pulseed 0.1.3 → 0.1.5

package/dist/execution/task/task-verifier-llm 2.d.ts

@@ -0,0 +1,24 @@
+import type { Task } from "../../types/task.js";
+import type { AgentResult } from "../adapter-layer.js";
+import type { Logger } from "../../runtime/logger.js";
+import type { VerifierDeps } from "./task-verifier-types.js";
+/**
+ * Wrap a promise with a timeout. Rejects with a TimeoutError if the promise
+ * does not resolve within `ms` milliseconds.
+ */
+export declare function withTimeout<T>(promise: Promise<T>, ms: number): Promise<T>;
+/**
+ * Call an async function with retry + exponential backoff.
+ * On each failure (including timeout), wait `backoffMs * 2^attempt` before retrying.
+ * After `maxRetries` retries, the last error is re-thrown.
+ */
+export declare function withRetry<T>(fn: () => Promise<T>, maxRetries: number, backoffMs: number, logger?: Logger, label?: string): Promise<T>;
+export declare function runLLMReview(deps: VerifierDeps, task: Task, executionResult: AgentResult, knowledgeBlock?: string, stateBlock?: string, modelTier?: 'main' | 'light'): Promise<{
+    passed: boolean;
+    partial: boolean;
+    description: string;
+    confidence: number;
+    criteria_met?: number;
+    criteria_total?: number;
+}>;
+//# sourceMappingURL=task-verifier-llm%202.d.ts.map

package/dist/execution/task/task-verifier-llm 2.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-verifier-llm 2.d.ts","sourceRoot":"","sources":["../../../src/execution/task/task-verifier-llm 2.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,qBAAqB,CAAC;AAChD,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AACvD,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAEtD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAK7D;;;GAGG;AACH,wBAAgB,WAAW,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,CAU1E;AAID;;;;GAIG;AACH,wBAAsB,SAAS,CAAC,CAAC,EAC/B,EAAE,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACpB,UAAU,EAAE,MAAM,EAClB,SAAS,EAAE,MAAM,EACjB,MAAM,CAAC,EAAE,MAAM,EACf,KAAK,CAAC,EAAE,MAAM,GACb,OAAO,CAAC,CAAC,CAAC,CAgBZ;AAID,wBAAsB,YAAY,CAChC,IAAI,EAAE,YAAY,EAClB,IAAI,EAAE,IAAI,EACV,eAAe,EAAE,WAAW,EAC5B,cAAc,SAAK,EACnB,UAAU,SAAK,EACf,SAAS,GAAE,MAAM,GAAG,OAAiB,GACpC,OAAO,CAAC;IAAE,MAAM,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,OAAO,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC;IAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAAC,cAAc,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CA2JzI"}

package/dist/execution/task/task-verifier-llm 2.js

@@ -0,0 +1,160 @@
+import { CompletionJudgerResponseSchema } from "./task-verifier-types.js";
+// ─── withTimeout ───
+/**
+ * Wrap a promise with a timeout. Rejects with a TimeoutError if the promise
+ * does not resolve within `ms` milliseconds.
+ */
+export function withTimeout(promise, ms) {
+    return new Promise((resolve, reject) => {
+        const timer = setTimeout(() => {
+            reject(new Error(`completion_judger timeout after ${ms}ms`));
+        }, ms);
+        promise.then((value) => { clearTimeout(timer); resolve(value); }, (err) => { clearTimeout(timer); reject(err); });
+    });
+}
+// ─── withRetry ───
+/**
+ * Call an async function with retry + exponential backoff.
+ * On each failure (including timeout), wait `backoffMs * 2^attempt` before retrying.
+ * After `maxRetries` retries, the last error is re-thrown.
+ */
+export async function withRetry(fn, maxRetries, backoffMs, logger, label) {
+    let lastErr;
+    for (let attempt = 0; attempt <= maxRetries; attempt++) {
+        try {
+            return await fn();
+        }
+        catch (err) {
+            lastErr = err;
+            if (attempt < maxRetries) {
+                const delay = backoffMs * Math.pow(2, attempt);
+                const msg = err instanceof Error ? err.message : String(err);
+                logger?.warn(`[completion_judger] ${label ?? "LLM call"} failed (attempt ${attempt + 1}/${maxRetries + 1}): ${msg} — retrying in ${delay}ms`);
+                await new Promise((res) => setTimeout(res, delay));
+            }
+        }
+    }
+    throw lastErr;
+}
+// ─── runLLMReview ───
+export async function runLLMReview(deps, task, executionResult, knowledgeBlock = "", stateBlock = "", modelTier = 'light') {
+    const timeoutMs = deps.completionJudgerConfig?.timeoutMs ?? 30_000;
+    const maxRetries = deps.completionJudgerConfig?.maxRetries ?? 2;
+    const retryBackoffMs = deps.completionJudgerConfig?.retryBackoffMs ?? 1_000;
+    // Create review session
+    const reviewSession = await deps.sessionManager.createSession("task_review", task.goal_id, task.id);
+    // Build review context (excludes executor self-report for bias prevention)
+    const reviewContext = deps.sessionManager.buildTaskReviewContext(task.goal_id, task.id);
+    const criteriaList = task.success_criteria
+        .map((c, i) => `${i + 1}. ${c.description} (blocking: ${c.is_blocking}, method: ${c.verification_method})`)
+        .join("\n");
+    const enrichmentBlocks = [knowledgeBlock, stateBlock].filter(Boolean).join("\n");
+    const prompt = `Evaluate task execution against success criteria.
+
+Task: ${task.work_description}
+Approach: ${task.approach}
+
+Criteria:
+${criteriaList}
+${enrichmentBlocks ? `\n${enrichmentBlocks}\n` : ""}
+Output (first 2000 chars):
+${executionResult.output.slice(0, 2000)}
+
+Status: ${executionResult.stopped_reason} | Success: ${executionResult.success}
+Context: ${reviewContext.map((s) => s.content).join(" ")}
+
+Return JSON:
+{"verdict": "pass"|"partial"|"fail", "reasoning": "...", "criteria_met": #, "criteria_total": #}`;
+    // Gateway path: route through PromptGateway when available
+    if (deps.gateway) {
+        let parsed;
+        try {
+            parsed = await withRetry(() => withTimeout(deps.gateway.execute({
+                purpose: "verification",
+                goalId: task.goal_id,
+                additionalContext: { review_prompt: prompt },
+                responseSchema: CompletionJudgerResponseSchema,
+                maxTokens: 1024,
+            }), timeoutMs), maxRetries, retryBackoffMs, deps.logger, `completion_judger for task ${task.id}`);
+        }
+        catch (err) {
+            const msg = err instanceof Error ? err.message : String(err);
+            deps.logger?.error(`[completion_judger] All retries exhausted for task ${task.id}: ${msg}`);
+            await deps.sessionManager.endSession(reviewSession.id, `completion_judger failed: ${msg}`);
+            return {
+                passed: false,
+                partial: false,
+                description: `completion_judger failed after ${maxRetries + 1} attempt(s): ${msg}`,
+                confidence: 0.0,
+            };
+        }
+        const verdictStr = parsed.verdict;
+        const result = {
+            passed: verdictStr === "pass",
+            partial: verdictStr === "partial",
+            description: parsed.reasoning || "LLM review completed",
+            confidence: verdictStr === "pass" ? 0.8 : verdictStr === "partial" ? 0.6 : 0.8,
+            criteria_met: parsed.criteria_met,
+            criteria_total: parsed.criteria_total,
+        };
+        await deps.sessionManager.endSession(reviewSession.id, `LLM review: ${verdictStr}`);
+        return result;
+    }
+    // Direct LLM path (fallback when no gateway)
+    let response;
+    try {
+        response = await withRetry(() => withTimeout((deps.reviewerLlmClient ?? deps.llmClient).sendMessage([{ role: "user", content: prompt }], {
+            system: "Review task results objectively against criteria. Ignore executor self-assessment.",
+            max_tokens: 1024,
+            model_tier: modelTier,
+        }), timeoutMs), maxRetries, retryBackoffMs, deps.logger, `completion_judger for task ${task.id}`);
+    }
+    catch (err) {
+        const msg = err instanceof Error ? err.message : String(err);
+        deps.logger?.error(`[completion_judger] All retries exhausted for task ${task.id}: ${msg}`);
+        await deps.sessionManager.endSession(reviewSession.id, `completion_judger failed: ${msg}`);
+        return {
+            passed: false,
+            partial: false,
+            description: `completion_judger failed after ${maxRetries + 1} attempt(s): ${msg}`,
+            confidence: 0.0,
+        };
+    }
+    try {
+        const rawJson = response.content.replace(/```json\n?/g, "").replace(/```/g, "").trim();
+        const parseResult = CompletionJudgerResponseSchema.safeParse(JSON.parse(rawJson));
+        if (!parseResult.success) {
+            deps.logger?.warn(`[completion_judger] Zod parse failed for task ${task.id}: ${parseResult.error.message}`);
+            await deps.sessionManager.endSession(reviewSession.id, "Failed to parse LLM review result");
+            return {
+                passed: false,
+                partial: false,
+                description: "Failed to parse LLM review result",
+                confidence: 0.3,
+            };
+        }
+        const parsed = parseResult.data;
+        const verdictStr = parsed.verdict;
+        const result = {
+            passed: verdictStr === "pass",
+            partial: verdictStr === "partial",
+            description: parsed.reasoning || "LLM review completed",
+            confidence: verdictStr === "pass" ? 0.8 : verdictStr === "partial" ? 0.6 : 0.8,
+            criteria_met: parsed.criteria_met,
+            criteria_total: parsed.criteria_total,
+        };
+        await deps.sessionManager.endSession(reviewSession.id, `LLM review: ${verdictStr}`);
+        return result;
+    }
+    catch {
+        deps.logger?.warn(`[completion_judger] JSON.parse failed for task ${task.id}`);
+        await deps.sessionManager.endSession(reviewSession.id, "Failed to parse LLM review result");
+        return {
+            passed: false,
+            partial: false,
+            description: "Failed to parse LLM review result",
+            confidence: 0.3,
+        };
+    }
+}
+//# sourceMappingURL=task-verifier-llm%202.js.map

package/dist/execution/task/task-verifier-llm 2.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-verifier-llm 2.js","sourceRoot":"","sources":["../../../src/execution/task/task-verifier-llm 2.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,8BAA8B,EAAE,MAAM,0BAA0B,CAAC;AAE1E,sBAAsB;AAEtB;;;GAGG;AACH,MAAM,UAAU,WAAW,CAAI,OAAmB,EAAE,EAAU;IAC5D,OAAO,IAAI,OAAO,CAAI,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACxC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;YAC5B,MAAM,CAAC,IAAI,KAAK,CAAC,mCAAmC,EAAE,IAAI,CAAC,CAAC,CAAC;QAC/D,CAAC,EAAE,EAAE,CAAC,CAAC;QACP,OAAO,CAAC,IAAI,CACV,CAAC,KAAK,EAAE,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EACnD,CAAC,GAAG,EAAE,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAC/C,CAAC;IACJ,CAAC,CAAC,CAAC;AACL,CAAC;AAED,oBAAoB;AAEpB;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,SAAS,CAC7B,EAAoB,EACpB,UAAkB,EAClB,SAAiB,EACjB,MAAe,EACf,KAAc;IAEd,IAAI,OAAgB,CAAC;IACrB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,IAAI,CAAC;YACH,OAAO,MAAM,EAAE,EAAE,CAAC;QACpB,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,GAAG,GAAG,CAAC;YACd,IAAI,OAAO,GAAG,UAAU,EAAE,CAAC;gBACzB,MAAM,KAAK,GAAG,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;gBAC/C,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAC7D,MAAM,EAAE,IAAI,CAAC,uBAAuB,KAAK,IAAI,UAAU,oBAAoB,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,MAAM,GAAG,kBAAkB,KAAK,IAAI,CAAC,CAAC;gBAC9I,MAAM,IAAI,OAAO,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC;YACrD,CAAC;QACH,CAAC;IACH,CAAC;IACD,MAAM,OAAO,CAAC;AAChB,CAAC;AAED,uBAAuB;AAEvB,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,IAAkB,EAClB,IAAU,EACV,eAA4B,EAC5B,cAAc,GAAG,EAAE,EACnB,UAAU,GAAG,EAAE,EACf,YAA8B,OAAO;IAErC,MAAM,SAAS,GAAG,IAAI,CAAC,sBAAsB,EAAE,SAAS,IAAI,MAAM,CAAC;IACnE,MAAM,UAAU,GAAG,IAAI,CAAC,sBAAsB,EAAE,UAAU,IAAI,CAAC,CAAC;IAChE,MAAM,cAAc,GAAG,IAAI,CAAC,sBAAsB,EAAE,cAAc,IAAI,KAAK,CAAC;IAE5E,wBAAwB;IACxB,MAAM,aAAa,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,aAAa,CAC3D,aAAa,EACb,IAAI,CAAC,OAAO,EACZ,IAAI,CAAC,EAAE,CACR,CAAC;IAEF,2EAA2E;IAC3E,MAAM,aAAa,GAAG,IAAI,CAAC,cAAc,CAAC,sBAAsB,CAC9D,IAAI,CAAC,OAAO,EACZ,IAAI,CAAC,EAAE,CACR,CAAC;IAEF,MAAM,YAAY,GAAG,IAAI,CAAC,gBAAgB;SACvC,GAAG,CACF,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACP,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,WAAW,eAAe,CAAC,CAAC,WAAW,aAAa,CAAC,CAAC,mBAAmB,GAAG,CAC9F;SACA,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,MAAM,gBAAgB,GAAG,CAAC,cAAc,EAAE,UAAU,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEjF,MAAM,MAAM,GAAG;;QAET,IAAI,CAAC,gBAAgB;YACjB,IAAI,CAAC,QAAQ;;;EAGvB,YAAY;EACZ,gBAAgB,CAAC,CAAC,CAAC,KAAK,gBAAgB,IAAI,CAAC,CAAC,CAAC,EAAE;;EAEjD,eAAe,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;;UAE7B,eAAe,CAAC,cAAc,eAAe,eAAe,CAAC,OAAO;WACnE,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;;;iGAGyC,CAAC;IAEhG,2DAA2D;IAC3D,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;QACjB,IAAI,MAAsD,CAAC;QAC3D,IAAI,CAAC;YACH,MAAM,GAAG,MAAM,SAAS,CACtB,GAAG,EAAE,CAAC,WAAW,CACf,IAAI,CAAC,OAAQ,CAAC,OAAO,CAAC;gBACpB,OAAO,EAAE,cAAc;gBACvB,MAAM,EAAE,IAAI,CAAC,OAAO;gBACpB,iBAAiB,EAAE,EAAE,aAAa,EAAE,MAAM,EAAE;gBAC5C,cAAc,EAAE,8BAA6F;gBAC7G,SAAS,EAAE,IAAI;aAChB,CAAC,EACF,SAAS,CACV,EACD,UAAU,EACV,cAAc,EACd,IAAI,CAAC,MAAM,EACX,8BAA8B,IAAI,CAAC,EAAE,EAAE,CACxC,CAAC;QACJ,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7D,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,sDAAsD,IAAI,CAAC,EAAE,KAAK,GAAG,EAAE,CAAC,CAAC;YAC5F,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,6BAA6B,GAAG,EAAE,CAAC,CAAC;YAC3F,OAAO;gBACL,MAAM,EAAE,KAAK;gBACb,OAAO,EAAE,KAAK;gBACd,WAAW,EAAE,kCAAkC,UAAU,GAAG,CAAC,gBAAgB,GAAG,EAAE;gBAClF,UAAU,EAAE,GAAG;aAChB,CAAC;QACJ,CAAC;QACD,MAAM,UAAU,GAAG,MAAM,CAAC,OAAO,CAAC;QAClC,MAAM,MAAM,GAAG;YACb,MAAM,EAAE,UAAU,KAAK,MAAM;YAC7B,OAAO,EAAE,UAAU,KAAK,SAAS;YACjC,WAAW,EAAE,MAAM,CAAC,SAAS,IAAI,sBAAsB;YACvD,UAAU,EAAE,UAAU,KAAK,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG;YAC9E,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,cAAc,EAAE,MAAM,CAAC,cAAc;SACtC,CAAC;QACF,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,eAAe,UAAU,EAAE,CAAC,CAAC;QACpF,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,6CAA6C;IAC7C,IAAI,QAAqB,CAAC;IAC1B,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,SAAS,CACxB,GAAG,EAAE,CAAC,WAAW,CACf,CAAC,IAAI,CAAC,iBAAiB,IAAI,IAAI,CAAC,SAAS,CAAC,CAAC,WAAW,CACpD,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,EACnC;YACE,MAAM,EAAE,oFAAoF;YAC5F,UAAU,EAAE,IAAI;YAChB,UAAU,EAAE,SAAS;SACtB,CACF,EACD,SAAS,CACV,EACD,UAAU,EACV,cAAc,EACd,IAAI,CAAC,MAAM,EACX,8BAA8B,IAAI,CAAC,EAAE,EAAE,CACxC,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC7D,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,sDAAsD,IAAI,CAAC,EAAE,KAAK,GAAG,EAAE,CAAC,CAAC;QAC5F,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,6BAA6B,GAAG,EAAE,CAAC,CAAC;QAC3F,OAAO;YACL,MAAM,EAAE,KAAK;YACb,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,kCAAkC,UAAU,GAAG,CAAC,gBAAgB,GAAG,EAAE;YAClF,UAAU,EAAE,GAAG;SAChB,CAAC;IACJ,CAAC;IAED,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACvF,MAAM,WAAW,GAAG,8BAA8B,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;QAClF,IAAI,CAAC,WAAW,CAAC,OAAO,EAAE,CAAC;YACzB,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,iDAAiD,IAAI,CAAC,EAAE,KAAK,WAAW,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;YAC5G,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,mCAAmC,CAAC,CAAC;YAC5F,OAAO;gBACL,MAAM,EAAE,KAAK;gBACb,OAAO,EAAE,KAAK;gBACd,WAAW,EAAE,mCAAmC;gBAChD,UAAU,EAAE,GAAG;aAChB,CAAC;QACJ,CAAC;QACD,MAAM,MAAM,GAAG,WAAW,CAAC,IAAI,CAAC;QAChC,MAAM,UAAU,GAAG,MAAM,CAAC,OAAO,CAAC;QAClC,MAAM,MAAM,GAAG;YACb,MAAM,EAAE,UAAU,KAAK,MAAM;YAC7B,OAAO,EAAE,UAAU,KAAK,SAAS;YACjC,WAAW,EAAE,MAAM,CAAC,SAAS,IAAI,sBAAsB;YACvD,UAAU,EAAE,UAAU,KAAK,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG;YAC9E,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,cAAc,EAAE,MAAM,CAAC,cAAc;SACtC,CAAC;QACF,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,eAAe,UAAU,EAAE,CAAC,CAAC;QACpF,OAAO,MAAM,CAAC;IAChB,CAAC;IAAC,MAAM,CAAC;QACP,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,kDAAkD,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC;QAC/E,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,mCAAmC,CAAC,CAAC;QAC5F,OAAO;YACL,MAAM,EAAE,KAAK;YACb,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,mCAAmC;YAChD,UAAU,EAAE,GAAG;SAChB,CAAC;IACJ,CAAC;AACH,CAAC"}

package/dist/execution/task/task-verifier-llm.d.ts

@@ -0,0 +1,24 @@
+import type { Task } from "../../types/task.js";
+import type { AgentResult } from "../adapter-layer.js";
+import type { Logger } from "../../runtime/logger.js";
+import type { VerifierDeps } from "./task-verifier-types.js";
+/**
+ * Wrap a promise with a timeout. Rejects with a TimeoutError if the promise
+ * does not resolve within `ms` milliseconds.
+ */
+export declare function withTimeout<T>(promise: Promise<T>, ms: number): Promise<T>;
+/**
+ * Call an async function with retry + exponential backoff.
+ * On each failure (including timeout), wait `backoffMs * 2^attempt` before retrying.
+ * After `maxRetries` retries, the last error is re-thrown.
+ */
+export declare function withRetry<T>(fn: () => Promise<T>, maxRetries: number, backoffMs: number, logger?: Logger, label?: string): Promise<T>;
+export declare function runLLMReview(deps: VerifierDeps, task: Task, executionResult: AgentResult, knowledgeBlock?: string, stateBlock?: string, modelTier?: 'main' | 'light'): Promise<{
+    passed: boolean;
+    partial: boolean;
+    description: string;
+    confidence: number;
+    criteria_met?: number;
+    criteria_total?: number;
+}>;
+//# sourceMappingURL=task-verifier-llm.d.ts.map

package/dist/execution/task/task-verifier-llm.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-verifier-llm.d.ts","sourceRoot":"","sources":["../../../src/execution/task/task-verifier-llm.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,qBAAqB,CAAC;AAChD,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AACvD,OAAO,KAAK,EAAE,MAAM,EAAE,MAAM,yBAAyB,CAAC;AAEtD,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAK7D;;;GAGG;AACH,wBAAgB,WAAW,CAAC,CAAC,EAAE,OAAO,EAAE,OAAO,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,MAAM,GAAG,OAAO,CAAC,CAAC,CAAC,CAU1E;AAID;;;;GAIG;AACH,wBAAsB,SAAS,CAAC,CAAC,EAC/B,EAAE,EAAE,MAAM,OAAO,CAAC,CAAC,CAAC,EACpB,UAAU,EAAE,MAAM,EAClB,SAAS,EAAE,MAAM,EACjB,MAAM,CAAC,EAAE,MAAM,EACf,KAAK,CAAC,EAAE,MAAM,GACb,OAAO,CAAC,CAAC,CAAC,CAgBZ;AAID,wBAAsB,YAAY,CAChC,IAAI,EAAE,YAAY,EAClB,IAAI,EAAE,IAAI,EACV,eAAe,EAAE,WAAW,EAC5B,cAAc,SAAK,EACnB,UAAU,SAAK,EACf,SAAS,GAAE,MAAM,GAAG,OAAiB,GACpC,OAAO,CAAC;IAAE,MAAM,EAAE,OAAO,CAAC;IAAC,OAAO,EAAE,OAAO,CAAC;IAAC,WAAW,EAAE,MAAM,CAAC;IAAC,UAAU,EAAE,MAAM,CAAC;IAAC,YAAY,CAAC,EAAE,MAAM,CAAC;IAAC,cAAc,CAAC,EAAE,MAAM,CAAA;CAAE,CAAC,CA2JzI"}

package/dist/execution/task/task-verifier-llm.js

@@ -0,0 +1,160 @@
+import { CompletionJudgerResponseSchema } from "./task-verifier-types.js";
+// ─── withTimeout ───
+/**
+ * Wrap a promise with a timeout. Rejects with a TimeoutError if the promise
+ * does not resolve within `ms` milliseconds.
+ */
+export function withTimeout(promise, ms) {
+    return new Promise((resolve, reject) => {
+        const timer = setTimeout(() => {
+            reject(new Error(`completion_judger timeout after ${ms}ms`));
+        }, ms);
+        promise.then((value) => { clearTimeout(timer); resolve(value); }, (err) => { clearTimeout(timer); reject(err); });
+    });
+}
+// ─── withRetry ───
+/**
+ * Call an async function with retry + exponential backoff.
+ * On each failure (including timeout), wait `backoffMs * 2^attempt` before retrying.
+ * After `maxRetries` retries, the last error is re-thrown.
+ */
+export async function withRetry(fn, maxRetries, backoffMs, logger, label) {
+    let lastErr;
+    for (let attempt = 0; attempt <= maxRetries; attempt++) {
+        try {
+            return await fn();
+        }
+        catch (err) {
+            lastErr = err;
+            if (attempt < maxRetries) {
+                const delay = backoffMs * Math.pow(2, attempt);
+                const msg = err instanceof Error ? err.message : String(err);
+                logger?.warn(`[completion_judger] ${label ?? "LLM call"} failed (attempt ${attempt + 1}/${maxRetries + 1}): ${msg} — retrying in ${delay}ms`);
+                await new Promise((res) => setTimeout(res, delay));
+            }
+        }
+    }
+    throw lastErr;
+}
+// ─── runLLMReview ───
+export async function runLLMReview(deps, task, executionResult, knowledgeBlock = "", stateBlock = "", modelTier = 'light') {
+    const timeoutMs = deps.completionJudgerConfig?.timeoutMs ?? 30_000;
+    const maxRetries = deps.completionJudgerConfig?.maxRetries ?? 2;
+    const retryBackoffMs = deps.completionJudgerConfig?.retryBackoffMs ?? 1_000;
+    // Create review session
+    const reviewSession = await deps.sessionManager.createSession("task_review", task.goal_id, task.id);
+    // Build review context (excludes executor self-report for bias prevention)
+    const reviewContext = deps.sessionManager.buildTaskReviewContext(task.goal_id, task.id);
+    const criteriaList = task.success_criteria
+        .map((c, i) => `${i + 1}. ${c.description} (blocking: ${c.is_blocking}, method: ${c.verification_method})`)
+        .join("\n");
+    const enrichmentBlocks = [knowledgeBlock, stateBlock].filter(Boolean).join("\n");
+    const prompt = `Evaluate task execution against success criteria.
+
+Task: ${task.work_description}
+Approach: ${task.approach}
+
+Criteria:
+${criteriaList}
+${enrichmentBlocks ? `\n${enrichmentBlocks}\n` : ""}
+Output (first 2000 chars):
+${executionResult.output.slice(0, 2000)}
+
+Status: ${executionResult.stopped_reason} | Success: ${executionResult.success}
+Context: ${reviewContext.map((s) => s.content).join(" ")}
+
+Return JSON:
+{"verdict": "pass"|"partial"|"fail", "reasoning": "...", "criteria_met": #, "criteria_total": #}`;
+    // Gateway path: route through PromptGateway when available
+    if (deps.gateway) {
+        let parsed;
+        try {
+            parsed = await withRetry(() => withTimeout(deps.gateway.execute({
+                purpose: "verification",
+                goalId: task.goal_id,
+                additionalContext: { review_prompt: prompt },
+                responseSchema: CompletionJudgerResponseSchema,
+                maxTokens: 1024,
+            }), timeoutMs), maxRetries, retryBackoffMs, deps.logger, `completion_judger for task ${task.id}`);
+        }
+        catch (err) {
+            const msg = err instanceof Error ? err.message : String(err);
+            deps.logger?.error(`[completion_judger] All retries exhausted for task ${task.id}: ${msg}`);
+            await deps.sessionManager.endSession(reviewSession.id, `completion_judger failed: ${msg}`);
+            return {
+                passed: false,
+                partial: false,
+                description: `completion_judger failed after ${maxRetries + 1} attempt(s): ${msg}`,
+                confidence: 0.0,
+            };
+        }
+        const verdictStr = parsed.verdict;
+        const result = {
+            passed: verdictStr === "pass",
+            partial: verdictStr === "partial",
+            description: parsed.reasoning || "LLM review completed",
+            confidence: verdictStr === "pass" ? 0.8 : verdictStr === "partial" ? 0.6 : 0.8,
+            criteria_met: parsed.criteria_met,
+            criteria_total: parsed.criteria_total,
+        };
+        await deps.sessionManager.endSession(reviewSession.id, `LLM review: ${verdictStr}`);
+        return result;
+    }
+    // Direct LLM path (fallback when no gateway)
+    let response;
+    try {
+        response = await withRetry(() => withTimeout((deps.reviewerLlmClient ?? deps.llmClient).sendMessage([{ role: "user", content: prompt }], {
+            system: "Review task results objectively against criteria. Ignore executor self-assessment.",
+            max_tokens: 1024,
+            model_tier: modelTier,
+        }), timeoutMs), maxRetries, retryBackoffMs, deps.logger, `completion_judger for task ${task.id}`);
+    }
+    catch (err) {
+        const msg = err instanceof Error ? err.message : String(err);
+        deps.logger?.error(`[completion_judger] All retries exhausted for task ${task.id}: ${msg}`);
+        await deps.sessionManager.endSession(reviewSession.id, `completion_judger failed: ${msg}`);
+        return {
+            passed: false,
+            partial: false,
+            description: `completion_judger failed after ${maxRetries + 1} attempt(s): ${msg}`,
+            confidence: 0.0,
+        };
+    }
+    try {
+        const rawJson = response.content.replace(/```json\n?/g, "").replace(/```/g, "").trim();
+        const parseResult = CompletionJudgerResponseSchema.safeParse(JSON.parse(rawJson));
+        if (!parseResult.success) {
+            deps.logger?.warn(`[completion_judger] Zod parse failed for task ${task.id}: ${parseResult.error.message}`);
+            await deps.sessionManager.endSession(reviewSession.id, "Failed to parse LLM review result");
+            return {
+                passed: false,
+                partial: false,
+                description: "Failed to parse LLM review result",
+                confidence: 0.3,
+            };
+        }
+        const parsed = parseResult.data;
+        const verdictStr = parsed.verdict;
+        const result = {
+            passed: verdictStr === "pass",
+            partial: verdictStr === "partial",
+            description: parsed.reasoning || "LLM review completed",
+            confidence: verdictStr === "pass" ? 0.8 : verdictStr === "partial" ? 0.6 : 0.8,
+            criteria_met: parsed.criteria_met,
+            criteria_total: parsed.criteria_total,
+        };
+        await deps.sessionManager.endSession(reviewSession.id, `LLM review: ${verdictStr}`);
+        return result;
+    }
+    catch {
+        deps.logger?.warn(`[completion_judger] JSON.parse failed for task ${task.id}`);
+        await deps.sessionManager.endSession(reviewSession.id, "Failed to parse LLM review result");
+        return {
+            passed: false,
+            partial: false,
+            description: "Failed to parse LLM review result",
+            confidence: 0.3,
+        };
+    }
+}
+//# sourceMappingURL=task-verifier-llm.js.map

package/dist/execution/task/task-verifier-llm.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-verifier-llm.js","sourceRoot":"","sources":["../../../src/execution/task/task-verifier-llm.ts"],"names":[],"mappings":"AAMA,OAAO,EAAE,8BAA8B,EAAE,MAAM,0BAA0B,CAAC;AAE1E,sBAAsB;AAEtB;;;GAGG;AACH,MAAM,UAAU,WAAW,CAAI,OAAmB,EAAE,EAAU;IAC5D,OAAO,IAAI,OAAO,CAAI,CAAC,OAAO,EAAE,MAAM,EAAE,EAAE;QACxC,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;YAC5B,MAAM,CAAC,IAAI,KAAK,CAAC,mCAAmC,EAAE,IAAI,CAAC,CAAC,CAAC;QAC/D,CAAC,EAAE,EAAE,CAAC,CAAC;QACP,OAAO,CAAC,IAAI,CACV,CAAC,KAAK,EAAE,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,EACnD,CAAC,GAAG,EAAE,EAAE,GAAG,YAAY,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAC/C,CAAC;IACJ,CAAC,CAAC,CAAC;AACL,CAAC;AAED,oBAAoB;AAEpB;;;;GAIG;AACH,MAAM,CAAC,KAAK,UAAU,SAAS,CAC7B,EAAoB,EACpB,UAAkB,EAClB,SAAiB,EACjB,MAAe,EACf,KAAc;IAEd,IAAI,OAAgB,CAAC;IACrB,KAAK,IAAI,OAAO,GAAG,CAAC,EAAE,OAAO,IAAI,UAAU,EAAE,OAAO,EAAE,EAAE,CAAC;QACvD,IAAI,CAAC;YACH,OAAO,MAAM,EAAE,EAAE,CAAC;QACpB,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,OAAO,GAAG,GAAG,CAAC;YACd,IAAI,OAAO,GAAG,UAAU,EAAE,CAAC;gBACzB,MAAM,KAAK,GAAG,SAAS,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,OAAO,CAAC,CAAC;gBAC/C,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;gBAC7D,MAAM,EAAE,IAAI,CAAC,uBAAuB,KAAK,IAAI,UAAU,oBAAoB,OAAO,GAAG,CAAC,IAAI,UAAU,GAAG,CAAC,MAAM,GAAG,kBAAkB,KAAK,IAAI,CAAC,CAAC;gBAC9I,MAAM,IAAI,OAAO,CAAC,CAAC,GAAG,EAAE,EAAE,CAAC,UAAU,CAAC,GAAG,EAAE,KAAK,CAAC,CAAC,CAAC;YACrD,CAAC;QACH,CAAC;IACH,CAAC;IACD,MAAM,OAAO,CAAC;AAChB,CAAC;AAED,uBAAuB;AAEvB,MAAM,CAAC,KAAK,UAAU,YAAY,CAChC,IAAkB,EAClB,IAAU,EACV,eAA4B,EAC5B,cAAc,GAAG,EAAE,EACnB,UAAU,GAAG,EAAE,EACf,YAA8B,OAAO;IAErC,MAAM,SAAS,GAAG,IAAI,CAAC,sBAAsB,EAAE,SAAS,IAAI,MAAM,CAAC;IACnE,MAAM,UAAU,GAAG,IAAI,CAAC,sBAAsB,EAAE,UAAU,IAAI,CAAC,CAAC;IAChE,MAAM,cAAc,GAAG,IAAI,CAAC,sBAAsB,EAAE,cAAc,IAAI,KAAK,CAAC;IAE5E,wBAAwB;IACxB,MAAM,aAAa,GAAG,MAAM,IAAI,CAAC,cAAc,CAAC,aAAa,CAC3D,aAAa,EACb,IAAI,CAAC,OAAO,EACZ,IAAI,CAAC,EAAE,CACR,CAAC;IAEF,2EAA2E;IAC3E,MAAM,aAAa,GAAG,IAAI,CAAC,cAAc,CAAC,sBAAsB,CAC9D,IAAI,CAAC,OAAO,EACZ,IAAI,CAAC,EAAE,CACR,CAAC;IAEF,MAAM,YAAY,GAAG,IAAI,CAAC,gBAAgB;SACvC,GAAG,CACF,CAAC,CAAC,EAAE,CAAC,EAAE,EAAE,CACP,GAAG,CAAC,GAAG,CAAC,KAAK,CAAC,CAAC,WAAW,eAAe,CAAC,CAAC,WAAW,aAAa,CAAC,CAAC,mBAAmB,GAAG,CAC9F;SACA,IAAI,CAAC,IAAI,CAAC,CAAC;IAEd,MAAM,gBAAgB,GAAG,CAAC,cAAc,EAAE,UAAU,CAAC,CAAC,MAAM,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAEjF,MAAM,MAAM,GAAG;;QAET,IAAI,CAAC,gBAAgB;YACjB,IAAI,CAAC,QAAQ;;;EAGvB,YAAY;EACZ,gBAAgB,CAAC,CAAC,CAAC,KAAK,gBAAgB,IAAI,CAAC,CAAC,CAAC,EAAE;;EAEjD,eAAe,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,IAAI,CAAC;;UAE7B,eAAe,CAAC,cAAc,eAAe,eAAe,CAAC,OAAO;WACnE,aAAa,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,OAAO,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC;;;iGAGyC,CAAC;IAEhG,2DAA2D;IAC3D,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;QACjB,IAAI,MAAsD,CAAC;QAC3D,IAAI,CAAC;YACH,MAAM,GAAG,MAAM,SAAS,CACtB,GAAG,EAAE,CAAC,WAAW,CACf,IAAI,CAAC,OAAQ,CAAC,OAAO,CAAC;gBACpB,OAAO,EAAE,cAAc;gBACvB,MAAM,EAAE,IAAI,CAAC,OAAO;gBACpB,iBAAiB,EAAE,EAAE,aAAa,EAAE,MAAM,EAAE;gBAC5C,cAAc,EAAE,8BAA6F;gBAC7G,SAAS,EAAE,IAAI;aAChB,CAAC,EACF,SAAS,CACV,EACD,UAAU,EACV,cAAc,EACd,IAAI,CAAC,MAAM,EACX,8BAA8B,IAAI,CAAC,EAAE,EAAE,CACxC,CAAC;QACJ,CAAC;QAAC,OAAO,GAAG,EAAE,CAAC;YACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;YAC7D,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,sDAAsD,IAAI,CAAC,EAAE,KAAK,GAAG,EAAE,CAAC,CAAC;YAC5F,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,6BAA6B,GAAG,EAAE,CAAC,CAAC;YAC3F,OAAO;gBACL,MAAM,EAAE,KAAK;gBACb,OAAO,EAAE,KAAK;gBACd,WAAW,EAAE,kCAAkC,UAAU,GAAG,CAAC,gBAAgB,GAAG,EAAE;gBAClF,UAAU,EAAE,GAAG;aAChB,CAAC;QACJ,CAAC;QACD,MAAM,UAAU,GAAG,MAAM,CAAC,OAAO,CAAC;QAClC,MAAM,MAAM,GAAG;YACb,MAAM,EAAE,UAAU,KAAK,MAAM;YAC7B,OAAO,EAAE,UAAU,KAAK,SAAS;YACjC,WAAW,EAAE,MAAM,CAAC,SAAS,IAAI,sBAAsB;YACvD,UAAU,EAAE,UAAU,KAAK,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG;YAC9E,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,cAAc,EAAE,MAAM,CAAC,cAAc;SACtC,CAAC;QACF,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,eAAe,UAAU,EAAE,CAAC,CAAC;QACpF,OAAO,MAAM,CAAC;IAChB,CAAC;IAED,6CAA6C;IAC7C,IAAI,QAAqB,CAAC;IAC1B,IAAI,CAAC;QACH,QAAQ,GAAG,MAAM,SAAS,CACxB,GAAG,EAAE,CAAC,WAAW,CACf,CAAC,IAAI,CAAC,iBAAiB,IAAI,IAAI,CAAC,SAAS,CAAC,CAAC,WAAW,CACpD,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,EACnC;YACE,MAAM,EAAE,oFAAoF;YAC5F,UAAU,EAAE,IAAI;YAChB,UAAU,EAAE,SAAS;SACtB,CACF,EACD,SAAS,CACV,EACD,UAAU,EACV,cAAc,EACd,IAAI,CAAC,MAAM,EACX,8BAA8B,IAAI,CAAC,EAAE,EAAE,CACxC,CAAC;IACJ,CAAC;IAAC,OAAO,GAAG,EAAE,CAAC;QACb,MAAM,GAAG,GAAG,GAAG,YAAY,KAAK,CAAC,CAAC,CAAC,GAAG,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,GAAG,CAAC,CAAC;QAC7D,IAAI,CAAC,MAAM,EAAE,KAAK,CAAC,sDAAsD,IAAI,CAAC,EAAE,KAAK,GAAG,EAAE,CAAC,CAAC;QAC5F,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,6BAA6B,GAAG,EAAE,CAAC,CAAC;QAC3F,OAAO;YACL,MAAM,EAAE,KAAK;YACb,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,kCAAkC,UAAU,GAAG,CAAC,gBAAgB,GAAG,EAAE;YAClF,UAAU,EAAE,GAAG;SAChB,CAAC;IACJ,CAAC;IAED,IAAI,CAAC;QACH,MAAM,OAAO,GAAG,QAAQ,CAAC,OAAO,CAAC,OAAO,CAAC,aAAa,EAAE,EAAE,CAAC,CAAC,OAAO,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACvF,MAAM,WAAW,GAAG,8BAA8B,CAAC,SAAS,CAAC,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC;QAClF,IAAI,CAAC,WAAW,CAAC,OAAO,EAAE,CAAC;YACzB,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,iDAAiD,IAAI,CAAC,EAAE,KAAK,WAAW,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC,CAAC;YAC5G,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,mCAAmC,CAAC,CAAC;YAC5F,OAAO;gBACL,MAAM,EAAE,KAAK;gBACb,OAAO,EAAE,KAAK;gBACd,WAAW,EAAE,mCAAmC;gBAChD,UAAU,EAAE,GAAG;aAChB,CAAC;QACJ,CAAC;QACD,MAAM,MAAM,GAAG,WAAW,CAAC,IAAI,CAAC;QAChC,MAAM,UAAU,GAAG,MAAM,CAAC,OAAO,CAAC;QAClC,MAAM,MAAM,GAAG;YACb,MAAM,EAAE,UAAU,KAAK,MAAM;YAC7B,OAAO,EAAE,UAAU,KAAK,SAAS;YACjC,WAAW,EAAE,MAAM,CAAC,SAAS,IAAI,sBAAsB;YACvD,UAAU,EAAE,UAAU,KAAK,MAAM,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,UAAU,KAAK,SAAS,CAAC,CAAC,CAAC,GAAG,CAAC,CAAC,CAAC,GAAG;YAC9E,YAAY,EAAE,MAAM,CAAC,YAAY;YACjC,cAAc,EAAE,MAAM,CAAC,cAAc;SACtC,CAAC;QACF,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,eAAe,UAAU,EAAE,CAAC,CAAC;QACpF,OAAO,MAAM,CAAC;IAChB,CAAC;IAAC,MAAM,CAAC;QACP,IAAI,CAAC,MAAM,EAAE,IAAI,CAAC,kDAAkD,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC;QAC/E,MAAM,IAAI,CAAC,cAAc,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,EAAE,mCAAmC,CAAC,CAAC;QAC5F,OAAO;YACL,MAAM,EAAE,KAAK;YACb,OAAO,EAAE,KAAK;YACd,WAAW,EAAE,mCAAmC;YAChD,UAAU,EAAE,GAAG;SAChB,CAAC;IACJ,CAAC;AACH,CAAC"}

package/dist/execution/task/task-verifier-rules 2.d.ts

@@ -0,0 +1,31 @@
+import type { Task } from "../../types/task.js";
+import type { VerificationResult } from "../../types/task.js";
+import type { AgentResult } from "../adapter-layer.js";
+import type { VerifierDeps } from "./task-verifier-types.js";
+export declare function runMechanicalVerification(deps: VerifierDeps, task: Task): Promise<{
+    applicable: boolean;
+    passed: boolean;
+    description: string;
+}>;
+/**
+ * Clamp a proposed dimension update to within ±30% absolute or ±30% relative
+ * of the current value (whichever is larger). Logs a warning when clamping occurs.
+ *
+ * Exported for unit testing.
+ */
+export declare function clampDimensionUpdate(current: number, proposed: number, logger?: import("../../runtime/logger.js").Logger, dimName?: string): number;
+/**
+ * Check whether a proposed dimension update moves in the intended direction.
+ * Returns true if the update should be applied, false if it should be skipped.
+ *
+ * Exported for unit testing.
+ */
+export declare function checkDimensionDirection(intendedDirection: "increase" | "decrease" | "neutral" | undefined, currentValue: number, proposedValue: number, logger?: {
+    warn: (msg: string) => void;
+}, dimName?: string): boolean;
+export declare function parseExecutorReport(executionResult: AgentResult): import("./task-verifier-types.js").ExecutorReport;
+export declare function isDirectionCorrect(verificationResult: VerificationResult): boolean;
+export declare function attemptRevert(deps: VerifierDeps, task: Task): Promise<boolean>;
+export declare function setDimensionIntegrity(deps: VerifierDeps, goalId: string, dimensionName: string, integrity: "ok" | "uncertain"): Promise<void>;
+export declare function appendTaskHistory(deps: VerifierDeps, goalId: string, task: Task): Promise<void>;
+//# sourceMappingURL=task-verifier-rules%202.d.ts.map

package/dist/execution/task/task-verifier-rules 2.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-verifier-rules 2.d.ts","sourceRoot":"","sources":["../../../src/execution/task/task-verifier-rules 2.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,IAAI,EAAE,MAAM,qBAAqB,CAAC;AAChD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,qBAAqB,CAAC;AAC9D,OAAO,KAAK,EAAa,WAAW,EAAY,MAAM,qBAAqB,CAAC;AAC5E,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,0BAA0B,CAAC;AAI7D,wBAAsB,yBAAyB,CAC7C,IAAI,EAAE,YAAY,EAClB,IAAI,EAAE,IAAI,GACT,OAAO,CAAC;IAAE,UAAU,EAAE,OAAO,CAAC;IAAC,MAAM,EAAE,OAAO,CAAC;IAAC,WAAW,EAAE,MAAM,CAAA;CAAE,CAAC,CAsFxE;AAID;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,MAAM,EACf,QAAQ,EAAE,MAAM,EAChB,MAAM,CAAC,EAAE,OAAO,yBAAyB,EAAE,MAAM,EACjD,OAAO,CAAC,EAAE,MAAM,GACf,MAAM,CAWR;AAID;;;;;GAKG;AACH,wBAAgB,uBAAuB,CACrC,iBAAiB,EAAE,UAAU,GAAG,UAAU,GAAG,SAAS,GAAG,SAAS,EAClE,YAAY,EAAE,MAAM,EACpB,aAAa,EAAE,MAAM,EACrB,MAAM,CAAC,EAAE;IAAE,IAAI,EAAE,CAAC,GAAG,EAAE,MAAM,KAAK,IAAI,CAAA;CAAE,EACxC,OAAO,CAAC,EAAE,MAAM,GACf,OAAO,CAuBT;AAID,wBAAgB,mBAAmB,CAAC,eAAe,EAAE,WAAW,GAAG,OAAO,0BAA0B,EAAE,cAAc,CAOnH;AAID,wBAAgB,kBAAkB,CAAC,kBAAkB,EAAE,kBAAkB,GAAG,OAAO,CAElF;AAID,wBAAsB,aAAa,CAAC,IAAI,EAAE,YAAY,EAAE,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,OAAO,CAAC,CA2CpF;AAID,wBAAsB,qBAAqB,CACzC,IAAI,EAAE,YAAY,EAClB,MAAM,EAAE,MAAM,EACd,aAAa,EAAE,MAAM,EACrB,SAAS,EAAE,IAAI,GAAG,WAAW,GAC5B,OAAO,CAAC,IAAI,CAAC,CAcf;AAID,wBAAsB,iBAAiB,CAAC,IAAI,EAAE,YAAY,EAAE,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,IAAI,GAAG,OAAO,CAAC,IAAI,CAAC,CAwBrG"}

package/dist/execution/task/task-verifier-rules 2.js

@@ -0,0 +1,207 @@
+import { z } from "zod";
+// ─── runMechanicalVerification ───
+export async function runMechanicalVerification(deps, task) {
+    // Mechanical prefixes that indicate a command can be run directly
+    const mechanicalPrefixes = ["npm", "npx", "pytest", "sh", "bash", "node", "make", "cargo", "go ", "gh "];
+    // Find the first success criterion with a mechanically-verifiable verification_method
+    const mechanicalCriterion = task.success_criteria.find((c) => {
+        const method = c.verification_method.toLowerCase().trim();
+        return mechanicalPrefixes.some((prefix) => method.startsWith(prefix));
+    });
+    if (!mechanicalCriterion) {
+        return {
+            applicable: false,
+            passed: false,
+            description: "No mechanical verification criteria applicable",
+        };
+    }
+    // If no adapter registry is available, fall back to assumed pass (backward compat)
+    if (!deps.adapterRegistry) {
+        return {
+            applicable: true,
+            passed: true,
+            description: "Mechanical verification criteria detected (no adapter: assumed pass)",
+        };
+    }
+    // Select the first available adapter from the registry for command execution
+    const availableAdapters = deps.adapterRegistry.listAdapters();
+    if (availableAdapters.length === 0) {
+        return {
+            applicable: true,
+            passed: true,
+            description: "Mechanical verification criteria detected (no adapters registered: assumed pass)",
+        };
+    }
+    const adapterType = availableAdapters[0];
+    let adapter;
+    try {
+        adapter = deps.adapterRegistry.getAdapter(adapterType);
+    }
+    catch {
+        return {
+            applicable: true,
+            passed: true,
+            description: "Mechanical verification criteria detected (adapter lookup failed: assumed pass)",
+        };
+    }
+    // Execute the verification command via the adapter
+    const verificationCommand = mechanicalCriterion.verification_method.trim();
+    const verificationTimeoutMs = 30_000; // 30 seconds default for L1 mechanical checks
+    const agentTask = {
+        prompt: verificationCommand,
+        timeout_ms: verificationTimeoutMs,
+        adapter_type: adapterType,
+    };
+    let result;
+    try {
+        result = await adapter.execute(agentTask);
+    }
+    catch (err) {
+        const errMsg = err instanceof Error ? err.message : String(err);
+        deps.logger?.error("runMechanicalVerification: adapter.execute() threw", { error: errMsg });
+        return {
+            applicable: true,
+            passed: false,
+            description: `Mechanical verification command threw: ${errMsg}`,
+        };
+    }
+    if (result.stopped_reason === "timeout") {
+        return {
+            applicable: true,
+            passed: false,
+            description: `Mechanical verification timed out after ${verificationTimeoutMs}ms (command: ${verificationCommand})`,
+        };
+    }
+    const passed = result.exit_code === 0 && result.success;
+    const description = passed
+        ? `Mechanical verification passed (exit 0): ${verificationCommand}`
+        : `Mechanical verification failed (exit ${result.exit_code ?? "null"}): ${verificationCommand}${result.error ? ` — ${result.error}` : ""}`;
+    return { applicable: true, passed, description };
+}
+// ─── P0 Guard 1: dimension_updates change magnitude limit (§3.2) ───
+/**
+ * Clamp a proposed dimension update to within ±30% absolute or ±30% relative
+ * of the current value (whichever is larger). Logs a warning when clamping occurs.
+ *
+ * Exported for unit testing.
+ */
+export function clampDimensionUpdate(current, proposed, logger, dimName) {
+    const absLimit = 0.3;
+    const relLimit = Math.abs(current) * 0.3;
+    const maxDelta = Math.max(absLimit, relLimit);
+    const clamped = Math.max(current - maxDelta, Math.min(current + maxDelta, proposed));
+    if (clamped !== proposed) {
+        logger?.warn(`dimension_update clamped: dim=${dimName}, proposed=${proposed}, applied=${clamped}, current=${current}`);
+    }
+    return clamped;
+}
+// ─── §4.5 Guard: dimension_updates direction check ───
+/**
+ * Check whether a proposed dimension update moves in the intended direction.
+ * Returns true if the update should be applied, false if it should be skipped.
+ *
+ * Exported for unit testing.
+ */
+export function checkDimensionDirection(intendedDirection, currentValue, proposedValue, logger, dimName) {
+    if (!intendedDirection || intendedDirection === "neutral")
+        return true;
+    const actualDirection = proposedValue > currentValue
+        ? "increase"
+        : proposedValue < currentValue
+            ? "decrease"
+            : "neutral";
+    if (intendedDirection === "increase" && actualDirection === "decrease") {
+        logger?.warn(`dimension_update direction mismatch: task intended ${intendedDirection}, but update suggests ${actualDirection} for dim ${dimName ?? "unknown"}`);
+        return false;
+    }
+    if (intendedDirection === "decrease" && actualDirection === "increase") {
+        logger?.warn(`dimension_update direction mismatch: task intended ${intendedDirection}, but update suggests ${actualDirection} for dim ${dimName ?? "unknown"}`);
+        return false;
+    }
+    return true;
+}
+// ─── parseExecutorReport ───
+export function parseExecutorReport(executionResult) {
+    return {
+        completed: executionResult.success,
+        summary: executionResult.output.slice(0, 500),
+        partial_results: [],
+        blockers: executionResult.error ? [executionResult.error] : [],
+    };
+}
+// ─── isDirectionCorrect ───
+export function isDirectionCorrect(verificationResult) {
+    return verificationResult.verdict === "partial";
+}
+// ─── attemptRevert ───
+export async function attemptRevert(deps, task) {
+    try {
+        const filesToRestore = task.scope_boundary.in_scope;
+        if (filesToRestore.length > 0) {
+            const { execFileSync } = await import("child_process");
+            execFileSync("git", ["restore", ...filesToRestore], { cwd: process.cwd(), encoding: "utf-8", stdio: ["pipe", "pipe", "pipe"] });
+            deps.logger?.info?.(`[attemptRevert] git restore succeeded for ${filesToRestore.length} files`);
+            return true;
+        }
+    }
+    catch {
+        // git not available or failed — fall back to LLM-based revert
+    }
+    try {
+        const revertSession = await deps.sessionManager.createSession("task_execution", task.goal_id, task.id);
+        const revertPrompt = `Revert task "${task.work_description}". Undo all changes in: ${task.scope_boundary.in_scope.join(", ")}.
+
+Return JSON: {"success": true|false, "reason": "..."}`;
+        const response = await deps.llmClient.sendMessage([{ role: "user", content: revertPrompt }], { system: "Revert failed task changes. Respond with JSON only.", max_tokens: 512, model_tier: "main" });
+        await deps.sessionManager.endSession(revertSession.id, response.content);
+        try {
+            const parsed = deps.llmClient.parseJSON(response.content, z.object({ success: z.boolean(), reason: z.string() }));
+            return parsed.success;
+        }
+        catch {
+            return false;
+        }
+    }
+    catch {
+        return false;
+    }
+}
+// ─── setDimensionIntegrity ───
+export async function setDimensionIntegrity(deps, goalId, dimensionName, integrity) {
+    const goalData = await deps.stateManager.readRaw(`goals/${goalId}/goal.json`);
+    if (goalData && typeof goalData === "object") {
+        const goal = goalData;
+        const dimensions = goal.dimensions;
+        if (dimensions) {
+            for (const dim of dimensions) {
+                if (dim.name === dimensionName) {
+                    dim.state_integrity = integrity;
+                }
+            }
+            await deps.stateManager.writeRaw(`goals/${goalId}/goal.json`, goal);
+        }
+    }
+}
+// ─── appendTaskHistory ───
+export async function appendTaskHistory(deps, goalId, task) {
+    const historyPath = `tasks/${goalId}/task-history.json`;
+    const existing = await deps.stateManager.readRaw(historyPath);
+    const history = Array.isArray(existing) ? existing : [];
+    const actual_elapsed_ms = task.started_at && task.completed_at
+        ? new Date(task.completed_at).getTime() - new Date(task.started_at).getTime()
+        : null;
+    const estimated_duration_ms = task.estimated_duration
+        ? deps.durationToMs(task.estimated_duration)
+        : null;
+    history.push({
+        task_id: task.id,
+        status: task.status,
+        primary_dimension: task.primary_dimension,
+        consecutive_failure_count: task.consecutive_failure_count,
+        completed_at: task.completed_at ?? new Date().toISOString(),
+        actual_elapsed_ms,
+        estimated_duration_ms,
+    });
+    await deps.stateManager.writeRaw(historyPath, history);
+}
+//# sourceMappingURL=task-verifier-rules%202.js.map