pi-harness-runtime 0.10.7 → 0.10.11

package/harness/forked-summarizer.ts

@@ -1,304 +0,0 @@
-/**
- * Forked Summarizer — RFC-0028
- *
- * Summarizes old messages via a separate LLM call (forked context).
- * This prevents the summarization request from hitting the same
- * context-too-long error that triggered compaction.
- */
-
-import type {
-	CompactableMessage,
-	CompactTriggerReason,
-	InvokeResult,
-} from "../packages/types/src/runtime-types.js";
-
-// --- Types ------------------------------------------------------------------
-
-export interface SummarizerConfig {
-	/** Model to use for summarization */
-	model: string;
-	/** Provider for API calls */
-	provider?: string;
-	/** Max tokens for summary output */
-	maxSummaryTokens: number;
-	/** Custom system prompt */
-	systemPrompt?: string;
-	/** Focus of summarization */
-	focusOn?: "work_done" | "decisions" | "all";
-}
-
-export interface SummarizerResult {
-	summary: string;
-	droppedCount: number;
-	tokensUsed?: number;
-}
-
-export interface InvokeOptions {
-	messages: Array<{
-		role: "system" | "user" | "assistant";
-		content: string;
-	}>;
-	model: string;
-	maxOutputTokens: number;
-	systemPrompt?: string;
-}
-
-// --- Default System Prompt ---------------------------------------------------
-
-const DEFAULT_SUMMARY_SYSTEM = `You are a conversation summarizer. Your task is to create concise, accurate summaries that preserve essential context for continuing a conversation.
-
-Focus on:
-- What was the goal/problem being addressed?
-- What approaches were tried and what were the results?
-- What decisions were made and why?
-- What remains to be done?
-- What files or code sections are relevant?
-
-IMPORTANT: Do NOT include tool_result content verbatim. Instead, summarize the outcomes.`;
-
-const WORK_DONE_FOCUS = `Focus specifically on:
-1. What work was accomplished?
-2. What files were created, modified, or deleted?
-3. What tests passed or failed?
-4. What errors or issues were encountered and resolved?
-5. What is the current state of the project?`;
-
-const DECISIONS_FOCUS = `Focus specifically on:
-1. What architectural decisions were made?
-2. What approaches were chosen over alternatives?
-3. What was the rationale for each decision?
-4. What constraints or requirements influenced the decisions?
-5. Any lessons learned or trade-offs considered?`;
-
-// --- Forked Summarizer ------------------------------------------------------
-
-export class ForkedSummarizer {
-	private readonly config: SummarizerConfig;
-	private readonly invokeAgent: (opts: InvokeOptions) => Promise<InvokeResult>;
-
-	constructor(
-		config: SummarizerConfig,
-		invokeAgent: (opts: InvokeOptions) => Promise<InvokeResult>,
-	) {
-		this.config = {
-			provider: "openai",
-			focusOn: "all",
-			...config,
-		};
-		this.invokeAgent = invokeAgent;
-	}
-
-	/**
-	 * Summarize old messages, keeping recent ones as context
-	 */
-	async summarize(
-		messages: CompactableMessage[],
-		options?: {
-			/** Number of recent messages to keep (default: 5) */
-			keepRecentCount?: number;
-			/** Override focus for this call */
-			focusOn?: "work_done" | "decisions" | "all";
-		},
-	): Promise<SummarizerResult> {
-		const keepRecentCount = options?.keepRecentCount ?? 5;
-		const focusOn = options?.focusOn ?? this.config.focusOn ?? "all";
-
-		// Split messages
-		const recentMessages = messages.slice(-keepRecentCount);
-		const oldMessages = messages.slice(0, -keepRecentCount);
-
-		if (oldMessages.length === 0) {
-			return { summary: "", droppedCount: 0 };
-		}
-
-		// Build summary prompt
-		const summaryPrompt = this.buildSummaryPrompt(oldMessages, { focusOn });
-
-		// Create summarization messages
-		const summarizationMessages = [
-			{
-				role: "system" as const,
-				content: this.config.systemPrompt ?? DEFAULT_SUMMARY_SYSTEM,
-			},
-			{
-				role: "user" as const,
-				content: summaryPrompt,
-			},
-		];
-
-		// Call summarization model
-		try {
-			const result = await this.invokeAgent({
-				messages: summarizationMessages,
-				model: this.config.model,
-				maxOutputTokens: this.config.maxSummaryTokens,
-			});
-
-			if (!result.success || !result.output) {
-				throw new Error(`Summarization failed: ${result.error ?? "unknown"}`);
-			}
-
-			return {
-				summary: result.output.trim(),
-				droppedCount: oldMessages.length,
-				tokensUsed: result.usage?.totalTokens,
-			};
-		} catch (error) {
-			// Return heuristic summary on failure
-			return {
-				summary: this.heuristicSummary(oldMessages, focusOn),
-				droppedCount: oldMessages.length,
-			};
-		}
-	}
-
-	/**
-	 * Build the summary prompt for the LLM
-	 */
-	private buildSummaryPrompt(
-		messages: CompactableMessage[],
-		options: { focusOn?: "work_done" | "decisions" | "all" },
-	): string {
-		const focusOn = options.focusOn ?? "all";
-		const formattedMessages = messages
-			.map((m) => `[${m.role}]\n${this.truncateContent(m.content, 2000)}`)
-			.join("\n\n---\n\n");
-
-		const focusInstructions = {
-			work_done: WORK_DONE_FOCUS,
-			decisions: DECISIONS_FOCUS,
-			all: "Provide a comprehensive summary covering all important aspects of the conversation.",
-		}[focusOn];
-
-		return `Summarize the following conversation concisely.
-
-${focusInstructions}
-
-## Conversation to Summarize
-${formattedMessages}
-
-## Output Format
-Provide a summary in 500-1000 tokens that:
-1. Captures the essential context
-2. Preserves key decisions and their rationale
-3. Notes any ongoing work or next steps
-4. Lists any important files or code sections referenced
-
-Do not include verbatim tool results — summarize their outcomes instead.`;
-	}
-
-	/**
-	 * Truncate content to max length
-	 */
-	private truncateContent(content: string, maxLength: number): string {
-		if (content.length <= maxLength) return content;
-		return content.substring(0, maxLength) + "... [truncated]";
-	}
-
-	/**
-	 * Heuristic summary when LLM summarization fails
-	 */
-	private heuristicSummary(
-		messages: CompactableMessage[],
-		focusOn: "work_done" | "decisions" | "all",
-	): string {
-		const summaries: string[] = [];
-		const keyFiles: Set<string> = new Set();
-		const decisions: string[] = [];
-
-		// Extract patterns from last 20 messages
-		for (const msg of messages.slice(-20)) {
-			const content = msg.content;
-
-			// Look for file references
-			const filePatterns = [
-				/created (?:file|module):?\s*([^\n.]+)/gi,
-				/modified (?:file|module):?\s*([^\n.]+)/gi,
-				/wrote to ([\w./-]+)/gi,
-				/updated ([\w./-]+)/gi,
-			];
-
-			for (const pattern of filePatterns) {
-				for (const match of content.matchAll(pattern)) {
-					if (match[1]) keyFiles.add(match[1].trim());
-				}
-			}
-
-			// Look for decisions
-			const decisionPatterns = [
-				/(?:decided|choosing|choice|chose):?\s*([^\n.]+)/gi,
-				/(?:approach|strategy):?\s*([^\n.]+)/gi,
-			];
-
-			for (const pattern of decisionPatterns) {
-				for (const match of content.matchAll(pattern)) {
-					if (match[1]) decisions.push(match[1].trim());
-				}
-			}
-
-			// Extract key user messages
-			if (msg.role === "user" && content.length > 50) {
-				summaries.push(`- User: ${content.substring(0, 300)}`);
-			}
-
-			// Extract significant actions from assistant
-			if (msg.role === "assistant" && msg.metadata?.action) {
-				summaries.push(`- Action: ${msg.metadata.action}`);
-			}
-		}
-
-		// Build heuristic summary
-		const parts: string[] = [
-			`[Context compacted — ${messages.length} messages summarized by heuristic]`,
-			"",
-		];
-
-		if (keyFiles.size > 0) {
-			parts.push("## Files Referenced");
-			for (const file of Array.from(keyFiles).slice(0, 10)) {
-				parts.push(`- ${file}`);
-			}
-			parts.push("");
-		}
-
-		if (decisions.length > 0) {
-			parts.push("## Key Points");
-			for (const decision of Array.from(new Set(decisions)).slice(0, 5)) {
-				parts.push(`- ${decision}`);
-			}
-			parts.push("");
-		}
-
-		if (summaries.length > 0) {
-			parts.push("## Summary");
-			for (const summary of summaries.slice(0, 10)) {
-				parts.push(summary);
-			}
-			parts.push("");
-		}
-
-		parts.push("Please continue from the recent messages provided.");
-
-		return parts.join("\n");
-	}
-}
-
-// --- Convenience Factory ----------------------------------------------------
-
-/**
- * Create a ForkedSummarizer with default config
- */
-export function createForkedSummarizer(
-	model: string,
-	invokeAgent: (opts: InvokeOptions) => Promise<InvokeResult>,
-	config?: Partial<SummarizerConfig>,
-): ForkedSummarizer {
-	return new ForkedSummarizer(
-		{
-			model,
-			maxSummaryTokens: 4096,
-			...config,
-		},
-		invokeAgent,
-	);
-}

package/harness/herdr-agents.ts

@@ -1,91 +0,0 @@
-/**
- * Herdr Agent Commands
- */
-import { existsSync, readdirSync, readFileSync } from "fs";
-import { join } from "path";
-import {
-	createHerdrBus,
-	getHerdrWorkspacePaths,
-	ensureHerdrWorkspace,
-} from "../packages/event-bus/src/herdr-bus.js";
-
-async function startReviewAgent(): Promise<void> {
-	console.log("[herdr:review] Starting review agent...");
-	const bus = createHerdrBus("review-agent");
-	ensureHerdrWorkspace();
-	bus.subscribe("code.written");
-	bus.startPolling(async (payload) => {
-		if (payload.topic === "code.written") {
-			const data = payload.data as { taskId: string; files: string[] };
-			console.log(
-				`[herdr:review] code.written: task=${data.taskId} files=${data.files.length}`,
-			);
-			for (const file of data.files ?? []) {
-				console.log(`[herdr:review] Reviewing: ${file}`);
-			}
-		}
-	});
-	console.log(`[herdr:review] Workspace: ${bus.getWorkspace()}`);
-	await new Promise(() => {});
-}
-
-async function startCodeAgent(): Promise<void> {
-	console.log("[herdr:code] Starting code agent...");
-	const bus = createHerdrBus("code-agent");
-	ensureHerdrWorkspace();
-	bus.subscribe("review.completed");
-	bus.startPolling(async (payload) => {
-		if (payload.topic === "review.completed") {
-			const data = payload.data as {
-				taskId: string;
-				reportFile: string;
-				status: string;
-			};
-			console.log(
-				`[herdr:code] Review done: task=${data.taskId} status=${data.status} report=${data.reportFile}`,
-			);
-		}
-	});
-	console.log(`[herdr:code] Workspace: ${bus.getWorkspace()}`);
-	await new Promise(() => {});
-}
-
-function showStatus(): void {
-	const paths = getHerdrWorkspacePaths();
-	console.log(`Workspace: ${paths.root}`);
-	const evPath = join(paths.root, "events.jsonl");
-	if (existsSync(evPath)) {
-		const lines = readFileSync(evPath, "utf-8")
-			.split("\n")
-			.filter(Boolean)
-			.slice(-10);
-		console.log("Recent events:");
-		for (const line of lines) {
-			try {
-				const { topic, ts } = JSON.parse(line);
-				console.log(`  ${ts} ${topic}`);
-			} catch {
-				/* skip */
-			}
-		}
-	}
-	if (existsSync(paths.subscriptions)) {
-		const agents = readdirSync(paths.subscriptions);
-		console.log(`Active agents: ${agents.join(", ") || "none"}`);
-	}
-}
-
-const [command] = process.argv.slice(2);
-switch (command) {
-	case "review":
-		startReviewAgent().catch(console.error);
-		break;
-	case "code":
-		startCodeAgent().catch(console.error);
-		break;
-	case "status":
-		showStatus();
-		break;
-	default:
-		console.log("Usage: bun harness/herdr-agents.ts <review|code|status>");
-}

package/harness/index.ts

@@ -1,232 +0,0 @@
-/**
- * Harness Runtime — Core Module Index
- *
- * Re-exports all harness components for easy importing.
- */
-
-// Re-export types from the types package
-export type {
-	JobStatus,
-	TaskStatus,
-	RuntimeCheckpoint,
-	RuntimeTask,
-	RuntimeEvent,
-	TaskGraph,
-	TaskNode,
-	RepairTask,
-	FailureType,
-	AttemptedFix,
-	RetryPolicy,
-	QuotaSignal,
-	QuotaState,
-	ProviderConfig,
-	ProviderCapability,
-	ProviderRequest,
-	ProviderResponse,
-	E2EScenario,
-	E2EStep,
-	E2EResult,
-	E2EReport,
-	ContextWindowUpdate,
-	ContextWindowStats,
-	ContextWindowConfig,
-	CompactableMessage,
-	CompactResult,
-	CompactTriggerReason,
-} from "../packages/types/src/runtime-types.js";
-
-// State Machine
-export { JobStateMachine } from "./job-state-machine.js";
-export type {
-	StateTransition,
-	TransitionResult,
-} from "./job-state-machine.js";
-export type { CheckpointManager } from "./job-state-machine.js";
-
-// Task Graph
-export type { TaskGraphManager } from "./task-graph.js";
-
-// Loop Runtime
-export { LoopRuntime } from "./loop-runtime.js";
-export type { LoopResult, LoopCallbacks } from "./loop-runtime.js";
-
-// Repair Engine
-export { RepairEngine } from "./repair-engine.js";
-export type { RepairResult } from "./repair-engine.js";
-
-// Auto Quota Resume (5h window auto-resume)
-export {
-	scheduleAutoResume,
-	cancelAutoResume,
-	getScheduledResume,
-} from "./auto-quota-resume.js";
-export type {} from "./auto-quota-resume.js";
-
-// Master Planner
-export { MasterPlanner } from "./master-planner.js";
-export type { PlanResult } from "./master-planner.js";
-
-// Context Window Manager (RFC-0010 — enhanced with thresholds)
-export {
-	ContextWindowManager,
-	microcompactToolResults,
-	parseTokenGapFromError,
-	AUTOCOMPACT_BUFFER_TOKENS,
-	WARNING_THRESHOLD_BUFFER_TOKENS,
-	BLOCKING_THRESHOLD_BUFFER_TOKENS,
-	MAX_CONSECUTIVE_COMPACT_FAILURES,
-} from "./context-window-manager.js";
-export type {
-	CompactThresholds,
-	TokenEstimate,
-} from "./context-window-manager.js";
-
-// Forked Summarizer (RFC-0028 Phase 2 — LLM-based compaction)
-export {
-	ForkedSummarizer,
-	createForkedSummarizer,
-} from "./forked-summarizer.js";
-export type {
-	SummarizerConfig,
-	SummarizerResult,
-	InvokeOptions,
-} from "./forked-summarizer.js";
-
-// Continue Prompt Generator (RFC-0029 Phase 5 — auto-resume)
-export {
-	ContinuePromptGenerator,
-	continuePromptGenerator,
-} from "./continue-prompt.js";
-export type {
-	ContinueContext,
-	MinimalContinueContext,
-} from "./continue-prompt.js";
-
-// BlackBoard
-export { createBlackboard, SharedBlackboard } from "./blackboard.js";
-
-// Agent Handoff
-export { AgentHandoffProtocol } from "./agent-handoff.js";
-
-// Auto Compact (RFC-0019)
-export { AutoCompactEngine } from "./auto-compact.js";
-export type { CompactionEvent, CompactionConfig } from "./auto-compact.js";
-
-// Context Compact Orchestrator (RFC-0028)
-export { CompactOrchestrator } from "./context-compact-orchestrator.js";
-export type {
-	CompactOrchestratorConfig,
-	CompactOrchestratorCallbacks,
-} from "./context-compact-orchestrator.js";
-
-// Output Limit Handler (RFC-0020)
-export { OutputLimitHandler } from "./output-limit-handler.js";
-export type {
-	OutputLimitConfig,
-	OutputLimitEvent,
-	ExpectedOutput,
-} from "./output-limit-handler.js";
-
-// Partial Recovery (RFC-0021)
-export { PartialRecovery, createPartialRecovery } from "./partial-recovery.js";
-export type {
-	PartialResponse,
-	RecoveryStatus,
-	MergeOptions,
-} from "./partial-recovery.js";
-
-// Session Memory (RFC-0030)
-export {
-	SessionMemoryManager,
-	createSessionMemoryManager,
-} from "./session-memory.js";
-export type {
-	SessionMemory,
-	Decision,
-	FileReference,
-	TestResult,
-} from "./session-memory.js";
-
-// Notification Events (RFC-0022)
-export {
-	HarnessNotificationEvents,
-	createNotificationConfigFromEnv,
-} from "./notification-events.js";
-
-// E2E Testing
-export { E2ETestEngine } from "./e2e/test-engine.js";
-export { PlaywrightE2ERunner } from "./e2e/playwright-runner.js";
-export type {
-	E2ERunner,
-	PlaywrightRunnerConfig,
-} from "./e2e/playwright-runner.js";
-export {
-	MiniMaxQuotaScraper,
-	MiniMaxQuotaManager,
-} from "./e2e/minimax-quota-scraper.js";
-export type {
-	MiniMaxQuotaData,
-	MiniMaxScraperConfig,
-} from "./e2e/minimax-quota-scraper.js";
-export {
-	QuotaStatusManager,
-	formatQuotaStatus,
-	createQuotaStatusManagerFromEnv,
-} from "./e2e/quota-status.js";
-export type { QuotaStatus, QuotaDisplayConfig } from "./e2e/quota-status.js";
-
-// Project Detector
-export { ProjectDetector } from "./project-detector/detector.js";
-
-// --- RFC-0056: Performance Optimizer ----------------------------------------
-export {
-	PerformanceOptimizer,
-	createPerformanceOptimizer,
-} from "../packages/performance-optimizer/src/index.js";
-export type {
-	PerformanceOptimizationRequest,
-	PerformanceOptimizationPlan,
-	PerformanceRecommendation,
-} from "../packages/performance-optimizer/src/types.js";
-
-// --- RFC-0057: Evaluation Engine -------------------------------------------
-export {
-	EvaluationEngine,
-	createEvaluationEngine,
-} from "../packages/evaluation-engine/src/index.js";
-export type {
-	EvaluationRequest,
-	EvaluationResult,
-	EvaluationFinding,
-} from "../packages/evaluation-engine/src/types.js";
-
-// --- RFC-0058: Learning Engine ---------------------------------------------
-export {
-	LearningEngine,
-	createLearningEngine,
-} from "../packages/learning-engine/src/index.js";
-export type {
-	LearningRequest,
-	LearnedExperience,
-} from "../packages/learning-engine/src/types.js";
-
-// --- RFC-0059: Experience Replay ------------------------------------------
-export {
-	ExperienceReplay,
-	createExperienceReplay,
-} from "../packages/experience-replay/src/index.js";
-export type {
-	ReplayRequest,
-	ReplayResult,
-} from "../packages/experience-replay/src/types.js";
-
-// --- RFC-0060: Memory Engine -----------------------------------------------
-export {
-	MemoryEngine,
-	createMemoryEngine,
-} from "../packages/memory-engine/src/index.js";
-export type {
-	OkfConcept,
-	KnowledgeQuery,
-	KnowledgeResult,
-} from "../packages/memory-engine/src/types.js";