@oh-my-pi/pi-agent-core 15.10.11 → 15.10.12

package/CHANGELOG.md

@@ -2,6 +2,17 @@
 
 ## [Unreleased]
 
+## [15.10.12] - 2026-06-10
+
+### Added
+
+- Added `AgentLoopConfig.getDisableReasoning` so callers can override `disableReasoning` per LLM call, mirroring `getReasoning`.
+- Added `transformProviderContext` to `AgentOptions`/`AgentLoopConfig`: an optional hook applied to the assembled provider context after conversion, normalization, and append-only handling, but before telemetry capture and provider send.
+
+### Fixed
+
+- Fixed `Agent` runs so explicit reasoning disablement is forwarded to provider stream options and re-resolved per continuation, keeping mid-run thinking-off changes in sync with the next provider request.
+
 ## [15.10.11] - 2026-06-10
 
 ### Changed
@@ -10,6 +21,7 @@
 - Catalog imports moved to the new `@oh-my-pi/pi-catalog` package: subpath imports (`calculateCost`, Codex wire constants) plus catalog values previously taken from the `@oh-my-pi/pi-ai` root (`getBundledModel`, `clampThinkingLevelForModel`), which pi-ai no longer re-exports; type-only `Model`/`Api`/`Effort` imports from pi-ai are unchanged
 
 ## [15.10.8] - 2026-06-09
+
 ### Added
 
 - Added optional `fetch` overrides to `SummaryOptions` and `compact`/`generateSummary` so remote compaction can use custom HTTP clients
@@ -18,6 +30,7 @@
 - Added the upstream provider that served a request (`AssistantMessage.upstreamProvider`, e.g. OpenRouter's routed provider) as a `pi.gen_ai.response.upstream_provider` chat-span telemetry attribute, alongside the existing response id and time-to-first-chunk.
 
 ## [15.10.5] - 2026-06-08
+
 ### Removed
 
 - Removed the `maxToolCallsPerTurn` option from `AgentOptions` and `AgentLoopConfig`, so assistant turns are no longer capped after a configured number of completed tool calls
@@ -55,7 +68,6 @@
 - Removed stale synthetic user-message tag filters from OpenAI remote compaction output preservation; developer messages are now dropped by role instead.
 - Tool executions now receive the active turn `AbortSignal` unconditionally.
 
-
 ## [15.10.2] - 2026-06-08
 
 ### Fixed
@@ -87,6 +99,7 @@
 - Surfaced Anthropic stream failures whose message starts with `Output blocked by conten` as normal assistant error lifecycle events, so interactive clients render content-filter blocks instead of silently dropping the streaming bubble at `agent_end`.
 
 ## [15.8.3] - 2026-06-03
+
 ### Added
 
 - Added `getReadToolPath(context)` to `@oh-my-pi/pi-agent-core/compaction/tool-protection` to extract a paired `read` tool call's `path` for embedders building read-targeted protection matchers
@@ -651,4 +664,4 @@ Initial release under @oh-my-pi scope. See previous releases at [badlogic/pi-mon
 
 - `Agent` constructor now has all options optional (empty options use defaults).
 
-- `queueMessage()` is now synchronous (no longer returns a Promise).
+- `queueMessage()` is now synchronous (no longer returns a Promise).

package/dist/types/agent.d.ts

@@ -1,4 +1,4 @@
-import { type ApiKeyResolveContext, type AssistantMessage, type AssistantMessageEvent, type CursorExecHandlers, type CursorToolResultHandler, type Effort, type ImageContent, type Message, type Model, type ProviderSessionState, type ServiceTier, type SimpleStreamOptions, type ThinkingBudgets, type ToolChoice } from "@oh-my-pi/pi-ai";
+import { type ApiKeyResolveContext, type AssistantMessage, type AssistantMessageEvent, type Context, type CursorExecHandlers, type CursorToolResultHandler, type Effort, type ImageContent, type Message, type Model, type ProviderSessionState, type ServiceTier, type SimpleStreamOptions, type ThinkingBudgets, type ToolChoice } from "@oh-my-pi/pi-ai";
 import type { AppendOnlyContextManager } from "./append-only-context";
 import type { HarmonyAuditEvent } from "./harmony-leak";
 import type { AgentEvent, AgentLoopConfig, AgentMessage, AgentState, AgentTool, AgentToolContext, AsideMessage, StreamFn, ToolCallContext } from "./types";
@@ -17,6 +17,11 @@ export interface AgentOptions {
      * Use for context pruning, injecting external context, etc.
      */
     transformContext?: (messages: AgentMessage[], signal?: AbortSignal) => Promise<AgentMessage[]>;
+    /**
+     * Optional transform applied after provider context assembly and before
+     * telemetry capture/provider send.
+     */
+    transformProviderContext?: (context: Context) => Context;
     /**
      * Steering mode: "all" = send all steering messages at once, "one-at-a-time" = one per turn
      */
@@ -294,6 +299,7 @@ export declare class Agent {
     setSystemPrompt(v: string[]): void;
     setModel(m: Model): void;
     setThinkingLevel(l: Effort | undefined): void;
+    setDisableReasoning(disabled: boolean): void;
     setSteeringMode(mode: "all" | "one-at-a-time"): void;
     getSteeringMode(): "all" | "one-at-a-time";
     setFollowUpMode(mode: "all" | "one-at-a-time"): void;

package/dist/types/compaction/compaction.d.ts

@@ -4,10 +4,10 @@
  * Pure functions for compaction logic. The session manager handles I/O,
  * and after compaction the session is reloaded.
  */
-import { type FetchImpl, type MessageAttribution, type Model, type Usage } from "@oh-my-pi/pi-ai";
+import { type FetchImpl, type MessageAttribution, type Model, type Tool, type Usage } from "@oh-my-pi/pi-ai";
 import { type AgentTelemetry } from "../telemetry";
 import { ThinkingLevel } from "../thinking";
-import type { AgentMessage, AgentTool } from "../types";
+import type { AgentMessage } from "../types";
 import type { SessionEntry } from "./entries";
 import { type ConvertToLlm } from "./messages";
 import { type FileOperations } from "./utils";
@@ -133,7 +133,7 @@ export interface HandoffOptions {
     /** Live agent system prompt — passed verbatim so providers hit the cached prefix. */
     systemPrompt: string[];
     /** Live agent tool list — same purpose. Forced to `toolChoice: "none"`. */
-    tools?: AgentTool<any>[];
+    tools?: Tool[];
     customInstructions?: string;
     convertToLlm?: ConvertToLlm;
     initiatorOverride?: MessageAttribution;

package/dist/types/types.d.ts

@@ -1,4 +1,4 @@
-import type { ApiKeyResolveContext, AssistantMessage, AssistantMessageEvent, AssistantMessageEventStream, Effort, ImageContent, Message, Model, SimpleStreamOptions, Static, streamSimple, TextContent, Tool, ToolChoice, ToolResultMessage, TSchema } from "@oh-my-pi/pi-ai";
+import type { ApiKeyResolveContext, AssistantMessage, AssistantMessageEvent, AssistantMessageEventStream, Context, Effort, ImageContent, Message, Model, SimpleStreamOptions, Static, streamSimple, TextContent, Tool, ToolChoice, ToolResultMessage, TSchema } from "@oh-my-pi/pi-ai";
 import type { AppendOnlyContextManager } from "./append-only-context";
 import type { HarmonyAuditEvent } from "./harmony-leak";
 import type { AgentRunCoverage, AgentRunSummary } from "./run-collector";
@@ -79,6 +79,12 @@ export interface AgentLoopConfig extends SimpleStreamOptions {
      * ```
      */
     transformContext?: (messages: AgentMessage[], signal?: AbortSignal) => Promise<AgentMessage[]>;
+    /**
+     * Optional transform applied to the final provider context after conversion,
+     * normalization, and append-only context handling, but before telemetry capture
+     * and provider send.
+     */
+    transformProviderContext?: (context: Context) => Context;
     /**
      * Resolves an API key dynamically for each LLM call.
      *
@@ -171,6 +177,14 @@ export interface AgentLoopConfig extends SimpleStreamOptions {
      * the next model call instead of waiting for the next prompt.
      */
     getReasoning?: () => Effort | undefined;
+    /**
+     * Dynamic reasoning-disable override, resolved per LLM call. When set,
+     * its return value overrides the static `disableReasoning` from
+     * `SimpleStreamOptions` for that request. Pair with `getReasoning` so
+     * mid-run transitions into and out of the explicit `off` state propagate
+     * to the next provider call.
+     */
+    getDisableReasoning?: () => boolean | undefined;
     /**
      * Called after a tool call has been validated and is about to execute.
      *
@@ -307,6 +321,7 @@ export interface AgentState {
     systemPrompt: string[];
     model: Model;
     thinkingLevel?: Effort;
+    disableReasoning?: boolean;
     tools: AgentTool<any>[];
     messages: AgentMessage[];
     isStreaming: boolean;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-agent-core",
-	"version": "15.10.11",
+	"version": "15.10.12",
 	"description": "General-purpose agent with transport abstraction, state management, and attachment support",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -35,10 +35,10 @@
 		"fmt": "biome format --write ."
 	},
 	"dependencies": {
-		"@oh-my-pi/pi-ai": "15.10.11",
-		"@oh-my-pi/pi-catalog": "15.10.11",
-		"@oh-my-pi/pi-natives": "15.10.11",
-		"@oh-my-pi/pi-utils": "15.10.11",
+		"@oh-my-pi/pi-ai": "15.10.12",
+		"@oh-my-pi/pi-catalog": "15.10.12",
+		"@oh-my-pi/pi-natives": "15.10.12",
+		"@oh-my-pi/pi-utils": "15.10.12",
 		"@opentelemetry/api": "^1.9.1"
 	},
 	"devDependencies": {

package/src/agent-loop.ts

@@ -829,6 +829,9 @@ async function streamAssistantResponse(
 			tools: normalizeTools(context.tools, !!config.intentTracing),
 		};
 	}
+	if (config.transformProviderContext) {
+		llmContext = config.transformProviderContext(llmContext);
+	}
 
 	const streamFunction = streamFn || streamSimple;
 
@@ -845,6 +848,7 @@ async function streamAssistantResponse(
 
 	const dynamicToolChoice = config.getToolChoice?.();
 	const dynamicReasoning = config.getReasoning?.();
+	const dynamicDisableReasoning = config.getDisableReasoning?.();
 	const harmonyMitigationEnabled = isHarmonyLeakMitigationTarget(config.model);
 	const harmonyAbortController = harmonyMitigationEnabled ? new AbortController() : undefined;
 	const requestSignal = harmonyAbortController
@@ -856,6 +860,7 @@ async function streamAssistantResponse(
 		harmonyRetryAttempt > 0 && config.temperature !== undefined ? config.temperature + 0.05 : config.temperature;
 	const effectiveToolChoice = dynamicToolChoice ?? config.toolChoice;
 	const effectiveReasoning = dynamicReasoning ?? config.reasoning;
+	const effectiveDisableReasoning = dynamicDisableReasoning ?? config.disableReasoning;
 
 	const chatStepNumber = stepCounter.count;
 	stepCounter.count += 1;
@@ -916,6 +921,7 @@ async function streamAssistantResponse(
 				metadata: resolvedMetadata,
 				toolChoice: effectiveToolChoice,
 				reasoning: effectiveReasoning,
+				disableReasoning: effectiveDisableReasoning,
 				temperature: effectiveTemperature,
 				signal: requestSignal,
 				onResponse: captureOnResponse,

package/src/agent.ts

@@ -6,6 +6,7 @@ import {
 	type ApiKeyResolveContext,
 	type AssistantMessage,
 	type AssistantMessageEvent,
+	type Context,
 	type CursorExecHandlers,
 	type CursorToolResultHandler,
 	type Effort,
@@ -93,6 +94,12 @@ export interface AgentOptions {
 	 */
 	transformContext?: (messages: AgentMessage[], signal?: AbortSignal) => Promise<AgentMessage[]>;
 
+	/**
+	 * Optional transform applied after provider context assembly and before
+	 * telemetry capture/provider send.
+	 */
+	transformProviderContext?: (context: Context) => Context;
+
 	/**
 	 * Steering mode: "all" = send all steering messages at once, "one-at-a-time" = one per turn
 	 */
@@ -265,6 +272,7 @@ export class Agent {
 		systemPrompt: [],
 		model: getBundledModel("google", "gemini-2.5-flash-lite-preview-06-17"),
 		thinkingLevel: undefined,
+		disableReasoning: false,
 		tools: [],
 		messages: [],
 		isStreaming: false,
@@ -277,6 +285,7 @@ export class Agent {
 	#abortController?: AbortController;
 	#convertToLlm: (messages: AgentMessage[]) => Message[] | Promise<Message[]>;
 	#transformContext?: (messages: AgentMessage[], signal?: AbortSignal) => Promise<AgentMessage[]>;
+	#transformProviderContext?: (context: Context) => Context;
 	#steeringQueue: AgentMessage[] = [];
 	#followUpQueue: AgentMessage[] = [];
 	#steeringMode: "all" | "one-at-a-time";
@@ -375,6 +384,7 @@ export class Agent {
 		this.afterToolCall = opts.afterToolCall;
 		this.#telemetry = opts.telemetry;
 		this.#appendOnlyContext = opts.appendOnlyContext;
+		this.#transformProviderContext = opts.transformProviderContext;
 	}
 
 	/**
@@ -658,6 +668,10 @@ export class Agent {
 		this.#state.thinkingLevel = l;
 	}
 
+	setDisableReasoning(disabled: boolean) {
+		this.#state.disableReasoning = disabled;
+	}
+
 	setSteeringMode(mode: "all" | "one-at-a-time") {
 		this.#steeringMode = mode;
 	}
@@ -942,6 +956,7 @@ export class Agent {
 		const config: AgentLoopConfig = {
 			model,
 			reasoning,
+			disableReasoning: this.#state.disableReasoning,
 			temperature: this.#temperature,
 			topP: this.#topP,
 			topK: this.#topK,
@@ -961,6 +976,7 @@ export class Agent {
 			kimiApiFormat: this.#kimiApiFormat,
 			preferWebsockets: this.#preferWebsockets,
 			convertToLlm: this.#convertToLlm,
+			transformProviderContext: this.#transformProviderContext,
 			transformContext: this.#transformContext,
 			onPayload: this.#onPayload,
 			onResponse: this.#onResponse,
@@ -985,6 +1001,7 @@ export class Agent {
 			onHarmonyLeak: this.#onHarmonyLeak,
 			getToolChoice,
 			getReasoning: () => this.#state.thinkingLevel,
+			getDisableReasoning: () => this.#state.disableReasoning,
 			getSteeringMessages: async () => {
 				if (skipInitialSteeringPoll) {
 					skipInitialSteeringPoll = false;

package/src/compaction/compaction.ts

@@ -12,6 +12,7 @@ import {
 	type Message,
 	type MessageAttribution,
 	type Model,
+	type Tool,
 	type Usage,
 } from "@oh-my-pi/pi-ai";
 import { clampThinkingLevelForModel } from "@oh-my-pi/pi-catalog/model-thinking";
@@ -19,7 +20,7 @@ import { countTokens } from "@oh-my-pi/pi-natives";
 import { logger, prompt } from "@oh-my-pi/pi-utils";
 import { type AgentTelemetry, instrumentedCompleteSimple } from "../telemetry";
 import { ThinkingLevel } from "../thinking";
-import type { AgentMessage, AgentTool } from "../types";
+import type { AgentMessage } from "../types";
 import type { CompactionEntry, SessionEntry } from "./entries";
 import { type ConvertToLlm, convertToLlm, createBranchSummaryMessage, createCustomMessage } from "./messages";
 import {
@@ -690,7 +691,7 @@ export interface HandoffOptions {
 	/** Live agent system prompt — passed verbatim so providers hit the cached prefix. */
 	systemPrompt: string[];
 	/** Live agent tool list — same purpose. Forced to `toolChoice: "none"`. */
-	tools?: AgentTool<any>[];
+	tools?: Tool[];
 	customInstructions?: string;
 	convertToLlm?: ConvertToLlm;
 	initiatorOverride?: MessageAttribution;

package/src/types.ts

@@ -3,6 +3,7 @@ import type {
 	AssistantMessage,
 	AssistantMessageEvent,
 	AssistantMessageEventStream,
+	Context,
 	Effort,
 	ImageContent,
 	Message,
@@ -107,6 +108,13 @@ export interface AgentLoopConfig extends SimpleStreamOptions {
 	 */
 	transformContext?: (messages: AgentMessage[], signal?: AbortSignal) => Promise<AgentMessage[]>;
 
+	/**
+	 * Optional transform applied to the final provider context after conversion,
+	 * normalization, and append-only context handling, but before telemetry capture
+	 * and provider send.
+	 */
+	transformProviderContext?: (context: Context) => Context;
+
 	/**
 	 * Resolves an API key dynamically for each LLM call.
 	 *
@@ -210,6 +218,15 @@ export interface AgentLoopConfig extends SimpleStreamOptions {
 	 */
 	getReasoning?: () => Effort | undefined;
 
+	/**
+	 * Dynamic reasoning-disable override, resolved per LLM call. When set,
+	 * its return value overrides the static `disableReasoning` from
+	 * `SimpleStreamOptions` for that request. Pair with `getReasoning` so
+	 * mid-run transitions into and out of the explicit `off` state propagate
+	 * to the next provider call.
+	 */
+	getDisableReasoning?: () => boolean | undefined;
+
 	/**
 	 * Called after a tool call has been validated and is about to execute.
 	 *
@@ -358,6 +375,7 @@ export interface AgentState {
 	systemPrompt: string[];
 	model: Model;
 	thinkingLevel?: Effort;
+	disableReasoning?: boolean;
 	tools: AgentTool<any>[];
 	messages: AgentMessage[]; // Can include attachments + custom message types
 	isStreaming: boolean;