- * generic `task`/`quick_task` spawn carries no `role`, or when one call clones
+ * generic `task`/`sonic` spawn carries no `role`, or when one call clones
  * the same agent ≥2× all without roles. Returns undefined when no nudge applies.
  */
 export declare function buildSpecializationAdvisory(agentName: string | undefined, items: TaskItem[], depthCapacity: boolean): string | undefined;

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

@@ -74,7 +74,7 @@ export interface TaskItem {
     isolated?: boolean;
 }
 export declare const taskSchema: import("arktype/internal/variants/object.ts").ObjectType<{
-    agent: string;
+    agent: import("arktype/internal/attributes.ts").Default<string, "task">;
     id?: string | undefined;
     description?: string | undefined;
     role?: string | undefined;
@@ -82,20 +82,20 @@ export declare const taskSchema: import("arktype/internal/variants/object.ts").O
     isolated?: boolean | undefined;
 }, {}>;
 declare const ALL_TASK_SCHEMAS: readonly [import("arktype/internal/variants/object.ts").ObjectType<{
-    agent: string;
+    agent: import("arktype/internal/attributes.ts").Default<string, "task">;
     id?: string | undefined;
     description?: string | undefined;
     role?: string | undefined;
     assignment: string;
     isolated?: boolean | undefined;
 }, {}>, import("arktype/internal/variants/object.ts").ObjectType<{
-    agent: string;
+    agent: import("arktype/internal/attributes.ts").Default<string, "task">;
     id?: string | undefined;
     description?: string | undefined;
     role?: string | undefined;
     assignment: string;
 }, {}>, import("arktype/internal/variants/object.ts").ObjectType<{
-    agent: string;
+    agent: import("arktype/internal/attributes.ts").Default<string, "task">;
     context: string;
     tasks: {
         id?: string | undefined;
@@ -105,7 +105,7 @@ declare const ALL_TASK_SCHEMAS: readonly [import("arktype/internal/variants/obje
         isolated?: boolean | undefined;
     }[];
 }, {}>, import("arktype/internal/variants/object.ts").ObjectType<{
-    agent: string;
+    agent: import("arktype/internal/attributes.ts").Default<string, "task">;
     context: string;
     tasks: {
         id?: string | undefined;
@@ -129,7 +129,7 @@ export declare function getTaskSchema(options: {
  * transcripts using the flat form keep working under either setting.
  */
 export interface TaskParams {
-    /** Agent type; required. */
+    /** Agent type to spawn; defaults to `"task"` (the general-purpose worker) when omitted. */
     agent?: string;
     /** Stable agent id (flat form); default = generated AdjectiveNoun. */
     id?: string;

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

@@ -84,6 +84,13 @@ export declare const TINY_MEMORY_LOCAL_MODELS: readonly [{
     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: "llama3.2:3b";
+    readonly repo: "onnx-community/Llama-3.2-3B-Instruct-ONNX";
+    readonly dtype: "q4";
+    readonly label: "Llama 3.2 3B";
+    readonly description: "Larger Llama 3.2 option for local memory/classifier tasks; higher quality potential at higher disk/RAM/latency cost.";
+    readonly contextNote: "Use when larger model capacity is preferred over faster load times.";
 }, {
     readonly key: "gemma-3-1b";
     readonly repo: "onnx-community/gemma-3-1b-it-ONNX";
@@ -106,7 +113,7 @@ export declare const TINY_MEMORY_LOCAL_MODELS: readonly [{
     readonly description: "Fastest load; solid all-rounder, slightly noisier extraction labels.";
     readonly contextNote: "Use when local startup cost is the priority.";
 }];
-export declare const TINY_MEMORY_MODEL_VALUES: readonly ["online", "qwen3-1.7b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
+export declare const TINY_MEMORY_MODEL_VALUES: readonly ["online", "qwen3-1.7b", "llama3.2:3b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
 export type TinyMemoryModelKey = (typeof TINY_MEMORY_MODEL_VALUES)[number];
 export type TinyMemoryLocalModelKey = (typeof TINY_MEMORY_LOCAL_MODELS)[number]["key"];
 export declare const TINY_MEMORY_MODEL_OPTIONS: ({
@@ -114,9 +121,9 @@ export declare const TINY_MEMORY_MODEL_OPTIONS: ({
     label: string;
     description: string;
 } | {
-    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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
+    value: "gemma-3-1b" | "lfm2-1.2b" | "llama3.2:3b" | "qwen2.5-1.5b" | "qwen3-1.7b";
+    label: "Gemma 3 1B" | "LFM2 1.2B" | "Llama 3.2 3B" | "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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Larger Llama 3.2 option for local memory/classifier tasks; higher quality potential at higher disk/RAM/latency cost.";
 })[];
 export declare function isTinyMemoryLocalModelKey(value: string): value is TinyMemoryLocalModelKey;
 export declare function getTinyMemoryModelSpec(key: TinyMemoryLocalModelKey): (typeof TINY_MEMORY_LOCAL_MODELS)[number];
@@ -173,6 +180,13 @@ export declare const TINY_LOCAL_MODELS: readonly [{
     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: "llama3.2:3b";
+    readonly repo: "onnx-community/Llama-3.2-3B-Instruct-ONNX";
+    readonly dtype: "q4";
+    readonly label: "Llama 3.2 3B";
+    readonly description: "Larger Llama 3.2 option for local memory/classifier tasks; higher quality potential at higher disk/RAM/latency cost.";
+    readonly contextNote: "Use when larger model capacity is preferred over faster load times.";
 }, {
     readonly key: "gemma-3-1b";
     readonly repo: "onnx-community/gemma-3-1b-it-ONNX";
@@ -203,14 +217,14 @@ export declare const TINY_LOCAL_MODELS: readonly [{
  * sub-1B title models.
  */
 export declare const ONLINE_AUTO_THINKING_MODEL_KEY = "online";
-export declare const AUTO_THINKING_MODEL_VALUES: readonly ["online", "qwen3-1.7b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
+export declare const AUTO_THINKING_MODEL_VALUES: readonly ["online", "qwen3-1.7b", "llama3.2:3b", "gemma-3-1b", "qwen2.5-1.5b", "lfm2-1.2b"];
 export type AutoThinkingModelKey = TinyMemoryModelKey;
 export declare const AUTO_THINKING_MODEL_OPTIONS: ({
     value: "online";
     label: string;
     description: string;
 } | {
-    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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels.";
+    value: "gemma-3-1b" | "lfm2-1.2b" | "llama3.2:3b" | "qwen2.5-1.5b" | "qwen3-1.7b";
+    label: "Gemma 3 1B" | "LFM2 1.2B" | "Llama 3.2 3B" | "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." | "Disabled for local inference: onnxruntime-node cannot run this ONNX export's RotaryEmbedding cache updates." | "Fastest load; solid all-rounder, slightly noisier extraction labels." | "Larger Llama 3.2 option for local memory/classifier tasks; higher quality potential at higher disk/RAM/latency cost.";
 })[];

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-coding-agent",
-	"version": "16.2.8",
+	"version": "16.2.11",
 	"description": "Coding agent CLI with read, bash, edit, write tools and session management",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -32,7 +32,8 @@
 	},
 	"scripts": {
 		"build": "bun scripts/build-binary.ts",
-		"check": "biome check . && bun run check:types",
+		"check": "biome check . && bun run check:docs && bun run check:types",
+		"check:docs": "bun scripts/generate-docs-index.ts --check",
 		"check:types": "tsgo -p tsconfig.json --noEmit",
 		"lint": "biome lint .",
 		"test": "bun ../../scripts/ci-test-ts.ts coding-agent-heavy --full",
@@ -55,17 +56,17 @@
 		"@agentclientprotocol/sdk": "0.25.0",
 		"@babel/parser": "^7.29.7",
 		"@mozilla/readability": "^0.6.0",
-		"@oh-my-pi/hashline": "16.2.8",
-		"@oh-my-pi/omp-stats": "16.2.8",
-		"@oh-my-pi/pi-agent-core": "16.2.8",
-		"@oh-my-pi/pi-ai": "16.2.8",
-		"@oh-my-pi/pi-catalog": "16.2.8",
-		"@oh-my-pi/pi-mnemopi": "16.2.8",
-		"@oh-my-pi/pi-natives": "16.2.8",
-		"@oh-my-pi/pi-tui": "16.2.8",
-		"@oh-my-pi/pi-utils": "16.2.8",
-		"@oh-my-pi/pi-wire": "16.2.8",
-		"@oh-my-pi/snapcompact": "16.2.8",
+		"@oh-my-pi/hashline": "16.2.11",
+		"@oh-my-pi/omp-stats": "16.2.11",
+		"@oh-my-pi/pi-agent-core": "16.2.11",
+		"@oh-my-pi/pi-ai": "16.2.11",
+		"@oh-my-pi/pi-catalog": "16.2.11",
+		"@oh-my-pi/pi-mnemopi": "16.2.11",
+		"@oh-my-pi/pi-natives": "16.2.11",
+		"@oh-my-pi/pi-tui": "16.2.11",
+		"@oh-my-pi/pi-utils": "16.2.11",
+		"@oh-my-pi/pi-wire": "16.2.11",
+		"@oh-my-pi/snapcompact": "16.2.11",
 		"@opentelemetry/api": "^1.9.1",
 		"@opentelemetry/context-async-hooks": "^2.7.1",
 		"@opentelemetry/exporter-trace-otlp-proto": "^0.218.0",

package/scripts/bundle-dist.ts

@@ -3,6 +3,7 @@
 import * as fs from "node:fs/promises";
 import * as path from "node:path";
 import { isEnoent } from "@oh-my-pi/pi-utils";
+import { assertDocsIndexFresh, buildDocsIndexPayload } from "./generate-docs-index";
 
 const packageDir = path.join(import.meta.dir, "..");
 const outDir = path.join(packageDir, "dist");
@@ -72,13 +73,31 @@ async function cleanBundleOutputs(): Promise<void> {
 	);
 }
 
+async function assertDocsEmbedPopulated(): Promise<void> {
+	// bundle-dist runs from prepack (which calls `gen:docs` first) or directly.
+	// Direct invocations must fail — the tarball ships src/, and an empty embed
+	// would make src/internal-urls/docs-index.ts fall through to the missing
+	// repo `docs/` tree at runtime in published packages (codex review, PR #3941).
+	const embedPath = path.join(packageDir, "src/internal-urls/docs-index.generated.txt");
+	const embed = await Bun.file(embedPath).text();
+	if (embed.length === 0) {
+		throw new Error(
+			"docs-index embed is empty. Run `bun run gen:docs` before `bun run gen:bundle`, or use `bun pm pack` which runs the prepack chain.",
+		);
+	}
+	const expected = await buildDocsIndexPayload();
+	assertDocsIndexFresh(embed, expected);
+}
+
 async function main(): Promise<void> {
 	const start = Bun.nanoseconds();
 	await cleanBundleOutputs();
-	// The npm bundle ships no stats dashboard sources or prebuilt dist/client,
-	// so embed the dashboard archive the same way compiled binaries do
-	// (scripts/build-binary.ts). Reset afterwards to keep the checked-in
-	// placeholder empty.
+	await assertDocsEmbedPopulated();
+	// The npm bundle ships no stats dashboard sources, so embed the dashboard
+	// archive the same way compiled binaries do (scripts/build-binary.ts). Reset
+	// afterwards to keep the checked-in placeholder empty. The docs embed stays
+	// populated on disk — postpack owns its reset so `bun pm pack` can pack a
+	// tarball whose src copy is still valid for subpath imports.
 	await runCommand(["bun", "--cwd=../stats", "run", "gen:stats"]);
 	try {
 		await runCommand([