@oh-my-pi/pi-ai 16.3.0 → 16.3.2

package/CHANGELOG.md

@@ -2,6 +2,12 @@
 
 ## [Unreleased]
 
+## [16.3.1] - 2026-07-02
+
+### Changed
+
+- Removed automated injection of reasoning suppression prompts in OpenAI responses
+
 ## [16.3.0] - 2026-07-02
 
 ### Added

package/dist/types/providers/openai-responses.d.ts

@@ -64,8 +64,8 @@ interface OpenAIResponsesProviderSessionState extends ProviderSessionState, Open
 }
 interface OpenAIResponsesChainState {
     /**
-     * Wire params of the last successful turn, with per-turn trailing
-     * scaffolding stripped from `input` (never carries previous_response_id).
+     * Wire params of the last successful turn; never carries
+     * `previous_response_id`.
      */
     lastParams?: OpenAIResponsesSamplingParams;
     lastResponseId?: string;
@@ -108,7 +108,6 @@ type OpenAIResponsesSamplingParams = ResponseCreateParamsStreaming & {
 export declare const streamOpenAIResponses: StreamFunction<"openai-responses">;
 export declare function buildParams(model: Model<"openai-responses">, context: Context, options: OpenAIResponsesOptions | undefined, providerSessionState: OpenAIResponsesProviderSessionState | undefined, strictToolsScope?: OpenAIStrictToolsScope, disableStrictToolsOverride?: boolean): {
     params: OpenAIResponsesSamplingParams;
-    trailingScaffoldingItems: number;
     strictToolsApplied: boolean;
 };
 /**

package/dist/types/providers/openai-shared.d.ts

@@ -465,12 +465,12 @@ export interface ApplyResponsesCompatPolicyOptions {
     reasoningSummary?: "auto" | "detailed" | "concise" | null;
     mapEffort?: (effort: string) => string;
 }
-export declare function applyResponsesCompatPolicy<P extends ResponseCreateParamsStreaming>(params: P, messages: ResponseInput, policy: OpenAICompatPolicy, options: ApplyResponsesCompatPolicyOptions | undefined): number;
+export declare function applyResponsesCompatPolicy<P extends ResponseCreateParamsStreaming>(params: P, policy: OpenAICompatPolicy, options: ApplyResponsesCompatPolicyOptions | undefined): void;
 /**
  * Apply reasoning-related Responses parameters. Default behavior comes from
  * catalog compat; include/omit arguments are explicit adapter-wrapper overrides.
  */
-export declare function applyResponsesReasoningParams<P extends ResponseCreateParamsStreaming>(params: P, model: Model<"openai-responses" | "azure-openai-responses" | "openai-codex-responses">, options: ReasoningOptions | undefined, messages: ResponseInput, mapEffort?: (effort: string) => string, includeEncryptedReasoning?: boolean, omitReasoningEffort?: boolean): number;
+export declare function applyResponsesReasoningParams<P extends ResponseCreateParamsStreaming>(params: P, model: Model<"openai-responses" | "azure-openai-responses" | "openai-codex-responses">, options: ReasoningOptions | undefined, mapEffort?: (effort: string) => string, includeEncryptedReasoning?: boolean, omitReasoningEffort?: boolean): void;
 /** Populate `output.usage` from a Responses-API `response.usage` payload. Does not invoke `calculateCost`. */
 export declare function populateResponsesUsageFromResponse(output: AssistantMessage, usage: {
     input_tokens?: number | null;

package/package.json

@@ -1,7 +1,7 @@
 {
 	"type": "module",
 	"name": "@oh-my-pi/pi-ai",
-	"version": "16.3.0",
+	"version": "16.3.2",
 	"description": "Unified LLM API with automatic model discovery and provider configuration",
 	"homepage": "https://omp.sh",
 	"author": "Can Boluk",
@@ -38,9 +38,9 @@
 	},
 	"dependencies": {
 		"@bufbuild/protobuf": "^2.12.0",
-		"@oh-my-pi/pi-catalog": "16.3.0",
-		"@oh-my-pi/pi-utils": "16.3.0",
-		"@oh-my-pi/pi-wire": "16.3.0",
+		"@oh-my-pi/pi-catalog": "16.3.2",
+		"@oh-my-pi/pi-utils": "16.3.2",
+		"@oh-my-pi/pi-wire": "16.3.2",
 		"arktype": "^2.2.0",
 		"zod": "^4"
 	},

package/src/providers/azure-openai-responses.ts

@@ -376,7 +376,7 @@ function buildParams(
 		}
 	}
 
-	applyResponsesReasoningParams(params, model, options, messages);
+	applyResponsesReasoningParams(params, model, options);
 
 	return params;
 }

package/src/providers/openai-responses.ts

@@ -166,8 +166,8 @@ interface OpenAIResponsesProviderSessionState
 
 interface OpenAIResponsesChainState {
 	/**
-	 * Wire params of the last successful turn, with per-turn trailing
-	 * scaffolding stripped from `input` (never carries previous_response_id).
+	 * Wire params of the last successful turn; never carries
+	 * `previous_response_id`.
 	 */
 	lastParams?: OpenAIResponsesSamplingParams;
 	lastResponseId?: string;
@@ -256,30 +256,19 @@ interface OpenAIResponsesChainedParams {
  * (same options, strict history prefix), chain via `previous_response_id` +
  * delta-only `input`; otherwise break the chain and replay the full transcript.
  *
- * The prefix check runs on the wire form of the conversation arguments alone:
- * per-turn trailing scaffolding is excluded from both sides and re-appended to
- * the delta, so a decoration that trails every request can never masquerade as
- * a history mutation.
+ * The prefix check runs on the wire form of the conversation arguments, so
+ * history mutations or option changes force a full replay.
  */
 function buildOpenAIResponsesChainedParams(
 	params: OpenAIResponsesSamplingParams,
-	trailingScaffoldingItems: number,
 	chain: OpenAIResponsesChainState,
 ): OpenAIResponsesChainedParams {
-	const historyParams =
-		trailingScaffoldingItems > 0 && Array.isArray(params.input)
-			? { ...params, input: params.input.slice(0, params.input.length - trailingScaffoldingItems) }
-			: params;
 	const deltaInput = chain.canAppend
-		? buildResponsesDeltaInput(chain.lastParams, chain.lastResponseItems, historyParams)
+		? buildResponsesDeltaInput(chain.lastParams, chain.lastResponseItems, params)
 		: null;
 	if (deltaInput && deltaInput.length > 0 && chain.lastResponseId) {
-		const scaffolding =
-			historyParams !== params && Array.isArray(params.input)
-				? params.input.slice(params.input.length - trailingScaffoldingItems)
-				: [];
 		return {
-			params: { ...params, previous_response_id: chain.lastResponseId, input: [...deltaInput, ...scaffolding] },
+			params: { ...params, previous_response_id: chain.lastResponseId, input: deltaInput },
 			previousResponseId: chain.lastResponseId,
 		};
 	}
@@ -416,9 +405,7 @@ const streamOpenAIResponsesOnce = (
 			const strictToolsScope = getOpenAIStrictToolsScope(model, baseUrl);
 			const builtParams = buildParams(model, context, options, providerSessionState, strictToolsScope);
 			const params = builtParams.params;
-			const { trailingScaffoldingItems } = builtParams;
 			let activeParams = params;
-			let activeTrailingScaffoldingItems = trailingScaffoldingItems;
 			const resolvedBaseUrl = (baseUrl ?? "https://api.openai.com/v1").replace(/\/+$/, "");
 			const requestReasoningEffortFallbacks = new Map<string, OpenAIReasoningEffortFallback>();
 			const attemptedReasoningEffortFallbacks = new Set<string>();
@@ -448,9 +435,7 @@ const streamOpenAIResponsesOnce = (
 			}
 			applyReasoningEffortFallbackForRequest(params);
 			let chained: OpenAIResponsesChainedParams =
-				chainState && !chainState.disabled
-					? buildOpenAIResponsesChainedParams(params, trailingScaffoldingItems, chainState)
-					: { params };
+				chainState && !chainState.disabled ? buildOpenAIResponsesChainedParams(params, chainState) : { params };
 			sentPreviousResponseId = chained.previousResponseId;
 			const idleTimeoutMs =
 				options?.streamIdleTimeoutMs ?? getOpenAIStreamIdleTimeoutMs(model.compat.streamIdleTimeoutMs);
@@ -589,11 +574,7 @@ const streamOpenAIResponsesOnce = (
 						if (chainState && !chainState.disabled) fallbackParams.store = true;
 						let fallbackChained: OpenAIResponsesChainedParams =
 							chainState && !chainState.disabled
-								? buildOpenAIResponsesChainedParams(
-										fallbackParams,
-										fallbackBuilt.trailingScaffoldingItems,
-										chainState,
-									)
+								? buildOpenAIResponsesChainedParams(fallbackParams, chainState)
 								: { params: fallbackParams };
 						sentPreviousResponseId = fallbackChained.previousResponseId;
 						fallbackChained = {
@@ -603,7 +584,6 @@ const streamOpenAIResponsesOnce = (
 						chained = fallbackChained;
 						rawRequestDump.body = chained.params;
 						activeParams = fallbackParams;
-						activeTrailingScaffoldingItems = fallbackBuilt.trailingScaffoldingItems;
 						activeStrictToolsApplied = fallbackBuilt.strictToolsApplied;
 						continue;
 					}
@@ -646,7 +626,6 @@ const streamOpenAIResponsesOnce = (
 					chained = { params: retryParams };
 					rawRequestDump.body = retryParams;
 					activeParams = currentParams;
-					activeTrailingScaffoldingItems = currentBuilt.trailingScaffoldingItems;
 					activeStrictToolsApplied = currentBuilt.strictToolsApplied;
 				}
 			}
@@ -710,14 +689,7 @@ const streamOpenAIResponsesOnce = (
 			output.providerPayload = createOpenAIResponsesHistoryPayload(model.provider, nativeOutputItems);
 			if (providerSessionState) providerSessionState.nativeHistoryReplayWarmed = true;
 			if (chainState) {
-				chainState.lastParams = structuredCloneJSON(
-					activeTrailingScaffoldingItems > 0 && Array.isArray(activeParams.input)
-						? {
-								...activeParams,
-								input: activeParams.input.slice(0, activeParams.input.length - activeTrailingScaffoldingItems),
-							}
-						: activeParams,
-				);
+				chainState.lastParams = structuredCloneJSON(activeParams);
 				if (output.responseId) {
 					chainState.lastResponseId = output.responseId;
 					chainState.lastResponseItems = sanitizeOpenAIResponsesHistoryItemsForReplay(
@@ -790,7 +762,7 @@ export function buildParams(
 	providerSessionState: OpenAIResponsesProviderSessionState | undefined,
 	strictToolsScope?: OpenAIStrictToolsScope,
 	disableStrictToolsOverride = false,
-): { params: OpenAIResponsesSamplingParams; trailingScaffoldingItems: number; strictToolsApplied: boolean } {
+): { params: OpenAIResponsesSamplingParams; strictToolsApplied: boolean } {
 	const policy = resolveOpenAICompatPolicy(model, {
 		endpoint: "responses",
 		reasoning: options?.reasoning,
@@ -914,7 +886,7 @@ export function buildParams(
 		filterReasoningHistory: options?.filterReasoningHistory,
 		omitReasoningEffort: options?.omitReasoningEffort,
 	});
-	const trailingScaffoldingItems = applyResponsesCompatPolicy(params, messages, reasoningPolicy, {
+	applyResponsesCompatPolicy(params, reasoningPolicy, {
 		reasoningSummary: options?.reasoningSummary,
 		mapEffort: effort =>
 			model.compat.reasoningEffortMap?.[effort as NonNullable<OpenAIResponsesOptions["reasoning"]>] ??
@@ -926,7 +898,7 @@ export function buildParams(
 
 	applyOpenAIExtraBody(params, options?.extraBody);
 
-	return { params, trailingScaffoldingItems, strictToolsApplied };
+	return { params, strictToolsApplied };
 }
 
 /**

package/src/providers/openai-shared.ts

@@ -76,7 +76,6 @@ import {
 } from "./github-copilot-headers";
 import type { ChatCompletionCreateParamsStreaming } from "./openai-chat-wire";
 import type { InputItem } from "./openai-codex/request-transformer";
-import responsesReasoningSuppressionPrompt from "./openai-responses-reasoning-suppression.md" with { type: "text" };
 import type {
 	ResponseContentPartAddedEvent,
 	ResponseCreateParamsStreaming,
@@ -2404,8 +2403,6 @@ export function applyCommonResponsesSamplingParams<P extends CommonResponsesPara
 	applyOpenAIServiceTier(params, options?.serviceTier, model.provider);
 }
 
-const RESPONSES_REASONING_SUPPRESSION_PROMPT = responsesReasoningSuppressionPrompt.trim();
-
 type ReasoningOptions = {
 	reasoning?: string;
 	reasoningSummary?: "auto" | "detailed" | "concise" | null;
@@ -2420,12 +2417,11 @@ export interface ApplyResponsesCompatPolicyOptions {
 
 export function applyResponsesCompatPolicy<P extends ResponseCreateParamsStreaming>(
 	params: P,
-	messages: ResponseInput,
 	policy: OpenAICompatPolicy,
 	options: ApplyResponsesCompatPolicyOptions | undefined,
-): number {
+): void {
 	const reasoning = policy.reasoning;
-	if (!reasoning.modelSupported) return 0;
+	if (!reasoning.modelSupported) return;
 	if (reasoning.includeEncryptedReasoning) {
 		const include = params.include ?? [];
 		if (!include.includes("reasoning.encrypted_content")) include.push("reasoning.encrypted_content");
@@ -2435,7 +2431,7 @@ export function applyResponsesCompatPolicy<P extends ResponseCreateParamsStreami
 	if (reasoning.disabled) {
 		if (reasoning.disableMode === "openrouter-enabled-false") {
 			params.reasoning = { enabled: false } as P["reasoning"];
-			return 0;
+			return;
 		}
 		if (
 			reasoning.disableMode === "lowest-effort" &&
@@ -2445,16 +2441,9 @@ export function applyResponsesCompatPolicy<P extends ResponseCreateParamsStreami
 			type ReasoningParam = NonNullable<ResponseCreateParamsStreaming["reasoning"]>;
 			params.reasoning = { effort: reasoning.wireEffort as ReasoningParam["effort"] } as P["reasoning"] &
 				ReasoningParam;
-			return 0;
-		}
-		if (policy.compat.requiresReasoningSuppressionPrompt && reasoning.requestedEffort === undefined) {
-			messages.push({
-				role: "developer",
-				content: [{ type: "input_text", text: RESPONSES_REASONING_SUPPRESSION_PROMPT }],
-			});
-			return 1;
+			return;
 		}
-		return 0;
+		return;
 	}
 
 	if (reasoning.requestedEffort !== undefined || options?.reasoningSummary !== undefined) {
@@ -2463,7 +2452,7 @@ export function applyResponsesCompatPolicy<P extends ResponseCreateParamsStreami
 				type ReasoningParam = NonNullable<ResponseCreateParamsStreaming["reasoning"]>;
 				params.reasoning = { summary: options.reasoningSummary || "auto" } as P["reasoning"] & ReasoningParam;
 			}
-			return 0;
+			return;
 		}
 
 		const requested = reasoning.requestedEffort ?? "medium";
@@ -2476,17 +2465,8 @@ export function applyResponsesCompatPolicy<P extends ResponseCreateParamsStreami
 			reasoningParams.summary = options?.reasoningSummary || "auto";
 		}
 		params.reasoning = reasoningParams as P["reasoning"];
-		return 0;
-	}
-
-	if (policy.compat.requiresReasoningSuppressionPrompt) {
-		messages.push({
-			role: "developer",
-			content: [{ type: "input_text", text: RESPONSES_REASONING_SUPPRESSION_PROMPT }],
-		});
-		return 1;
+		return;
 	}
-	return 0;
 }
 
 /**
@@ -2497,14 +2477,12 @@ export function applyResponsesReasoningParams<P extends ResponseCreateParamsStre
 	params: P,
 	model: Model<"openai-responses" | "azure-openai-responses" | "openai-codex-responses">,
 	options: ReasoningOptions | undefined,
-	messages: ResponseInput,
 	mapEffort?: (effort: string) => string,
 	includeEncryptedReasoning?: boolean,
 	omitReasoningEffort?: boolean,
-): number {
+): void {
 	return applyResponsesCompatPolicy(
 		params,
-		messages,
 		resolveOpenAICompatPolicy(model, {
 			endpoint: "responses",
 			reasoning: options?.reasoning,

package/src/providers/openai-responses-reasoning-suppression.md

@@ -1 +0,0 @@
-Keep internal reasoning brief. Continue following the task and use tools normally.