pi-openai-toolkit 0.9.0 → 0.11.0

package/src/runtime.ts

@@ -54,6 +54,12 @@ export type ResponsesRuntime = {
 	headers?: ProviderHeaders;
 	responsesPath: string;
 	responsesUrl: string;
+	/**
+	 * Conversation identity for Codex prompt-cache affinity headers. Synthetic
+	 * compaction/review requests must present the same id the live transcript
+	 * requests used or the backend treats them as a new conversation.
+	 */
+	sessionId?: string;
 	currentModel: RuntimeModel;
 };
 
@@ -229,6 +235,15 @@ export function getRuntimeModelDescriptor(model: RuntimeModel | undefined): {
 	};
 }
 
+/** Best-effort conversation id; header emission then simply skips the affinity headers. */
+export function resolveRuntimeSessionId(ctx: ExtensionContext): string | undefined {
+	try {
+		return ctx.sessionManager.getSessionId();
+	} catch {
+		return undefined;
+	}
+}
+
 async function resolveNativeCompactionEnvironmentForModel(
 	ctx: ExtensionContext,
 	currentModel: RuntimeModel | undefined,
@@ -336,6 +351,7 @@ async function resolveNativeCompactionEnvironmentForModel(
 			responsesUrl: buildResponsesUrl(baseUrl, descriptor.api),
 			compactPath: buildCompactPath(descriptor.api),
 			compactUrl: buildCompactUrl(baseUrl, descriptor.api),
+			sessionId: resolveRuntimeSessionId(ctx),
 			payload: requestPayload,
 			currentModel,
 		},
@@ -361,6 +377,7 @@ export async function resolveResponsesEnvironment(
 			headers: runtime.headers,
 			responsesPath: runtime.responsesPath,
 			responsesUrl: runtime.responsesUrl,
+			sessionId: runtime.sessionId,
 			currentModel: runtime.currentModel,
 		},
 	};

package/src/types.ts

@@ -97,6 +97,12 @@ export type ImageGenerationConfig = {
 	enabled: boolean;
 };
 
+export type CodexAstraConfig = {
+	enabled: boolean;
+	/** Exact provider/model keys that receive the stable-effort `configuration_update` rewrite. */
+	models: string[];
+};
+
 /** `side-effect` reviews bash/write/edit plus extras; `all` reviews every tool call. */
 export type AutoModeGate = "side-effect" | "all";
 
@@ -111,16 +117,58 @@ export type AutoModeConfig = {
 	extraTools: string[];
 	/** Reviewer deadline in ms; expiry degrades to human confirmation or fail-closed. */
 	timeoutMs: number;
+	/** Give the reviewer a bounded transcript so it can judge user authorization. */
+	transcript: boolean;
+	/** Let the reviewer spend read-only tool calls establishing facts before deciding. */
+	evidenceTools: boolean;
+	maxEvidenceRounds: number;
+	/** Non-blocking trajectory pre-scorer that can satisfy ordinary gated calls. */
+	classifier: AutoModeClassifierConfig;
+	/** End a turn that keeps getting denied instead of negotiating forever. */
+	circuitBreaker: AutoModeCircuitBreakerConfig;
+};
+
+export type AutoModeClassifierConfig = {
+	enabled: boolean;
+	/** "provider/model-id"; falls back to `autoMode.reviewerModel` when unset. */
+	model?: string;
+	timeoutMs: number;
+	/** How many tool calls may elapse between scoring and use before a score is stale. */
+	maxLag: number;
+};
+
+export type AutoModeCircuitBreakerConfig = {
+	/** Consecutive reviewer denials in one turn that end the turn. 0 disables. */
+	consecutiveDenials: number;
+	/** Denials inside `windowSize` reviewed calls that end the turn. 0 disables. */
+	recentDenials: number;
+	windowSize: number;
 };
 
 export const REVIEWER_TIMEOUT_MIN_MS = 1_000;
 export const REVIEWER_TIMEOUT_MAX_MS = 120_000;
 
+/** Read-only investigation rounds the reviewer may spend before it must answer. */
+export const EVIDENCE_ROUNDS_MIN = 0;
+export const EVIDENCE_ROUNDS_MAX = 8;
+
+export const CLASSIFIER_MAX_LAG_MIN = 0;
+export const CLASSIFIER_MAX_LAG_MAX = 20;
+
+export const DEFAULT_CLASSIFIER_TIMEOUT_MS = 15_000;
+
+/** Denial counts are small integers; 0 disables the matching trigger. */
+export const BREAKER_LIMIT_MIN = 0;
+export const BREAKER_LIMIT_MAX = 100;
+export const BREAKER_WINDOW_MIN = 1;
+export const BREAKER_WINDOW_MAX = 200;
+
 export type ToolkitConfig = {
 	compaction: CompactionConfig;
 	webSearch: WebSearchConfig;
 	imageGeneration: ImageGenerationConfig;
 	autoMode: AutoModeConfig;
+	codexAstra: CodexAstraConfig;
 };
 
 export type LoadedToolkitConfig = {
@@ -483,6 +531,11 @@ export const DEFAULT_IMAGE_GENERATION_CONFIG: ImageGenerationConfig = {
 	enabled: false,
 };
 
+export const DEFAULT_CODEX_ASTRA_CONFIG: CodexAstraConfig = {
+	enabled: false,
+	models: [],
+};
+
 export const DEFAULT_AUTO_MODE_CONFIG: AutoModeConfig = {
 	enabled: true,
 	models: [],
@@ -490,6 +543,20 @@ export const DEFAULT_AUTO_MODE_CONFIG: AutoModeConfig = {
 	gate: "side-effect",
 	extraTools: [],
 	timeoutMs: 30_000,
+	transcript: true,
+	evidenceTools: true,
+	maxEvidenceRounds: 3,
+	classifier: {
+		enabled: false,
+		model: undefined,
+		timeoutMs: DEFAULT_CLASSIFIER_TIMEOUT_MS,
+		maxLag: 2,
+	},
+	circuitBreaker: {
+		consecutiveDenials: 3,
+		recentDenials: 10,
+		windowSize: 50,
+	},
 };
 
 export const DEFAULT_TOOLKIT_CONFIG: ToolkitConfig = {
@@ -497,4 +564,5 @@ export const DEFAULT_TOOLKIT_CONFIG: ToolkitConfig = {
 	webSearch: DEFAULT_WEB_SEARCH_CONFIG,
 	imageGeneration: DEFAULT_IMAGE_GENERATION_CONFIG,
 	autoMode: DEFAULT_AUTO_MODE_CONFIG,
+	codexAstra: DEFAULT_CODEX_ASTRA_CONFIG,
 };