codeharness 0.35.0 → 0.35.2

package/dist/{chunk-PUZ5PWYL.js → chunk-KJF2YA5T.js}

@@ -2895,7 +2895,7 @@ function generateDockerfileTemplate(projectDir, stackOrDetections) {
 }
 
 // src/modules/infra/init-project.ts
-var HARNESS_VERSION = true ? "0.35.0" : "0.0.0-dev";
+var HARNESS_VERSION = true ? "0.35.2" : "0.0.0-dev";
 function failResult(opts, error) {
   return {
     status: "fail",

package/dist/{docker-CTLGRLXP.js → docker-EIWOFRFK.js}

@@ -16,7 +16,7 @@ import {
   stopCollectorOnly,
   stopSharedStack,
   stopStack
-} from "./chunk-PUZ5PWYL.js";
+} from "./chunk-KJF2YA5T.js";
 export {
   checkRemoteEndpoint,
   cleanupOrphanedContainers,

package/dist/index.js

@@ -40,7 +40,7 @@ import {
   validateDockerfile,
   warn,
   writeState
-} from "./chunk-PUZ5PWYL.js";
+} from "./chunk-KJF2YA5T.js";
 
 // src/index.ts
 import { Command } from "commander";
@@ -3386,8 +3386,7 @@ async function dispatchTaskWithResult(task, taskName, storyKey, definition, stat
   }
   const isEpicSentinel = storyKey.startsWith("__epic_") || storyKey === PER_RUN_SENTINEL;
   const TASK_PROMPTS = {
-    "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. Wrap output in <story-spec>...</story-spec> tags.`,
-    "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? Include <verdict>pass</verdict> or <verdict>fail</verdict> in your response. If fail, include <issues>...</issues> with specific feedback per AC.`,
+    "create-story": (key) => `Create the story spec for ${key}. Read the epic definitions and architecture docs. Write a complete story file with acceptance criteria, tasks, and dev notes. CRITICAL: Every AC must be testable by a blind QA agent using ONLY a user guide + browser/API/CLI access. No AC should reference source code, internal data structures, or implementation details like O(1) complexity. Each AC must describe observable behavior that can be verified through UI interaction (agent-browser), API calls (curl), CLI commands (docker exec), or log inspection (docker logs). Wrap output in <story-spec>...</story-spec> tags.`,
     "implement": (key) => `Implement story ${key}`,
     "check": (key) => `Run automated checks for story ${key}. Execute the project's test suite, linter, and coverage tool. Include <verdict>pass</verdict> or <verdict>fail</verdict> in your response.`,
     "review": (key) => `Review the implementation of story ${key}. Check for correctness, security issues, architecture violations, and AC coverage. Include <verdict>pass</verdict> or <verdict>fail</verdict> in your response. If fail, include <issues>...</issues>.`,
@@ -3610,6 +3609,14 @@ async function executeLoopBlock(loopBlock, state, config, workItems, initialCont
   if (loopBlock.loop.length === 0) {
     return { state: currentState, errors, tasksCompleted, halted: false, lastContract: lastOutputContract };
   }
+  const lastAgentTaskInLoop = (() => {
+    for (let i = loopBlock.loop.length - 1; i >= 0; i--) {
+      const tn = loopBlock.loop[i];
+      const t = config.workflow.tasks[tn];
+      if (t && t.agent !== null) return tn;
+    }
+    return loopBlock.loop[loopBlock.loop.length - 1];
+  })();
   while (true) {
     const nextIteration = currentState.iteration + 1;
     const allCurrentIterationDone = currentState.iteration > 0 && loopBlock.loop.every((tn) => {
@@ -3752,88 +3759,91 @@ async function executeLoopBlock(loopBlock, state, config, workItems, initialCont
           propagateVerifyFlags(taskName, dispatchResult.contract, projectDir);
           accumulatedCostUsd += dispatchResult.contract?.cost_usd ?? 0;
           tasksCompleted++;
-          let verdict = null;
-          try {
-            verdict = parseVerdict(dispatchResult.output);
-          } catch (parseErr) {
-            if (parseErr instanceof VerdictParseError && parseErr.retryable) {
-              warn(`workflow-engine: verdict parse failed, retrying evaluator for ${taskName}`);
-              try {
-                const retryResult = await dispatchTaskWithResult(
-                  task,
-                  taskName,
-                  PER_RUN_SENTINEL,
-                  definition,
-                  currentState,
-                  config,
-                  void 0,
-                  lastOutputContract ?? void 0
-                );
-                currentState = retryResult.updatedState;
-                lastOutputContract = retryResult.contract;
-                propagateVerifyFlags(taskName, retryResult.contract, projectDir);
-                tasksCompleted++;
-                verdict = parseVerdict(retryResult.output);
-              } catch {
-                verdict = buildAllUnknownVerdict(
-                  workItems,
-                  "Evaluator failed to produce valid JSON after retry"
-                );
+          const isLastTaskInLoop = taskName === lastAgentTaskInLoop;
+          if (isLastTaskInLoop) {
+            let verdict = null;
+            try {
+              verdict = parseVerdict(dispatchResult.output);
+            } catch (parseErr) {
+              if (parseErr instanceof VerdictParseError && parseErr.retryable) {
+                warn(`workflow-engine: verdict parse failed, retrying evaluator for ${taskName}`);
+                try {
+                  const retryResult = await dispatchTaskWithResult(
+                    task,
+                    taskName,
+                    PER_RUN_SENTINEL,
+                    definition,
+                    currentState,
+                    config,
+                    void 0,
+                    lastOutputContract ?? void 0
+                  );
+                  currentState = retryResult.updatedState;
+                  lastOutputContract = retryResult.contract;
+                  propagateVerifyFlags(taskName, retryResult.contract, projectDir);
+                  tasksCompleted++;
+                  verdict = parseVerdict(retryResult.output);
+                } catch {
+                  verdict = buildAllUnknownVerdict(
+                    workItems,
+                    "Evaluator failed to produce valid JSON after retry"
+                  );
+                }
               }
             }
-          }
-          if (!verdict) {
-            const tagged = parseVerdictTag(dispatchResult.output);
-            if (tagged) {
-              verdict = {
-                verdict: tagged.verdict,
-                score: { passed: tagged.verdict === "pass" ? 1 : 0, failed: tagged.verdict === "fail" ? 1 : 0, unknown: 0, total: 1 },
-                findings: []
+            if (!verdict) {
+              const tagged = parseVerdictTag(dispatchResult.output);
+              if (tagged) {
+                verdict = {
+                  verdict: tagged.verdict,
+                  score: { passed: tagged.verdict === "pass" ? 1 : 0, failed: tagged.verdict === "fail" ? 1 : 0, unknown: 0, total: 1 },
+                  findings: []
+                };
+              }
+            }
+            lastVerdict = verdict;
+            if (verdict) {
+              const score = {
+                iteration: currentState.iteration,
+                passed: verdict.score.passed,
+                failed: verdict.score.failed,
+                unknown: verdict.score.unknown,
+                total: verdict.score.total,
+                timestamp: (/* @__PURE__ */ new Date()).toISOString()
+              };
+              currentState = {
+                ...currentState,
+                evaluator_scores: [...currentState.evaluator_scores, score]
+              };
+            } else {
+              const totalItems = workItems.length;
+              const score = {
+                iteration: currentState.iteration,
+                passed: 0,
+                failed: 0,
+                unknown: totalItems,
+                total: totalItems,
+                timestamp: (/* @__PURE__ */ new Date()).toISOString()
+              };
+              currentState = {
+                ...currentState,
+                evaluator_scores: [...currentState.evaluator_scores, score]
               };
             }
-          }
-          lastVerdict = verdict;
-          if (verdict) {
-            const score = {
-              iteration: currentState.iteration,
-              passed: verdict.score.passed,
-              failed: verdict.score.failed,
-              unknown: verdict.score.unknown,
-              total: verdict.score.total,
-              timestamp: (/* @__PURE__ */ new Date()).toISOString()
-            };
-            currentState = {
-              ...currentState,
-              evaluator_scores: [...currentState.evaluator_scores, score]
-            };
-          } else {
-            const totalItems = workItems.length;
-            const score = {
-              iteration: currentState.iteration,
-              passed: 0,
-              failed: 0,
-              unknown: totalItems,
-              total: totalItems,
-              timestamp: (/* @__PURE__ */ new Date()).toISOString()
-            };
-            currentState = {
-              ...currentState,
-              evaluator_scores: [...currentState.evaluator_scores, score]
-            };
+            const cbDecision = evaluateProgress(currentState.evaluator_scores);
+            if (cbDecision.halt) {
+              currentState = {
+                ...currentState,
+                circuit_breaker: {
+                  triggered: true,
+                  reason: cbDecision.reason,
+                  score_history: cbDecision.scoreHistory
+                }
+              };
+              writeWorkflowState(currentState, projectDir);
+            }
           }
           writeWorkflowState(currentState, projectDir);
-          const cbDecision = evaluateProgress(currentState.evaluator_scores);
-          if (cbDecision.halt) {
-            currentState = {
-              ...currentState,
-              circuit_breaker: {
-                triggered: true,
-                reason: cbDecision.reason,
-                score_history: cbDecision.scoreHistory
-              }
-            };
-            writeWorkflowState(currentState, projectDir);
-          }
         } catch (err) {
           const engineError = handleDispatchError(err, taskName, PER_RUN_SENTINEL);
           errors.push(engineError);
@@ -11189,7 +11199,7 @@ function registerTeardownCommand(program) {
     } else if (otlpMode === "remote-routed") {
       if (!options.keepDocker) {
         try {
-          const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-CTLGRLXP.js");
+          const { stopCollectorOnly: stopCollectorOnly2 } = await import("./docker-EIWOFRFK.js");
           stopCollectorOnly2();
           result.docker.stopped = true;
           if (!isJson) {
@@ -11221,7 +11231,7 @@ function registerTeardownCommand(program) {
         info("Shared stack: kept running (other projects may use it)");
       }
     } else if (isLegacyStack) {
-      const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-CTLGRLXP.js");
+      const { isStackRunning: isStackRunning2, stopStack } = await import("./docker-EIWOFRFK.js");
       let stackRunning = false;
       try {
         stackRunning = isStackRunning2(composeFile);
@@ -14208,7 +14218,7 @@ function registerDriversCommand(program) {
 }
 
 // src/index.ts
-var VERSION = true ? "0.35.0" : "0.0.0-dev";
+var VERSION = true ? "0.35.2" : "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": "0.35.0",
+  "version": "0.35.2",
   "type": "module",
   "description": "CLI for codeharness — makes autonomous coding agents produce software that actually works",
   "bin": {

package/templates/workflows/default.yaml

@@ -4,11 +4,6 @@ tasks:
     session: fresh
     source_access: true
     model: claude-opus-4-6
-  negotiate-acs:
-    agent: negotiator
-    session: fresh
-    source_access: true
-    model: claude-sonnet-4-6
   implement:
     agent: dev
     session: fresh
@@ -52,10 +47,6 @@ tasks:
 
 story_flow:
   - create-story
-  - negotiate-acs
-  - loop:
-      - create-story
-      - negotiate-acs
   - implement
   - check
   - review