@workermill/agent 0.8.2 → 0.8.5

package/dist/plan-validator.d.ts

@@ -18,7 +18,7 @@ export interface PlannedStory {
     priority: number;
     estimatedEffort: "small" | "medium" | "large";
     dependencies: string[];
-    acceptanceCriteria: string[];
+    acceptanceCriteria?: string[];
     targetFiles?: string[];
     scope?: string;
 }
@@ -39,7 +39,7 @@ export interface CriticResult {
         suggestedChanges?: string[];
     }>;
 }
-declare const AUTO_APPROVAL_THRESHOLD = 80;
+declare const AUTO_APPROVAL_THRESHOLD = 85;
 /**
  * Parse execution plan JSON from raw Claude CLI output.
  * Mirrors server-side parseExecutionPlan() in planning-agent-local.ts.

package/dist/plan-validator.js

@@ -17,7 +17,7 @@ import { api } from "./api.js";
 // CONSTANTS
 // ============================================================================
 const MAX_TARGET_FILES = 15;
-const AUTO_APPROVAL_THRESHOLD = 80;
+const AUTO_APPROVAL_THRESHOLD = 85;
 // ============================================================================
 // PLAN PARSING
 // ============================================================================
@@ -217,11 +217,13 @@ Review this execution plan against the PRD:
 
 **DO check for:**
 1. **Missing Requirements** - Does the plan cover what the PRD asks for?
-2. **Vague Instructions** - Will the worker know what to do?
+2. **Scope Clarity** - Is each story's description a brief file scope label (1 line)? Stories should NOT rewrite ticket requirements.
 3. **Security Issues** - Only for tasks involving auth, user data, or external input
-4. **Unrealistic Scope** - Any step targeting >5 files MUST score below 80 (auto-rejection threshold). Each step should modify at most 5 files. If a step needs more, split it into multiple steps first.
+4. **Unrealistic Scope** - Any step targeting >5 files MUST score below 85 (auto-rejection threshold). Each step should modify at most 5 files. If a step needs more, split it into multiple steps first.
 5. **Missing Operational Steps** - If the PRD requires deployment, provisioning, migrations, or running commands, does the plan include operational steps? Writing code is not the same as deploying it.
 6. **Overlapping File Scope** - If two or more steps share the same targetFiles, this causes parallel merge conflicts. Steps MUST NOT overlap on targetFiles. Deduct 10 points per shared file across steps.
+7. **Serialization Bottleneck** - If more than half the stories depend on a single story that targets >5 files, the plan has a bottleneck. Deduct 15 points — split the foundation or allow more parallel work.
+8. **Requirement Rewriting** - If any story description contains implementation details, acceptance criteria, or rewritten requirements from the PRD, deduct 15 points per offending story. Story descriptions must be ONE-LINE file scope labels (e.g., "Database layer — migrations and entity definitions"). The original ticket is the spec.
 
 ## Scoring Guide
 
@@ -236,7 +238,7 @@ Respond with ONLY a JSON object (no markdown, no explanation):
 {"approved": boolean, "score": number, "risks": ["risk1", "risk2"], "suggestions": ["suggestion1", "suggestion2"], "storyFeedback": [{"storyId": "step-0", "feedback": "specific feedback", "suggestedChanges": ["change1"]}]}
 
 Rules:
-- approved = true if score >= 80 AND plan is right-sized for task
+- approved = true if score >= 85 AND plan is right-sized for task
 - risks = specific issues (empty array if none)
 - suggestions = actionable improvements (empty array if none)
 - storyFeedback = per-step feedback (optional, only for steps that need changes)`;

package/dist/planner.js

@@ -80,9 +80,10 @@ const logQueue = [];
 let logDrainPromise = null;
 async function drainLogQueue() {
     while (logQueue.length > 0) {
-        const entry = logQueue.shift();
+        // Drain up to 50 entries per batch POST
+        const batch = logQueue.splice(0, 50);
         try {
-            await api.post("/api/control-center/logs", entry, { timeout: 5_000 });
+            await api.post("/api/control-center/logs/batch", { entries: batch }, { timeout: 5_000 });
         }
         catch {
             // Best-effort — drop on failure
@@ -218,7 +219,7 @@ function runClaudeCli(claudePath, model, prompt, env, taskId, startTime, cwd) {
             console.log(`${ts()} ${taskLabel} ${chalk.dim(msg)}`);
         }
         // Flush buffered LLM text to dashboard every 1s (complete lines only)
-        const textFlushInterval = setInterval(() => flushTextBuffer(), 1_000);
+        const textFlushInterval = setInterval(() => flushTextBuffer(), 500);
         // SSE progress updates every 2s — drives PlanningTerminalBar in dashboard
         // (same cadence as local dev's progressInterval in planning-agent-local.ts)
         const sseProgressInterval = setInterval(() => {
@@ -624,9 +625,9 @@ export async function planTask(task, config, credentials) {
             // 2e. Check critic result
             if (!criticResult) {
                 // Critic failed (timeout, parse error, etc.) — post plan without critic gate
-                const msg = `${PREFIX} Critic validation failed — posting plan without critic score`;
+                const msg = `${PREFIX} ⚠️ CRITIC BYPASSED — Critic validation failed (timeout/parse error). Posting plan WITHOUT quality gate.`;
                 console.log(`${ts()} ${taskLabel} ${chalk.yellow("⚠")} ${msg}`);
-                await postLog(task.id, msg);
+                await postLog(task.id, msg, "error", "warning");
                 const planningDurationMs = Date.now() - startTime;
                 return await postValidatedPlan(task.id, plan, config.agentId, taskLabel, elapsed, undefined, undefined, criticHistory, totalFileCapTruncations, planningDurationMs, iteration);
             }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@workermill/agent",
-  "version": "0.8.2",
+  "version": "0.8.5",
   "description": "WorkerMill Remote Agent - Run AI workers locally with your Claude Max subscription",
   "type": "module",
   "main": "./dist/index.js",