@oh-my-pi/pi-coding-agent 13.11.1 → 13.12.0

package/src/prompts/system/system-prompt.md

@@ -297,6 +297,7 @@ When a tool call fails, read the full error before doing anything else. When a f
 - You **MUST** exhaust tools/context/files first — explore.
 ## 7. Verification
 - Test everything rigorously → Future contributor cannot break behavior without failure. Prefer unit/e2e.
+- You **MUST NOT** rely on mocks — they invent behaviors that never happen in production and hide real bugs.
 - You **SHOULD** run only tests you added/modified unless asked otherwise.
 - You **MUST NOT** yield without proof when non-trivial work, self-assessment is deceptive: tests, linters, type checks, repro steps… exhaust all external verification.
 

package/src/sdk.ts

@@ -227,17 +227,17 @@ export {
 	BashTool,
 	// Tool classes and factories
 	BUILTIN_TOOLS,
-	HIDDEN_TOOLS,
 	createTools,
 	EditTool,
 	FindTool,
 	GrepTool,
+	HIDDEN_TOOLS,
 	loadSshTool,
 	PythonTool,
 	ReadTool,
 	ResolveTool,
-	WriteTool,
 	type ToolSession,
+	WriteTool,
 };
 
 // Helper Functions
@@ -702,19 +702,7 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 		}
 	}
 
-	// For subagent sessions using GitHub Copilot, add X-Initiator header
-	// to ensure proper billing (agent-initiated vs user-initiated)
 	const taskDepth = options.taskDepth ?? 0;
-	const forceCopilotAgentInitiator = taskDepth > 0;
-	if (forceCopilotAgentInitiator && model?.provider === "github-copilot") {
-		model = {
-			...model,
-			headers: {
-				...model.headers,
-				"X-Initiator": "agent",
-			},
-		};
-	}
 
 	let thinkingLevel = options.thinkingLevel;
 
@@ -1340,6 +1328,16 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 		if (!obfuscator?.hasSecrets()) return converted;
 		return obfuscateMessages(obfuscator, converted);
 	};
+	const transformContext = extensionRunner
+		? async (messages: AgentMessage[], _signal?: AbortSignal) => {
+				return await extensionRunner.emitContext(messages);
+			}
+		: undefined;
+	const onPayload = extensionRunner
+		? async (payload: unknown, _model?: Model) => {
+				return await extensionRunner.emitBeforeProviderRequest(payload);
+			}
+		: undefined;
 
 	const setToolUIContext = (uiContext: ExtensionUIContext, hasUI: boolean) => {
 		toolContextStore.setUIContext(uiContext, hasUI);
@@ -1362,17 +1360,9 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 			tools: initialTools,
 		},
 		convertToLlm: convertToLlmFinal,
-		onPayload: extensionRunner
-			? async (payload, _model) => {
-					return extensionRunner.emitBeforeProviderRequest(payload);
-				}
-			: undefined,
+		onPayload,
 		sessionId: sessionManager.getSessionId(),
-		transformContext: extensionRunner
-			? async messages => {
-					return extensionRunner.emitContext(messages);
-				}
-			: undefined,
+		transformContext,
 		steeringMode: settings.get("steeringMode") ?? "one-at-a-time",
 		followUpMode: settings.get("followUpMode") ?? "one-at-a-time",
 		interruptMode: settings.get("interruptMode") ?? "immediate",
@@ -1447,9 +1437,11 @@ export async function createAgentSession(options: CreateAgentSessionOptions = {}
 		skillsSettings: settings.getGroup("skills") as Required<SkillsSettings>,
 		modelRegistry,
 		toolRegistry,
+		transformContext,
+		onPayload,
+		convertToLlm: convertToLlmFinal,
 		rebuildSystemPrompt,
 		ttsrManager,
-		forceCopilotAgentInitiator,
 		obfuscator,
 		asyncJobManager,
 		pendingActionStore,

package/src/session/agent-session.ts

@@ -34,6 +34,7 @@ import type {
 	Model,
 	ProviderSessionState,
 	ServiceTier,
+	SimpleStreamOptions,
 	TextContent,
 	ToolCall,
 	ToolChoice,
@@ -104,6 +105,7 @@ import { outputMeta } from "../tools/output-meta";
 import { resolveToCwd } from "../tools/path-utils";
 import type { PendingActionStore } from "../tools/pending-action";
 import { getLatestTodoPhasesFromEntries, type TodoItem, type TodoPhase } from "../tools/todo-write";
+import { clampTimeout } from "../tools/tool-timeouts";
 import { parseCommandArgs } from "../utils/command-args";
 import { resolveFileDisplayMode } from "../utils/file-display-mode";
 import { extractFileMentions, generateFileMentionMessages } from "../utils/file-mentions";
@@ -126,6 +128,7 @@ import {
 	bashExecutionToText,
 	type CompactionSummaryMessage,
 	type CustomMessage,
+	convertToLlm,
 	type FileMentionMessage,
 	type HookMessage,
 	type PythonExecutionMessage,
@@ -145,6 +148,8 @@ export type AgentSessionEvent =
 			aborted: boolean;
 			willRetry: boolean;
 			errorMessage?: string;
+			/** True when compaction was skipped for a benign reason (no model, no candidates, nothing to compact). */
+			skipped?: boolean;
 	  }
 	| { type: "auto_retry_start"; attempt: number; maxAttempts: number; delayMs: number; errorMessage: string }
 	| { type: "auto_retry_end"; success: boolean; attempt: number; finalError?: string }
@@ -191,12 +196,16 @@ export interface AgentSessionConfig {
 	modelRegistry: ModelRegistry;
 	/** Tool registry for LSP and settings */
 	toolRegistry?: Map<string, AgentTool>;
+	/** Current session pre-LLM message transform pipeline */
+	transformContext?: (messages: AgentMessage[], signal?: AbortSignal) => AgentMessage[] | Promise<AgentMessage[]>;
+	/** Provider payload hook used by the active session request path */
+	onPayload?: SimpleStreamOptions["onPayload"];
+	/** Current session message-to-LLM conversion pipeline */
+	convertToLlm?: (messages: AgentMessage[]) => Message[] | Promise<Message[]>;
 	/** System prompt builder that can consider tool availability */
 	rebuildSystemPrompt?: (toolNames: string[], tools: Map<string, AgentTool>) => Promise<string>;
 	/** TTSR manager for time-traveling stream rules */
 	ttsrManager?: TtsrManager;
-	/** Force X-Initiator: agent for GitHub Copilot model selections in this session. */
-	forceCopilotAgentInitiator?: boolean;
 	/** Secret obfuscator for deobfuscating streaming edit content */
 	obfuscator?: SecretObfuscator;
 	/** Pending action store for preview/apply workflows */
@@ -381,9 +390,11 @@ export class AgentSession {
 
 	// Tool registry and prompt builder for extensions
 	#toolRegistry: Map<string, AgentTool>;
+	#transformContext: (messages: AgentMessage[], signal?: AbortSignal) => AgentMessage[] | Promise<AgentMessage[]>;
+	#onPayload: SimpleStreamOptions["onPayload"] | undefined;
+	#convertToLlm: (messages: AgentMessage[]) => Message[] | Promise<Message[]>;
 	#rebuildSystemPrompt: ((toolNames: string[], tools: Map<string, AgentTool>) => Promise<string>) | undefined;
 	#baseSystemPrompt: string;
-	#forceCopilotAgentInitiator = false;
 
 	// TTSR manager for time-traveling stream rules
 	#ttsrManager: TtsrManager | undefined = undefined;
@@ -425,10 +436,12 @@ export class AgentSession {
 		this.#skillsSettings = config.skillsSettings;
 		this.#modelRegistry = config.modelRegistry;
 		this.#toolRegistry = config.toolRegistry ?? new Map();
+		this.#transformContext = config.transformContext ?? (messages => messages);
+		this.#onPayload = config.onPayload;
+		this.#convertToLlm = config.convertToLlm ?? convertToLlm;
 		this.#rebuildSystemPrompt = config.rebuildSystemPrompt;
 		this.#baseSystemPrompt = this.agent.state.systemPrompt;
 		this.#ttsrManager = config.ttsrManager;
-		this.#forceCopilotAgentInitiator = config.forceCopilotAgentInitiator ?? false;
 		this.#obfuscator = config.obfuscator;
 		this.agent.providerSessionState = this.#providerSessionState;
 		this.#pendingActionStore = config.pendingActionStore;
@@ -580,7 +593,7 @@ export class AgentSession {
 
 					this.#addPendingTtsrInjections(matches);
 
-					if (this.#shouldInterruptForTtsrMatch(matchContext)) {
+					if (this.#shouldInterruptForTtsrMatch(matches, matchContext)) {
 						// Abort the stream immediately — do not gate on extension callbacks
 						this.#ttsrAbortPending = true;
 						this.#ensureTtsrResumePromise();
@@ -1014,18 +1027,17 @@ export class AgentSession {
 		return -1;
 	}
 
-	#shouldInterruptForTtsrMatch(matchContext: TtsrMatchContext): boolean {
-		const mode = this.#ttsrManager?.getSettings().interruptMode ?? "always";
-		if (mode === "never") {
-			return false;
-		}
-		if (mode === "prose-only") {
-			return matchContext.source === "text" || matchContext.source === "thinking";
-		}
-		if (mode === "tool-only") {
-			return matchContext.source === "tool";
+	#shouldInterruptForTtsrMatch(matches: Rule[], matchContext: TtsrMatchContext): boolean {
+		const globalMode = this.#ttsrManager?.getSettings().interruptMode ?? "always";
+		for (const rule of matches) {
+			const mode = rule.interruptMode ?? globalMode;
+			if (mode === "never") continue;
+			if (mode === "prose-only" && (matchContext.source === "text" || matchContext.source === "thinking"))
+				return true;
+			if (mode === "tool-only" && matchContext.source === "tool") return true;
+			if (mode === "always") return true;
 		}
-		return true;
+		return false;
 	}
 
 	#queueDeferredTtsrInjectionIfNeeded(assistantMsg: AssistantMessage): void {
@@ -1442,6 +1454,7 @@ export class AgentSession {
 				aborted: event.aborted,
 				willRetry: event.willRetry,
 				errorMessage: event.errorMessage,
+				skipped: event.skipped,
 			});
 		} else if (event.type === "auto_retry_start") {
 			await this.#extensionRunner.emit({
@@ -1526,7 +1539,7 @@ export class AgentSession {
 		if (drained === false && deliveryState) {
 			logger.warn("Async job completion deliveries still pending during dispose", { ...deliveryState });
 		}
-		await this.sessionManager.flush();
+		await this.sessionManager.close();
 		for (const state of this.#providerSessionState.values()) {
 			state.close();
 		}
@@ -1551,19 +1564,6 @@ export class AgentSession {
 		return this.agent.state.model;
 	}
 
-	#applySessionModelOverrides(model: Model): Model {
-		if (!this.#forceCopilotAgentInitiator || model.provider !== "github-copilot") {
-			return model;
-		}
-		return {
-			...model,
-			headers: {
-				...model.headers,
-				"X-Initiator": "agent",
-			},
-		};
-	}
-
 	/** Current thinking level */
 	get thinkingLevel(): ThinkingLevel | undefined {
 		return this.#thinkingLevel;
@@ -1714,6 +1714,31 @@ export class AgentSession {
 		return this.agent.state.messages;
 	}
 
+	/** Convert session messages using the same pre-LLM pipeline as the active session. */
+	async convertMessagesToLlm(messages: AgentMessage[], signal?: AbortSignal): Promise<Message[]> {
+		const transformedMessages = await this.#transformContext(messages, signal);
+		return await this.#convertToLlm(transformedMessages);
+	}
+
+	/** Apply session-level stream hooks to a direct side request. */
+	prepareSimpleStreamOptions(options: SimpleStreamOptions): SimpleStreamOptions {
+		if (!this.#onPayload) return options;
+		if (!options.onPayload) {
+			return { ...options, onPayload: this.#onPayload };
+		}
+		const sessionOnPayload = this.#onPayload;
+		const requestOnPayload = options.onPayload;
+		return {
+			...options,
+			onPayload: async (payload, model) => {
+				const sessionPayload = await sessionOnPayload(payload, model);
+				const sessionResolvedPayload = sessionPayload ?? payload;
+				const requestPayload = await requestOnPayload(sessionResolvedPayload, model);
+				return requestPayload ?? sessionResolvedPayload;
+			},
+		};
+	}
+
 	/** Current steering mode */
 	get steeringMode(): "all" | "one-at-a-time" {
 		return this.agent.getSteeringMode();
@@ -2905,14 +2930,12 @@ export class AgentSession {
 	 * Cycle to next thinking level.
 	 * @returns New level, or undefined if model doesn't support thinking
 	 */
-	cycleThinkingLevel(): Effort | undefined {
+	cycleThinkingLevel(): ThinkingLevel | undefined {
 		if (!this.model?.reasoning) return undefined;
 
-		const levels = this.getAvailableThinkingLevels();
-		const currentIndex =
-			this.thinkingLevel && this.thinkingLevel !== ThinkingLevel.Off && this.thinkingLevel !== ThinkingLevel.Inherit
-				? levels.indexOf(this.thinkingLevel)
-				: -1;
+		const levels = [ThinkingLevel.Off, ...this.getAvailableThinkingLevels()];
+		const currentLevel = this.thinkingLevel === ThinkingLevel.Inherit ? ThinkingLevel.Off : this.thinkingLevel;
+		const currentIndex = currentLevel ? levels.indexOf(currentLevel) : -1;
 		const nextIndex = (currentIndex + 1) % levels.length;
 		const nextLevel = levels[nextIndex];
 		if (!nextLevel) return undefined;
@@ -3685,7 +3708,7 @@ export class AgentSession {
 		if (currentModel) {
 			this.#closeProviderSessionsForModelSwitch(currentModel, model);
 		}
-		this.agent.setModel(this.#applySessionModelOverrides(model));
+		this.agent.setModel(model);
 	}
 
 	#closeCodexProviderSessionsForHistoryRewrite(): void {
@@ -3853,6 +3876,7 @@ export class AgentSession {
 					result: undefined,
 					aborted: false,
 					willRetry: false,
+					skipped: true,
 				});
 				return;
 			}
@@ -3865,6 +3889,7 @@ export class AgentSession {
 					result: undefined,
 					aborted: false,
 					willRetry: false,
+					skipped: true,
 				});
 				return;
 			}
@@ -3879,6 +3904,7 @@ export class AgentSession {
 					result: undefined,
 					aborted: false,
 					willRetry: false,
+					skipped: true,
 				});
 				if (!willRetry && this.agent.hasQueuedMessages()) {
 					this.#scheduleAgentContinue({
@@ -3966,6 +3992,7 @@ export class AgentSession {
 								promptOverride: hookPrompt,
 								extraContext: hookContext,
 								remoteInstructions: this.#baseSystemPrompt,
+								initiatorOverride: "agent",
 							});
 							break;
 						} catch (error) {
@@ -4429,6 +4456,7 @@ export class AgentSession {
 				onChunk,
 				signal: this.#bashAbortController.signal,
 				sessionKey: this.sessionId,
+				timeout: clampTimeout("bash") * 1000,
 			});
 
 			this.recordBashResult(command, result, options);

package/src/session/blob-store.ts

@@ -74,6 +74,21 @@ export function parseBlobRef(data: string): string | null {
 	return data.slice(BLOB_PREFIX.length);
 }
 
+/** Identify provider transport image data URLs so persistence can externalize and restore them losslessly. */
+export function isImageDataUrl(data: string): boolean {
+	return data.startsWith("data:image/") && data.includes(";base64,");
+}
+
+/**
+ * Externalize a provider image data URL to the blob store, returning a blob reference.
+ * The full data URL string is preserved so transport-native history can be reconstructed on resume.
+ */
+export async function externalizeImageDataUrl(blobStore: BlobStore, dataUrl: string): Promise<string> {
+	if (isBlobRef(dataUrl)) return dataUrl;
+	const { ref } = await blobStore.put(Buffer.from(dataUrl, "utf8"));
+	return ref;
+}
+
 /**
  * Externalize an image's base64 data to the blob store, returning a blob reference.
  * If the data is already a blob reference, returns it unchanged.
@@ -85,6 +100,23 @@ export async function externalizeImageData(blobStore: BlobStore, base64Data: str
 	return ref;
 }
 
+/**
+ * Resolve an externalized provider image data URL back to its original string.
+ * If the data is not a blob reference, returns it unchanged.
+ * If the blob is missing, logs a warning and returns the reference as-is.
+ */
+export async function resolveImageDataUrl(blobStore: BlobStore, data: string): Promise<string> {
+	const hash = parseBlobRef(data);
+	if (!hash) return data;
+
+	const buffer = await blobStore.get(hash);
+	if (!buffer) {
+		logger.warn("Blob not found for persisted image data URL", { hash });
+		return data;
+	}
+	return buffer.toString("utf8");
+}
+
 /**
  * Resolve a blob reference back to base64 data.
  * If the data is not a blob reference, returns it unchanged.

package/src/session/compaction/compaction.ts

@@ -5,7 +5,14 @@
  * and after compaction the session is reloaded.
  */
 import type { AgentMessage } from "@oh-my-pi/pi-agent-core";
-import { type AssistantMessage, completeSimple, Effort, type Model, type Usage } from "@oh-my-pi/pi-ai";
+import {
+	type AssistantMessage,
+	completeSimple,
+	Effort,
+	type MessageAttribution,
+	type Model,
+	type Usage,
+} from "@oh-my-pi/pi-ai";
 import {
 	CODEX_BASE_URL,
 	getCodexAccountId,
@@ -935,6 +942,7 @@ export interface SummaryOptions {
 	extraContext?: string[];
 	remoteEndpoint?: string;
 	remoteInstructions?: string;
+	initiatorOverride?: MessageAttribution;
 }
 
 export async function generateSummary(
@@ -990,7 +998,7 @@ export async function generateSummary(
 	const response = await completeSimple(
 		model,
 		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
-		{ maxTokens, signal, apiKey, reasoning: Effort.High },
+		{ maxTokens, signal, apiKey, reasoning: Effort.High, initiatorOverride: options?.initiatorOverride },
 	);
 
 	if (response.stopReason === "error") {
@@ -1039,7 +1047,7 @@ async function generateShortSummary(
 			systemPrompt: SUMMARIZATION_SYSTEM_PROMPT,
 			messages: [{ role: "user", content: [{ type: "text", text: promptText }], timestamp: Date.now() }],
 		},
-		{ maxTokens, signal, apiKey, reasoning: Effort.High },
+		{ maxTokens, signal, apiKey, reasoning: Effort.High, initiatorOverride: options?.initiatorOverride },
 	);
 
 	if (response.stopReason === "error") {
@@ -1218,6 +1226,7 @@ export async function compact(
 		extraContext: options?.extraContext,
 		remoteEndpoint: settings.remoteEnabled === false ? undefined : settings.remoteEndpoint,
 		remoteInstructions: options?.remoteInstructions,
+		initiatorOverride: options?.initiatorOverride,
 	};
 
 	let preserveData = withOpenAiRemoteCompactionPreserveData(previousPreserveData, undefined);
@@ -1266,7 +1275,14 @@ export async function compact(
 						summaryOptions,
 					)
 				: Promise.resolve("No prior history."),
-			generateTurnPrefixSummary(turnPrefixMessages, model, settings.reserveTokens, apiKey, signal),
+			generateTurnPrefixSummary(
+				turnPrefixMessages,
+				model,
+				settings.reserveTokens,
+				apiKey,
+				signal,
+				summaryOptions.initiatorOverride,
+			),
 		]);
 		// Merge into single summary
 		summary = `${historyResult}\n\n---\n\n**Turn Context (split turn):**\n\n${turnPrefixResult}`;
@@ -1297,7 +1313,11 @@ export async function compact(
 		settings.reserveTokens,
 		apiKey,
 		signal,
-		{ extraContext: options?.extraContext, remoteEndpoint: summaryOptions.remoteEndpoint },
+		{
+			extraContext: options?.extraContext,
+			remoteEndpoint: summaryOptions.remoteEndpoint,
+			initiatorOverride: summaryOptions.initiatorOverride,
+		},
 	);
 
 	// Compute file lists and append to summary
@@ -1327,6 +1347,7 @@ async function generateTurnPrefixSummary(
 	reserveTokens: number,
 	apiKey: string,
 	signal?: AbortSignal,
+	initiatorOverride?: MessageAttribution,
 ): Promise<string> {
 	const maxTokens = Math.floor(0.5 * reserveTokens); // Smaller budget for turn prefix
 
@@ -1344,7 +1365,7 @@ async function generateTurnPrefixSummary(
 	const response = await completeSimple(
 		model,
 		{ systemPrompt: SUMMARIZATION_SYSTEM_PROMPT, messages: summarizationMessages },
-		{ maxTokens, signal, apiKey, reasoning: Effort.High },
+		{ maxTokens, signal, apiKey, reasoning: Effort.High, initiatorOverride },
 	);
 
 	if (response.stopReason === "error") {

package/src/session/history-storage.ts

@@ -1,7 +1,7 @@
 import { Database, type Statement } from "bun:sqlite";
 import * as fs from "node:fs";
 import * as path from "node:path";
-import { getAgentDir, logger } from "@oh-my-pi/pi-utils";
+import { getHistoryDbPath, logger } from "@oh-my-pi/pi-utils";
 
 export interface HistoryEntry {
 	id: number;
@@ -78,7 +78,7 @@ END;
 		this.#lastPromptCache = last?.prompt ?? null;
 	}
 
-	static open(dbPath: string = path.join(getAgentDir(), "history.db")): HistoryStorage {
+	static open(dbPath: string = getHistoryDbPath()): HistoryStorage {
 		if (!HistoryStorage.#instance) {
 			HistoryStorage.#instance = new HistoryStorage(dbPath);
 		}