codeharness 0.34.0 → 0.34.1

package/dist/{chunk-QMWMRFGH.js → chunk-G2RR744K.js}

@@ -2895,7 +2895,7 @@ function generateDockerfileTemplate(projectDir, stackOrDetections) {
 }
 
 // src/modules/infra/init-project.ts
-var HARNESS_VERSION = true ? "0.34.0" : "0.0.0-dev";
+var HARNESS_VERSION = true ? "0.34.1" : "0.0.0-dev";
 function failResult(opts, error) {
   return {
     status: "fail",

package/dist/{docker-QYSLNCQ2.js → docker-MEPWHG4P.js}

@@ -16,7 +16,7 @@ import {
   stopCollectorOnly,
   stopSharedStack,
   stopStack
-} from "./chunk-QMWMRFGH.js";
+} from "./chunk-G2RR744K.js";
 export {
   checkRemoteEndpoint,
   cleanupOrphanedContainers,

package/dist/index.js

@@ -40,7 +40,7 @@ import {
   validateDockerfile,
   warn,
   writeState
-} from "./chunk-QMWMRFGH.js";
+} from "./chunk-G2RR744K.js";
 
 // src/index.ts
 import { Command } from "commander";
@@ -3068,11 +3068,15 @@ function parseVerdict(output) {
   return verdict;
 }
 function parseSimpleVerdict(output) {
-  const jsonPattern = /\{[^{}]*"verdict"\s*:\s*"(pass|fail)"[^{}]*\}/;
-  const match = jsonPattern.exec(output);
-  if (!match) return null;
+  const jsonPattern = /\{[^{}]*"verdict"\s*:\s*"(pass|fail)"[^{}]*\}/g;
+  let lastMatch = null;
+  let m;
+  while ((m = jsonPattern.exec(output)) !== null) {
+    lastMatch = m;
+  }
+  if (!lastMatch) return null;
   try {
-    const parsed = JSON.parse(match[0]);
+    const parsed = JSON.parse(lastMatch[0]);
     if (parsed.verdict === "pass" || parsed.verdict === "fail") {
       return { verdict: parsed.verdict };
     }
@@ -3384,8 +3388,8 @@ async function dispatchTaskWithResult(task, taskName, storyKey, definition, stat
   }
   const isEpicSentinel = storyKey.startsWith("__epic_") || storyKey === PER_RUN_SENTINEL;
   const TASK_PROMPTS = {
-    "create-story": (key) => `Create the story spec for ${key}. Read the epic definitions and architecture docs, then write a complete story file with acceptance criteria, tasks, and dev notes.`,
-    "negotiate-acs": (key) => `Review the ACs in story ${key} for testability. Can each AC be verified by a blind QA agent with only Docker access and user documentation? Output a verdict JSON.`,
+    "create-story": (key) => `Create or revise the story spec for ${key}. Read the epic definitions and architecture docs. If previous feedback is provided (from AC negotiation or review), revise the story to address that feedback. Write a complete story file with acceptance criteria, tasks, and dev notes.`,
+    "negotiate-acs": (key) => `Review the ACs in story ${key} for testability. Can each AC be verified by a blind QA agent with only Docker access and user documentation? Your response MUST end with exactly one JSON line: {"verdict": "pass"} or {"verdict": "fail", "issues": ["..."]}`,
     "implement": (key) => `Implement story ${key}`,
     "check": (key) => `Run automated checks for story ${key}. Execute the project's test suite, linter, and coverage tool. Report pass/fail results.`,
     "review": (key) => `Review the implementation of story ${key}. Check for correctness, security issues, architecture violations, and AC coverage. Output a verdict JSON at the end.`,
@@ -11185,7 +11189,7 @@ function registerTeardownCommand(program) {
     } else if (otlpMode === "remote-routed") {
       if (!options.keepDocker) {
         try {
-          const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-QYSLNCQ2.js");
+          const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-MEPWHG4P.js");
           stopCollectorOnly2();
           result.docker.stopped = true;
           if (!isJson) {
@@ -11217,7 +11221,7 @@ function registerTeardownCommand(program) {
         info("Shared stack: kept running (other projects may use it)");
       }
     } else if (isLegacyStack) {
-      const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-QYSLNCQ2.js");
+      const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-MEPWHG4P.js");
       let stackRunning = false;
       try {
         stackRunning = isStackRunning2(composeFile);
@@ -14204,7 +14208,7 @@ function registerDriversCommand(program) {
 }
 
 // src/index.ts
-var VERSION = true ? "0.34.0" : "0.0.0-dev";
+var VERSION = true ? "0.34.1" : "0.0.0-dev";
 function createProgram() {
   const program = new Command();
   program.name("codeharness").description("Makes autonomous coding agents produce software that actually works").version(VERSION).option("--json", "Output in machine-readable JSON format");

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "codeharness",
-  "version": "0.34.0",
+  "version": "0.34.1",
   "type": "module",
   "description": "CLI for codeharness — makes autonomous coding agents produce software that actually works",
   "bin": {

package/templates/agents/negotiator.yaml

@@ -29,16 +29,14 @@ prompt_template: |
      - Observable output (logs, responses, behavior)
   3. If an AC requires reading source code, inspecting file contents, or understanding implementation details to verify — it FAILS testability
 
-  ## Output
+  ## Output — MANDATORY FORMAT
 
-  If ALL ACs are testable, output:
-  ```json
+  Your response MUST end with EXACTLY one of these two JSON lines (no code block, no markdown, just raw JSON as the LAST line of your output):
+
+  If ALL ACs are testable:
   {"verdict": "pass"}
-  ```
 
-  If ANY AC fails testability, output:
-  ```json
-  {"verdict": "fail", "issues": ["AC N: [reason it's untestable]. Suggested rewrite: [specific rewrite]"]}
-  ```
+  If ANY AC fails testability:
+  {"verdict": "fail", "issues": ["AC 1: reason and suggested rewrite", "AC 3: reason and suggested rewrite"]}
 
-  Be specific. Don't just say "untestable" — explain WHY and provide a concrete rewrite.
+  You may include analysis BEFORE the verdict line, but the LAST line of your response MUST be the raw JSON verdict. This is machine-parsed — the loop cannot exit without it.