pi-harness-runtime 1.1.77 → 1.1.78

package/packages/jev-judge/src/auto-continue.ts

@@ -0,0 +1,317 @@
+/**
+ * Auto-Continue Judge - Agent decision making when user is unavailable
+ *
+ * Uses Jev to decide whether an agent should continue autonomously
+ * when waiting for user input that may not come.
+ *
+ * Scenario: Agent completes 2/5 tasks, asks "Continue?" User asleep.
+ * Jev decides: proceed, wait, or proceed with caution.
+ */
+
+import { JevJudge } from "./index.js";
+import type { JevJudgeConfig, JevEvaluationResult } from "./types.js";
+
+/**
+ * Task tracking state
+ */
+export interface TaskState {
+  completedTasks: string[];
+  remainingTasks: string[];
+  totalTasks: number;
+  waitTimeMinutes: number;
+  userResponded: boolean;
+  lastUserActivity?: Date;
+  sessionStartTime: Date;
+}
+
+/**
+ * Auto-continue decision
+ */
+export interface AutoContinueDecision {
+  action: "proceed" | "wait" | "proceed_with_caution";
+  probability: number; // 0-1 confidence to proceed
+  riskLevel: "minimal" | "low" | "medium" | "high" | "critical";
+  reasoning: string;
+  urgency: "critical" | "high" | "normal" | "low";
+  estimatedImpact: "minimal" | "moderate" | "significant" | "major";
+  confidence: "high" | "medium" | "low";
+  waitRecommendation?: number; // minutes to wait if action is "wait"
+}
+
+/**
+ * Auto-Continue Judge configuration
+ */
+export interface AutoContinueConfig extends JevJudgeConfig {
+  /** Minimum probability to auto-proceed (default: 0.7) */
+  proceedThreshold?: number;
+  /** Minimum probability to proceed with caution (default: 0.5) */
+  cautionThreshold?: number;
+  /** Maximum wait time before force decision (default: 30 minutes) */
+  maxWaitMinutes?: number;
+  /** Default action if Jev fails (default: "wait") */
+  fallbackAction?: "proceed" | "wait" | "proceed_with_caution";
+}
+
+/**
+ * Default configuration
+ */
+const DEFAULT_CONFIG = {
+  proceedThreshold: 0.7,
+  cautionThreshold: 0.5,
+  maxWaitMinutes: 30,
+  fallbackAction: "wait" as const,
+};
+
+/**
+ * Auto-Continue Judge
+ *
+ * Judges whether an agent should continue autonomously when user is unavailable.
+ * Integrates with pi-coding-agent hooks for automatic decision making.
+ */
+export class AutoContinueJudge {
+  private jev: JevJudge;
+  private config: Required<AutoContinueConfig>;
+
+  constructor(config: AutoContinueConfig) {
+    this.config = {
+      ...DEFAULT_CONFIG,
+      ...config,
+    } as Required<AutoContinueConfig>;
+
+    this.jev = new JevJudge(config);
+  }
+
+  /**
+   * Judge whether to continue or wait
+   */
+  async decide(
+    taskState: TaskState
+  ): Promise<AutoContinueDecision> {
+    const state = this.buildState(taskState);
+
+    try {
+      const result = await this.jev.evaluate(state, {
+        shouldProceed: {
+          type: "noul",
+          instructions:
+            "The agent should continue with remaining tasks without waiting for user confirmation. Consider: task completion %, risk of wrong action, time sensitivity, user unavailability.",
+        },
+        riskLevel: {
+          type: "score",
+          instructions: "Risk level if agent proceeds without user confirmation",
+          min: 1,
+          max: 5,
+        },
+        urgency: {
+          type: "choice",
+          instructions: "How urgent are the remaining tasks?",
+          options: ["critical", "high", "normal", "low"],
+        },
+        estimatedImpact: {
+          type: "choice",
+          instructions: "Impact if the agent makes wrong decision by proceeding",
+          options: ["minimal", "moderate", "significant", "major"],
+        },
+        waitBenefit: {
+          type: "noul",
+          instructions:
+            "Would waiting for user confirmation significantly improve the outcome or prevent a serious mistake?",
+        },
+        taskComplexity: {
+          type: "score",
+          instructions:
+            "How complex are the remaining tasks compared to completed ones?",
+          min: 1,
+          max: 5,
+        },
+      });
+
+      return this.buildDecision(result, taskState);
+    } catch (error) {
+      console.error("[AutoContinueJudge] Jev call failed:", error);
+      return this.fallbackDecision(taskState);
+    }
+  }
+
+  /**
+   * Quick decision - simple yes/no to proceed
+   */
+  async shouldContinue(
+    taskState: TaskState
+  ): Promise<{ shouldContinue: boolean; probability: number; confidence: string }> {
+    const result = await this.decide(taskState);
+
+    return {
+      shouldContinue: result.action !== "wait",
+      probability: result.probability,
+      confidence: result.confidence,
+    };
+  }
+
+  /**
+   * Build state for Jev
+   */
+  private buildState(taskState: TaskState): string {
+    const completionPercent = Math.round(
+      (taskState.completedTasks.length / taskState.totalTasks) * 100
+    );
+
+    const state = {
+      task_summary: {
+        completed: taskState.completedTasks.length,
+        remaining: taskState.remainingTasks.length,
+        total: taskState.totalTasks,
+        completion_percent: completionPercent,
+        completed_list: taskState.completedTasks,
+        remaining_list: taskState.remainingTasks,
+      },
+      user_status: {
+        responded: taskState.userResponded,
+        wait_time_minutes: taskState.waitTimeMinutes,
+        last_activity: taskState.lastUserActivity?.toISOString() ?? "unknown",
+        session_duration_minutes: Math.round(
+          (Date.now() - taskState.sessionStartTime.getTime()) / 60000
+        ),
+      },
+    };
+
+    return JSON.stringify(state, null, 2);
+  }
+
+  /**
+   * Build decision from Jev result
+   */
+  private buildDecision(
+    result: JevEvaluationResult,
+    _taskState: TaskState
+  ): AutoContinueDecision {
+    const shouldProceed = result.decisions.shouldProceed;
+    const probability = shouldProceed.probability ?? 0.5;
+
+    const riskDecision = result.decisions.riskLevel;
+    const riskScore = (riskDecision?.response as { score: number })?.score ?? 3;
+    const riskLevel = this.scoreToRiskLevel(riskScore);
+
+    const urgencyDecision = result.decisions.urgency as {
+      response: { choice: string };
+    };
+    const urgency = (urgencyDecision?.response?.choice as AutoContinueDecision["urgency"]) ?? "normal";
+
+    const impactDecision = result.decisions.estimatedImpact as {
+      response: { choice: string };
+    };
+    const estimatedImpact = (impactDecision?.response?.choice as AutoContinueDecision["estimatedImpact"]) ?? "moderate";
+
+    const waitBenefit = (result.decisions.waitBenefit?.response as { noul: number })?.noul ?? 0.5;
+
+    // Determine action based on thresholds
+    let action: AutoContinueDecision["action"];
+    let confidence: AutoContinueDecision["confidence"];
+    let reasoning: string;
+
+    if (probability >= this.config.proceedThreshold) {
+      action = "proceed";
+      confidence = probability > 0.85 ? "high" : "medium";
+      reasoning = `High confidence to proceed (${(probability * 100).toFixed(0)}%)`;
+    } else if (probability >= this.config.cautionThreshold && riskScore <= 3) {
+      action = "proceed_with_caution";
+      confidence = "medium";
+      reasoning = `Proceed with caution (${(probability * 100).toFixed(0)}%), risk level ${riskScore}/5`;
+    } else if (waitBenefit < 0.3 && riskScore <= 2) {
+      action = "proceed_with_caution";
+      confidence = "low";
+      reasoning = `Low wait benefit, proceeding despite lower confidence`;
+    } else {
+      action = "wait";
+      confidence = "high";
+      reasoning = `Waiting recommended (${((1 - probability) * 100).toFixed(0)}% confidence to wait)`;
+    }
+
+    // Calculate recommended wait time
+    let waitRecommendation: number | undefined;
+    if (action === "wait") {
+      waitRecommendation = Math.min(
+        this.config.maxWaitMinutes,
+        Math.round(this.config.maxWaitMinutes * (1 - probability))
+      );
+    }
+
+    return {
+      action,
+      probability,
+      riskLevel,
+      reasoning,
+      urgency,
+      estimatedImpact,
+      confidence,
+      waitRecommendation,
+    };
+  }
+
+  /**
+   * Convert score to risk level
+   */
+  private scoreToRiskLevel(score: number): AutoContinueDecision["riskLevel"] {
+    if (score <= 1.5) return "minimal";
+    if (score <= 2.5) return "low";
+    if (score <= 3.5) return "medium";
+    if (score <= 4.5) return "high";
+    return "critical";
+  }
+
+  /**
+   * Fallback decision when Jev fails
+   */
+  private fallbackDecision(taskState: TaskState): AutoContinueDecision {
+    const completionPercent = taskState.completedTasks.length / taskState.totalTasks;
+
+    // If >80% done and waited >15 min, proceed with caution
+    if (completionPercent >= 0.8 && taskState.waitTimeMinutes >= 15) {
+      return {
+        action: "proceed_with_caution",
+        probability: 0.6,
+        riskLevel: "medium",
+        reasoning: "Fallback: high completion, waited long time",
+        urgency: "normal",
+        estimatedImpact: "moderate",
+        confidence: "low",
+        waitRecommendation: 10,
+      };
+    }
+
+    return {
+      action: this.config.fallbackAction,
+      probability: 0.5,
+      riskLevel: "medium",
+      reasoning: "Fallback: Jev unavailable, using default behavior",
+      urgency: "normal",
+      estimatedImpact: "moderate",
+      confidence: "low",
+      waitRecommendation: this.config.maxWaitMinutes,
+    };
+  }
+}
+
+/**
+ * Helper: Create task state from todo list
+ */
+export function createTaskState(
+  completedTasks: string[],
+  remainingTasks: string[],
+  lastUserActivity?: Date,
+  sessionStartTime?: Date
+): TaskState {
+  const totalTasks = completedTasks.length + remainingTasks.length;
+
+  return {
+    completedTasks,
+    remainingTasks,
+    totalTasks,
+    waitTimeMinutes: lastUserActivity
+      ? Math.round((Date.now() - lastUserActivity.getTime()) / 60000)
+      : 0,
+    userResponded: false,
+    lastUserActivity,
+    sessionStartTime: sessionStartTime ?? new Date(),
+  };
+}

package/packages/jev-judge/src/index.ts

@@ -44,6 +44,45 @@ import type {
 // @ts-ignore - openai is installed at workspace root
 import OpenAI from "openai";
 
+/**
+ * Environment variable keys for API keys
+ */
+export const ENV_KEYS = {
+  TYPESAFE_API_KEY: "TYPESAFE_API_KEY",
+  OPENROUTER_API_KEY: "OPENROUTER_API_KEY",
+} as const;
+
+/**
+ * Get API key from environment
+ * Priority: TYPESAFE_API_KEY > OPENROUTER_API_KEY
+ */
+export function getApiKeyFromEnv(): string | undefined {
+  return (
+    process.env[ENV_KEYS.TYPESAFE_API_KEY] ||
+    process.env[ENV_KEYS.OPENROUTER_API_KEY]
+  );
+}
+
+/**
+ * Check if Jev API key is available
+ */
+export function hasJevApiKey(): boolean {
+  return !!getApiKeyFromEnv();
+}
+
+/**
+ * Create JevJudge with auto-detected API key from environment
+ */
+export function createJevJudge(config?: Partial<JevJudgeConfig>): JevJudge {
+  const apiKey = getApiKeyFromEnv();
+  if (!apiKey) {
+    throw new Error(
+      `Jev API key not found. Set ${ENV_KEYS.TYPESAFE_API_KEY} or ${ENV_KEYS.OPENROUTER_API_KEY} in environment.`
+    );
+  }
+  return new JevJudge({ apiKey, ...config });
+}
+
 /**
  * Default configuration
  */
@@ -131,7 +170,6 @@ Respond with a JSON object mapping each question ID to its answer.`,
       ],
       response_format: {
         type: "json_object",
-        schema: this.buildResponseSchema(questions),
       },
       temperature: 0.1, // Low temperature for consistent decisions
     });
@@ -263,33 +301,8 @@ Respond with a JSON object mapping each question ID to its answer.`,
 
   // Private helpers
 
-  private buildResponseSchema(questions: JevQuestions): Record<string, unknown> {
-    const schema: Record<string, unknown> = {};
-
-    for (const [id, question] of Object.entries(questions)) {
-      switch ((question as JevQuestion).type) {
-        case "noul":
-          schema[id] = { type: "object", properties: {
-            noul: { type: "number", minimum: 0, maximum: 1 }
-          }, required: ["noul"] };
-          break;
-        case "choice":
-          schema[id] = { type: "object", properties: {
-            choice: { type: "string", enum: (question as { options: string[] }).options },
-            confidence: { type: "number", minimum: 0, maximum: 1 }
-          }, required: ["choice"] };
-          break;
-        case "score":
-          schema[id] = { type: "object", properties: {
-            score: { type: "number" },
-            confidence: { type: "number", minimum: 0, maximum: 1 }
-          }, required: ["score"] };
-          break;
-      }
-    }
-
-    return { type: "object", properties: schema, additionalProperties: false };
-  }
+  // Note: Response schema removed - using json_object mode without strict schema
+  // This allows Jev to return flexible responses
 
   private buildDecision<T extends string>(
     questionId: T,
@@ -343,3 +356,5 @@ export * from "./types.js";
 export { E2EJudge } from "./e2e-judge.js";
 export { FlakyDetector } from "./flaky-detector.js";
 export { DecisionEngine } from "./decision-engine.js";
+export { AutoContinueJudge, createTaskState } from "./auto-continue.js";
+export type { TaskState, AutoContinueDecision, AutoContinueConfig } from "./auto-continue.js";

package/packages/knowledge-graph/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/knowledge-graph",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Knowledge Graph extraction and provenance linking for SKILL.md files",
   "type": "module",
   "main": "./dist/index.js",

package/packages/knowledge-retrieval/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/knowledge-retrieval",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Knowledge retrieval via TencentDB-Agent-Memory MCP (RFC-0105)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/laravel-plugin/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/laravel-plugin",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Laravel framework analysis — artisan commands, Eloquent models, Blade",
   "type": "module",
   "main": "./dist/index.js",

package/packages/learning-engine/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/learning-engine",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Learning engine for extracting lessons from runtime execution (RFC-0058)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/mcp-adapter/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/mcp-adapter",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "type": "module",
   "description": "MCP server adapter for pi-harness — exposes harness capabilities as MCP tools/resources/prompts (RFC-0068)",
   "main": "dist/index.js",

package/packages/memory-engine/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/memory-engine",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Memory engine with OKF integration for durable knowledge management (RFC-0060)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/milestone-manager/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/milestone-manager",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Track milestones and alerts when items are at risk",
   "type": "module",
   "main": "./dist/index.js",

package/packages/model-registry/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/model-registry",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Model Registry for pi-harness-runtime — centralized model metadata with pricing",
   "type": "module",
   "main": "./dist/index.js",

package/packages/nextjs-plugin/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/nextjs-plugin",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Next.js framework detection and analysis for pi-harness-runtime (RFC-0062)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/notification/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/notification",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "pi-harness-runtime — notification",
   "type": "module",
   "main": "./dist/index.js",

package/packages/observability/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/observability",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Comprehensive observability system with structured logging, tracing, metrics, and alerting",
   "type": "module",
   "main": "./dist/index.js",

package/packages/okf-indexer/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/okf-indexer",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "OKF document indexer for SKILL.md files (RFC-0106)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/performance-optimizer/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/performance-optimizer",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Performance optimization analyzer for pi-harness runtime (RFC-0056)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/privilege-broker/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/privilege-broker",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Privilege Broker — capability registry and command executor for autonomous runtime (RFC-0101 §7)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/project-analyzer/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/project-analyzer",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Project analysis system for pi-harness-runtime - detects frameworks, rules, and project structure",
   "type": "module",
   "main": "./dist/index.js",

package/packages/project-bootstrap/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/project-bootstrap",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Project scaffolding from typed ProjectSpec",
   "type": "module",
   "main": "./dist/index.js",

package/packages/projection-engine/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/projection-engine",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "type": "module",
   "main": "src/index.ts",
   "exports": {

package/packages/prompt-compiler/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/prompt-compiler",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Deterministic prompt compiler for pi-harness-runtime",
   "type": "module",
   "main": "./dist/index.js",

package/packages/prompt-versioning/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/prompt-versioning",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "type": "module",
   "description": "Version control for prompts with rollback support (RFC-0019)",
   "main": "dist/index.js",

package/packages/provider-adapter-sdk/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/provider-adapter-sdk",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "SDK for building, testing, and registering AI provider adapters",
   "type": "module",
   "main": "./dist/index.js",

package/packages/provider-router/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/provider-router",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Enhanced Provider Router for pi-harness-runtime — intelligent routing with capabilities, costs, and quotas",
   "type": "module",
   "main": "./dist/provider-router.js",

package/packages/provider-selector/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/provider-selector",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Multi-Provider Selector — route requests by cost, quality, and latency (RFC-0012)",
   "type": "module",
   "main": "./dist/index.js",

package/packages/providers/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/providers",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "pi-harness-runtime — providers",
   "type": "module",
   "main": "./dist/index.js",

package/packages/quota-manager/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/quota-manager",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "pi-harness-runtime — quota-manager",
   "type": "module",
   "main": "./dist/index.js",

package/packages/rate-limiter/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/rate-limiter",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "type": "module",
   "description": "Rate limiting with token bucket algorithm (RFC-0023)",
   "main": "dist/index.js",

package/packages/react-vite-plugin/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/react-vite-plugin",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "React/Vite framework analysis — components, config, TypeScript setup",
   "type": "module",
   "main": "./dist/index.js",

package/packages/release-manager/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi/release-manager",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "type": "module",
   "description": "Release management for pi-harness",
   "main": "dist/index.js",

package/packages/requirement-compiler/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@pi-harness/requirement-compiler",
-  "version": "1.1.77",
+  "version": "1.1.78",
   "description": "Compiles raw requirements into validated CompiledRequirement with ambiguity detection and risk tagging",
   "type": "module",
   "main": "./dist/index.js",