karajan-code 2.4.1 → 2.6.0

package/src/orchestrator/flow-runner.js

@@ -0,0 +1,2085 @@
+import { createAgent } from "../agents/index.js";
+import {
+  loadSession,
+  markSessionStatus,
+  pauseSession,
+  resumeSessionWithAnswer,
+  saveSession,
+  addCheckpoint
+} from "../session-store.js";
+import { generateDiff } from "../review/diff-generator.js";
+import { resolveRole } from "../config.js";
+import { resolveReviewProfile } from "../review/profiles.js";
+import { msg, getLang } from "../utils/messages.js";
+import { RepeatDetector, getRepeatThreshold } from "../repeat-detector.js";
+import { emitProgress, makeEvent } from "../utils/events.js";
+import {
+  prepareGitAutomation,
+  finalizeGitAutomation
+} from "../git/automation.js";
+import { scanDiff } from "../guards/output-guard.js";
+import { scanPerfDiff } from "../guards/perf-guard.js";
+import { classifyIntent } from "../guards/intent-guard.js";
+import { CoderRole } from "../roles/coder-role.js";
+import { invokeSolomon } from "./solomon-escalation.js";
+import { PipelineContext } from "./pipeline-context.js";
+import { runTriageStage, runResearcherStage, runArchitectStage, runPlannerStage, runDiscoverStage, runHuReviewerStage } from "./pre-loop-stages.js";
+import { runDomainCuratorStage } from "./stages/domain-curator-stage.js";
+import { persistInlineDomain } from "../domains/domain-loader.js";
+import { runCoderStage, runRefactorerStage, runTddCheckStage, runSonarStage, runSonarCloudStage, runReviewerStage } from "./iteration-stages.js";
+import { runTesterStage, runSecurityStage, runImpeccableStage, runFinalAuditStage } from "./post-loop-stages.js";
+import { needsSubPipeline, runHuSubPipeline } from "./hu-sub-pipeline.js";
+import { waitForCooldown, MAX_STANDBY_RETRIES } from "./standby.js";
+import { detectTestFramework } from "../utils/project-detect.js";
+import { runPreflightChecks } from "./preflight-checks.js";
+import { detectRtk } from "../utils/rtk-detect.js";
+import { createRtkRunner, RtkSavingsTracker } from "../utils/rtk-wrapper.js";
+import { setRunner as setDiffRunner, setProjectDir as setDiffProjectDir } from "../review/diff-generator.js";
+import { setRunner as setGitRunner } from "../utils/git.js";
+import { detectNeededSkills, autoInstallSkills, cleanupAutoInstalledSkills } from "../skills/skill-detector.js";
+import { isOpenSkillsAvailable } from "../skills/openskills-client.js";
+import { refineSkillsSemantically, resolveSkillsMode } from "../skills/semantic-detector.js";
+
+// Extracted modules
+import {
+  loadProductContext,
+  resolvePipelineFlags, handleDryRun, createBudgetManager,
+  initializeSession, applyTriageOverrides, applyAutoSimplify,
+  applyFlagOverrides, resolvePipelinePolicies, autoInit,
+  updateGitignoreForStack
+} from "./config-init.js";
+import {
+  tryCiComment, handleCiEarlyPrOrPush, handleCiReviewDispatch,
+  formatBlockingIssues
+} from "./ci-integration.js";
+import {
+  handleCheckpoint, checkSessionTimeout, checkBudgetExceeded,
+  takeCheckpointSnapshot
+} from "./flow-control.js";
+import {
+  createJournalDir, writePreLoopJournal, writeIterationsJournal,
+  writeDecisionsJournal, writeTreeJournal, writeSummaryJournal,
+  formatIteration, formatDecision, buildPlanSummary
+} from "./session-journal.js";
+
+// Public re-exports (loadProductContext, shouldAutoContinueCheckpoint,
+// parseCheckpointAnswer) live in src/orchestrator.js (the barrel).
+
+
+// PG card "In Progress" logic moved to src/planning-game/pipeline-adapter.js → initPgAdapter()
+
+async function runPlanningPhases({ config, logger, emitter, eventBase, session, stageResults, pipelineFlags, coderRole, trackBudget, task, askQuestion, brainCtx }) {
+  let researchContext = null;
+  let plannedTask = task;
+
+  // Brain: track compression across pre-loop roles
+  const brainCompress = brainCtx?.enabled
+    ? (await import("./brain-coordinator.js")).processRoleOutput
+    : null;
+
+  if (pipelineFlags.researcherEnabled) {
+    const researcherResult = await runResearcherStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget });
+    researchContext = researcherResult.researchContext;
+    stageResults.researcher = researcherResult.stageResult;
+    if (brainCompress) brainCompress(brainCtx, { roleName: "researcher", output: researcherResult.stageResult, iteration: 0 });
+  }
+
+  // --- Architect (between researcher and planner) ---
+  let architectContext = null;
+  if (pipelineFlags.architectEnabled) {
+    const architectResult = await runArchitectStage({
+      config, logger, emitter, eventBase, session, coderRole, trackBudget,
+      researchContext,
+      discoverResult: stageResults.discover || null,
+      triageLevel: stageResults.triage?.level || null,
+      askQuestion
+    });
+    architectContext = architectResult.architectContext;
+    stageResults.architect = architectResult.stageResult;
+    if (brainCompress) brainCompress(brainCtx, { roleName: "architect", output: architectResult.stageResult, iteration: 0 });
+  }
+
+  const triageDecomposition = stageResults.triage?.shouldDecompose ? stageResults.triage.subtasks : null;
+  if (pipelineFlags.plannerEnabled) {
+    const plannerRole = resolveRole(config, "planner");
+    const plannerResult = await runPlannerStage({ config, logger, emitter, eventBase, session, plannerRole, researchContext, architectContext, triageDecomposition, trackBudget });
+    plannedTask = plannerResult.plannedTask;
+    stageResults.planner = plannerResult.stageResult;
+    if (brainCompress) brainCompress(brainCtx, { roleName: "planner", output: plannerResult.stageResult, iteration: 0 });
+
+    await tryCiComment({
+      config, session, logger,
+      agent: "Planner",
+      body: `Plan: ${plannerResult.stageResult?.summary || plannedTask}`
+    });
+  }
+
+  return { plannedTask };
+}
+
+
+
+async function handleStandbyResult({ stageResult, session, emitter, eventBase, i, stage, logger, config, askQuestion }) {
+  if (stageResult?.action !== "standby") {
+    return { handled: false };
+  }
+
+  const isOutage = stageResult.standbyInfo.isProviderOutage;
+  const agent = stageResult.standbyInfo.agent;
+  const cooldownUntil = stageResult.standbyInfo.cooldownUntil;
+  const cooldownMs = stageResult.standbyInfo.cooldownMs;
+
+  // Rate limit = out of normal flow → Solomon decides immediately
+  const solomonResult = await invokeSolomon({
+    config, logger, emitter, eventBase, stage: `${stage}_rate_limit`, askQuestion, session,
+    iteration: i,
+    conflict: {
+      stage: `${stage}_rate_limit`,
+      task: session.task,
+      iterationCount: i,
+      maxIterations: config?.max_iterations || 5,
+      cooldownUntil,
+      cooldownMs,
+      history: [{
+        agent: stage,
+        feedback: `Agent "${agent}" rate-limited. ${isOutage ? "Provider outage (5xx)." : "API rate limit (429)."} ${cooldownUntil ? `Cooldown until ${cooldownUntil}.` : ""}`
+      }]
+    }
+  });
+
+  if (solomonResult.action === "approve") {
+    // Solomon says skip is safe — only allowed after exhausting alternatives
+    logger.info(`Solomon: skip ${stage} after evaluating risk (agent "${agent}" rate-limited)`);
+    emitProgress(emitter, makeEvent(`${stage}:rate_limit`, { ...eventBase, stage }, {
+      status: "ok",
+      message: `Solomon: skip ${stage} (low risk, agent "${agent}" unavailable)`,
+      detail: { agent, solomonAction: "approve" }
+    }));
+    return { handled: true, action: "skip" };
+  }
+
+  if (solomonResult.action === "continue") {
+    // Solomon says: wait for cooldown, or retry with alternative agent
+    const altAgent = solomonResult.alternativeAgent;
+    const waitTarget = solomonResult.waitUntil;
+
+    if (waitTarget) {
+      const waitMs = Math.max(0, new Date(waitTarget).getTime() - Date.now());
+      if (waitMs > 0 && waitMs < 10 * 60 * 1000) {
+        logger.info(`Solomon: wait ${Math.round(waitMs / 1000)}s for cooldown, then retry ${stage}`);
+        await new Promise(r => setTimeout(r, waitMs));
+      }
+    }
+
+    if (altAgent) {
+      logger.info(`Solomon: retry ${stage} with alternative agent "${altAgent}"`);
+      session._alternative_agent = { stage, provider: altAgent };
+    }
+
+    if (solomonResult.humanGuidance) {
+      session.last_reviewer_feedback = `Solomon guidance: ${solomonResult.humanGuidance}`;
+    }
+    await saveSession(session);
+    return { handled: true, action: "retry_reviewer_only" };
+  }
+
+  if (solomonResult.action === "pause") {
+    return {
+      handled: true,
+      action: "return",
+      result: { paused: true, sessionId: session.id, question: solomonResult.question, context: "rate_limit" }
+    };
+  }
+
+  // Solomon couldn't resolve — pause
+  await pauseSession(session, {
+    question: `Agent "${agent}" rate-limited. Solomon could not resolve.`,
+    context: { iteration: i, stage, reason: "rate_limit" }
+  });
+  return {
+    handled: true,
+    action: "return",
+    result: { paused: true, sessionId: session.id, question: `Agent "${agent}" rate-limited`, context: "rate_limit" }
+  };
+}
+
+
+function emitSolomonAlerts(alerts, emitter, eventBase, logger) {
+  for (const alert of alerts) {
+    emitProgress(emitter, makeEvent("brain:rules-alert", { ...eventBase, stage: "brain" }, {
+      status: alert.severity === "critical" ? "fail" : "warn",
+      message: alert.message,
+      detail: alert.detail
+    }));
+    logger.warn(`Rules alert [${alert.rule}]: ${alert.message}`);
+  }
+}
+
+async function handleSolomonCheck({ config, session, emitter, eventBase, logger, task, i, askQuestion, ciEnabled, blockingIssues, brainCtx }) {
+  if (config.pipeline?.solomon?.enabled === false) return { action: "continue" };
+
+  try {
+    const { evaluateRules, buildRulesContext } = await import("./solomon-rules.js");
+    const rulesContext = await buildRulesContext({ session, task, iteration: i, blockingIssues });
+    const rulesResult = evaluateRules(rulesContext, config.solomon?.rules);
+
+    if (rulesResult.alerts.length > 0) {
+      emitSolomonAlerts(rulesResult.alerts, emitter, eventBase, logger);
+      // Brain gateway: when Brain is the orchestrator, rule alerts are telemetry.
+      // On critical alerts Brain consults Solomon (AI judge). Only if Solomon can't
+      // resolve does Brain escalate to human. Solomon-rules never prompts directly.
+      if (!brainCtx?.enabled) {
+        const pauseResult = await checkSolomonCriticalAlerts({ rulesResult, askQuestion, session, i });
+        if (pauseResult) return pauseResult;
+      } else if (rulesResult.hasCritical) {
+        const criticalAlerts = rulesResult.alerts.filter(a => a.severity === "critical");
+        brainCtx.ruleAlerts = brainCtx.ruleAlerts || [];
+        brainCtx.ruleAlerts.push(...criticalAlerts);
+        logger.info(`Brain: ${criticalAlerts.length} critical rule alert(s) — consulting Solomon`);
+
+        const { invokeSolomon } = await import("./solomon-escalation.js");
+        const alertSummary = criticalAlerts.map(a => a.message).join("; ");
+        const solomonOpinion = await invokeSolomon({
+          config, logger, emitter, eventBase, stage: "brain-dilemma", askQuestion, session, iteration: i,
+          conflict: {
+            stage: "brain-dilemma",
+            task,
+            iterationCount: i,
+            maxIterations: config.max_iterations,
+            dilemma: `Brain detected critical rule alerts: ${alertSummary}. Should we continue iterating, pause for human, or stop?`,
+            ruleAlerts: criticalAlerts,
+            blockingIssues: blockingIssues || [],
+            history: (session.checkpoints || []).filter(cp => cp.stage === "reviewer").slice(-5).map(cp => ({ iteration: cp.iteration, feedback: cp.note || "" }))
+          }
+        });
+
+        if (solomonOpinion.action === "pause") {
+          logger.info("Brain: Solomon advised pause — escalating to human");
+          return { action: "return", result: { paused: true, sessionId: session.id, question: solomonOpinion.question || `Brain+Solomon paused: ${alertSummary}`, context: "brain_solomon_dilemma" } };
+        }
+        if (solomonOpinion.action === "approve") {
+          logger.info("Brain: Solomon advised proceeding — treating as approved");
+          return { action: "continue", approved: true };
+        }
+        // action === "continue" | "subtask" | fallback → Brain continues the loop
+        logger.info(`Brain: Solomon said '${solomonOpinion.action}' — continuing loop`);
+      }
+    }
+
+    if (ciEnabled && session.ci_pr_number) {
+      const alerts = rulesResult.alerts || [];
+      const alertMsg = alerts.length > 0
+        ? alerts.map(a => `- [${a.severity}] ${a.message}`).join("\n")
+        : "No anomalies detected";
+      await tryCiComment({
+        config, session, logger,
+        agent: "Solomon",
+        body: `Supervisor check iteration ${i}: ${alertMsg}`
+      });
+    }
+  } catch (err) {
+    logger.warn(`Solomon rules evaluation failed: ${err.message}`);
+  }
+
+  return { action: "continue" };
+}
+
+async function checkSolomonCriticalAlerts({ rulesResult, askQuestion, session, i }) {
+  if (!rulesResult.hasCritical || !askQuestion) return null;
+
+  const alertSummary = rulesResult.alerts
+    .filter(a => a.severity === "critical")
+    .map(a => a.message)
+    .join("\n");
+  const question = [
+    "Solomon detected critical issues:",
+    alertSummary,
+    "",
+    "What should we do?",
+    "1 = Continue anyway",
+    "2 = Pause the session",
+    "3 = Stop the session",
+    "",
+    "Type 1, 2, or 3:"
+  ].join("\n");
+  const answer = await askQuestion(question, { iteration: i, stage: "solomon" });
+  const trimmed = (answer || "").trim().toLowerCase();
+  const shouldPause = !answer
+    || trimmed === "2" || trimmed === "3"
+    || trimmed.startsWith("pause") || trimmed.startsWith("stop");
+  if (shouldPause) {
+    await pauseSession(session, {
+      question: `Solomon supervisor paused: ${alertSummary}`,
+      context: { iteration: i, stage: "solomon", alerts: rulesResult.alerts }
+    });
+    return { action: "pause", result: { paused: true, sessionId: session.id, reason: "solomon_alert" } };
+  }
+  return null;
+}
+
+
+async function handlePostLoopStages({ config, session, emitter, eventBase, coderRole, trackBudget, i, task, stageResults, ciEnabled, testerEnabled, securityEnabled, askQuestion, logger, brainCtx }) {
+  const postLoopDiff = await generateDiff({ baseRef: session.session_start_sha });
+
+  if (testerEnabled) {
+    const testerResult = await runTesterStage({
+      config, logger, emitter, eventBase, session, coderRole, trackBudget,
+      iteration: i, task, diff: postLoopDiff, askQuestion
+    });
+    if (testerResult.action === "pause") return { action: "return", result: testerResult.result };
+    if (testerResult.action === "continue") {
+      const summary = testerResult.stageResult?.summary || "Tester found issues";
+      session.last_reviewer_feedback = `Tester FAILED — fix these issues:\n${summary}`;
+      await saveSession(session);
+      if (testerResult.stageResult) stageResults.tester = testerResult.stageResult;
+      // Brain: push tester failure into feedback queue + compress for next coder iteration
+      if (brainCtx?.enabled) {
+        const { processRoleOutput } = await import("./brain-coordinator.js");
+        processRoleOutput(brainCtx, { roleName: "tester", output: testerResult.stageResult, iteration: i });
+      }
+      return { action: "continue" };
+    }
+    if (testerResult.stageResult) {
+      stageResults.tester = testerResult.stageResult;
+      await tryCiComment({ config, session, logger, agent: "Tester", body: `Tests: ${testerResult.stageResult.summary || "completed"}` });
+    }
+  }
+
+  if (securityEnabled) {
+    const securityResult = await runSecurityStage({
+      config, logger, emitter, eventBase, session, coderRole, trackBudget,
+      iteration: i, task, diff: postLoopDiff, askQuestion, brainCtx
+    });
+    if (securityResult.action === "pause") return { action: "return", result: securityResult.result };
+    if (securityResult.action === "continue") {
+      const summary = securityResult.stageResult?.summary || "Security found issues";
+      session.last_reviewer_feedback = `Security FAILED — fix these issues:\n${summary}`;
+      await saveSession(session);
+      if (securityResult.stageResult) stageResults.security = securityResult.stageResult;
+      // Brain: push security findings into feedback queue + compress for next coder iteration
+      if (brainCtx?.enabled) {
+        const { processRoleOutput } = await import("./brain-coordinator.js");
+        processRoleOutput(brainCtx, { roleName: "security", output: securityResult.stageResult, iteration: i });
+      }
+      return { action: "continue" };
+    }
+    if (securityResult.stageResult) {
+      stageResults.security = securityResult.stageResult;
+      await tryCiComment({ config, session, logger, agent: "Security", body: `Security scan: ${securityResult.stageResult.summary || "completed"}` });
+    }
+  }
+
+  // Final audit — last quality gate before declaring success
+  const auditResult = await runFinalAuditStage({
+    config, logger, emitter, eventBase, session, coderRole, trackBudget,
+    iteration: i, task, diff: postLoopDiff
+  });
+  if (auditResult.stageResult) {
+    stageResults.audit = auditResult.stageResult;
+    await tryCiComment({ config, session, logger, agent: "Audit", body: `Final audit: ${auditResult.stageResult.summary || "completed"}` });
+  }
+  if (auditResult.action === "retry") {
+    // Audit found actionable issues — loop back to coder
+    session.last_reviewer_feedback = auditResult.feedback;
+    await saveSession(session);
+    return { action: "continue" };
+  }
+
+  return { action: "proceed" };
+}
+
+async function finalizeApprovedSession({ config, gitCtx, task, logger, session, stageResults, emitter, eventBase, budgetSummary, pgCard, pgProject, review, i, rtkTracker }) {
+  const gitResult = await finalizeGitAutomation({ config, gitCtx, task, logger, session, stageResults });
+
+  // Accumulate final commits for PG card lifecycle tracking
+  if (gitResult?.commits?.length) {
+    const { accumulateCommit } = await import("../planning-game/pipeline-adapter.js");
+    for (const c of gitResult.commits) accumulateCommit(session, c);
+  }
+
+  if (stageResults.planner?.ok) {
+    stageResults.planner.completedSteps = [...(stageResults.planner.steps ?? [])];
+  }
+  session.budget = budgetSummary();
+  await markSessionStatus(session, "approved");
+
+  const { markPgCardToValidate } = await import("../planning-game/pipeline-adapter.js");
+  await markPgCardToValidate({ pgCard, pgProject, config, session, gitResult, logger });
+
+  const deferredIssues = session.deferred_issues || [];
+  const rtkSavings = rtkTracker?.hasData() ? rtkTracker.summary() : undefined;
+  if (rtkSavings) session.rtk_savings = rtkSavings;
+  await saveSession(session);
+
+  // --- Journal: write final files ---
+  const journalDir = session._journalDir;
+  if (journalDir) {
+    try {
+      await writeIterationsJournal(journalDir, session._journalIterations || []);
+      // decisions.md: writes only when Solomon was invoked (AC of KJC-TSK-0287).
+      // Pulls structured data from session.solomonRulings + session.brainDecisions.
+      const { writeDecisionsJournal: writeDecisionsJournalV2 } =
+        await import("../session/journal/decisions-writer.js");
+      const decisionsResult = await writeDecisionsJournalV2(session, { journalDir, logger });
+      // KJC-TSK-0288: use the new tree-writer with directory grouping
+      const { writeTreeJournal: writeTreeJournalV2 } =
+        await import("../session/journal/tree-writer.js");
+      const treeResult = await writeTreeJournalV2(journalDir, session.session_start_sha, { logger });
+
+      const journalFiles = [...(session._journalFiles || [])];
+      if (session._journalIterations?.length) journalFiles.push("iterations.md");
+      if (decisionsResult.written) journalFiles.push("decisions.md");
+      if (treeResult.written) journalFiles.push("tree.txt");
+
+      // KJC-TSK-0289: use the richer summary writer (stages table, budget
+      // breakdown, brain/solomon counts, typed links to other journal files).
+      const { writeSummaryJournal: writeSummaryJournalV2 } =
+        await import("../session/journal/summary-writer.js");
+      const startedAt = session._startedAt ? new Date(session._startedAt).toISOString() : undefined;
+      await writeSummaryJournalV2(journalDir, {
+        task: session.task,
+        result: "APPROVED",
+        sessionId: session.id,
+        iterations: i,
+        durationMs: Date.now() - (session._startedAt || Date.now()),
+        budget: budgetSummary(),
+        stages: stageResults,
+        commits: gitResult?.commits || [],
+        files: journalFiles,
+        startedAt,
+        finishedAt: new Date().toISOString(),
+        brainDecisions: Array.isArray(session.brainDecisions) ? session.brainDecisions.length : 0,
+        solomonInvocations: Array.isArray(session.solomonRulings) ? session.solomonRulings.length : 0,
+      }, { logger });
+      logger.info(`Session journal written to ${journalDir}`);
+    } catch (err) {
+      logger.warn(`Journal write failed (non-blocking): ${err.message}`);
+    }
+  }
+
+  const endDetail = { approved: true, iterations: i, stages: stageResults, git: gitResult, budget: budgetSummary(), deferredIssues };
+  if (rtkSavings) endDetail.rtk_savings = rtkSavings;
+  emitProgress(
+    emitter,
+    makeEvent("session:end", { ...eventBase, stage: "done" }, {
+      message: deferredIssues.length > 0
+        ? `Session approved (${deferredIssues.length} deferred issue(s) tracked as tech debt)`
+        : "Session approved",
+      detail: endDetail
+    })
+  );
+  const result = { approved: true, sessionId: session.id, review, git: gitResult, deferredIssues };
+  if (rtkSavings) result.rtk_savings = rtkSavings;
+  return result;
+}
+
+// PG card "To Validate" logic moved to src/planning-game/pipeline-adapter.js → markPgCardToValidate()
+
+async function handleReviewerRetryAndSolomon({ config, session, emitter, eventBase, logger, review, task, i, askQuestion }) {
+  session.last_reviewer_feedback = review.blocking_issues
+    .map((x) => {
+      const parts = [`[${x.severity || "high"}] ${x.id || "ISSUE"}: ${x.description || "Missing description"}`];
+      if (x.file) parts.push(`  File: ${x.file}${x.line ? `:${x.line}` : ""}`);
+      if (x.suggested_fix) parts.push(`  Fix: ${x.suggested_fix}`);
+      return parts.join("\n");
+    })
+    .join("\n\n");
+  session.reviewer_retry_count = (session.reviewer_retry_count || 0) + 1;
+  await saveSession(session);
+
+  const maxReviewerRetries = config.session.max_reviewer_retries ?? config.session.fail_fast_repeats;
+  if (session.reviewer_retry_count < maxReviewerRetries) {
+    return { action: "continue" };
+  }
+
+  emitProgress(
+    emitter,
+    makeEvent("solomon:escalate", { ...eventBase, stage: "reviewer" }, {
+      message: `Reviewer sub-loop limit reached (${session.reviewer_retry_count}/${maxReviewerRetries})`,
+      detail: { subloop: "reviewer", retryCount: session.reviewer_retry_count, limit: maxReviewerRetries }
+    })
+  );
+
+  const solomonResult = await invokeSolomon({
+    config, logger, emitter, eventBase, stage: "reviewer", askQuestion, session, iteration: i,
+    conflict: {
+      stage: "reviewer",
+      task,
+      iterationCount: session.reviewer_retry_count,
+      maxIterations: maxReviewerRetries,
+      history: [{ agent: "reviewer", feedback: session.last_reviewer_feedback }]
+    }
+  });
+
+  if (solomonResult.action === "pause") {
+    return { action: "return", result: { paused: true, sessionId: session.id, question: solomonResult.question, context: "reviewer_fail_fast" } };
+  }
+  if (solomonResult.action === "continue") {
+    if (solomonResult.humanGuidance) {
+      session.last_reviewer_feedback += `\nUser guidance: ${solomonResult.humanGuidance}`;
+    }
+    session.reviewer_retry_count = 0;
+    await saveSession(session);
+    return { action: "continue" };
+  }
+  if (solomonResult.action === "subtask") {
+    return { action: "return", result: { paused: true, sessionId: session.id, subtask: solomonResult.subtask, context: "reviewer_subtask" } };
+  }
+
+  return { action: "continue" };
+}
+
+
+async function runPreLoopStages({ config, logger, emitter, eventBase, session, flags, pipelineFlags, coderRole, trackBudget, task, askQuestion, pgTaskId, pgProject, stageResults, brainCtx }) {
+  // --- HU Reviewer (first stage, before everything else, opt-in) ---
+  const huFile = flags.huFile || null;
+  if (flags.enableHuReviewer !== undefined) pipelineFlags.huReviewerEnabled = Boolean(flags.enableHuReviewer);
+  if (pipelineFlags.huReviewerEnabled && huFile) {
+    const huResult = await runHuReviewerStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget, huFile, askQuestion, pgStories: null });
+    stageResults.huReviewer = huResult.stageResult;
+  }
+
+  // --- Intent classifier (deterministic pre-triage, opt-in) ---
+  if (config.guards?.intent?.enabled) {
+    const intentResult = classifyIntent(task, config);
+    stageResults.intent = intentResult;
+    if (intentResult.classified) {
+      emitProgress(emitter, makeEvent("intent:classified", { ...eventBase, stage: "intent" }, {
+        message: `Intent classified: ${intentResult.taskType} (${intentResult.level}) — ${intentResult.message}`,
+        detail: intentResult
+      }));
+    }
+  }
+
+  // --- Discover (pre-triage, opt-in) ---
+  if (flags.enableDiscover !== undefined) pipelineFlags.discoverEnabled = Boolean(flags.enableDiscover);
+  if (pipelineFlags.discoverEnabled) {
+    const discoverResult = await runDiscoverStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget });
+    stageResults.discover = discoverResult.stageResult;
+  }
+
+  // --- Triage (always on) ---
+  const triageResult = await runTriageStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget });
+  applyTriageOverrides(pipelineFlags, triageResult.roleOverrides);
+  stageResults.triage = triageResult.stageResult;
+
+  // --- Brain decisor (opt-in, intent-driven routing) ---
+  //
+  // When enabled, translates triage output + task text + CLI overrides into a
+  // structured Decision and applies it as pipelineFlags overrides. The rest
+  // of the orchestrator continues to branch on pipelineFlags as before — the
+  // Brain is a router that adjusts WHICH roles run, not HOW they run.
+  //
+  // Flag precedence:
+  //   1. flags.brain === "off"    → forced off
+  //   2. flags.brain === "on"     → forced on
+  //   3. config.brain.decisor.enabled (if set)
+  //   4. globalThis.__KJ_DEFAULT_BRAIN_DECISOR (test harness default = false)
+  //   5. true in production
+  const brainDefault = globalThis.__KJ_DEFAULT_BRAIN_DECISOR ?? true;
+  let brainDecisorEnabled = config?.brain?.decisor?.enabled ?? brainDefault;
+  if (flags?.brain === "off") brainDecisorEnabled = false;
+  if (flags?.brain === "on") brainDecisorEnabled = true;
+  if (brainDecisorEnabled) {
+    try {
+      const { buildDecision, applyDecisionToFlags } = await import("../brain/decisor.js");
+      const { createTracker, recordDecision, checkLimits } = await import("../brain/decision-tracker.js");
+      const tracker = createTracker(session, { max: config?.brain?.decisor?.maxDecisions });
+      const limitStatus = checkLimits(tracker);
+      if (limitStatus.status === "ok") {
+        let decision = buildDecision({
+          triage: triageResult.stageResult,
+          task,
+          config,
+          overrides: {
+            // Commander variadic --force-role a b → flags.forceRole = ["a","b"].
+            // Back-compat: accept both singular (commander) and plural (legacy / API).
+            forceRoles: flags?.forceRole || flags?.forceRoles || [],
+            skipRoles: flags?.skipRole || flags?.skipRoles || [],
+          },
+        });
+        // If the decision is low-confidence or ambiguous, consult Solomon
+        // for a refined routing suggestion. Solomon's ruling is advisory —
+        // if it fails or returns nothing useful, the baseline decision stands.
+        if (decision.consultSolomon) {
+          try {
+            const { consultSolomonForRouting } = await import("../brain/solomon-consult.js");
+            decision = await consultSolomonForRouting({
+              decision, triage: triageResult.stageResult, task, config, logger,
+              emitter, eventBase, session,
+            });
+          } catch (err) {
+            logger.warn(`Brain → Solomon consult failed (non-blocking): ${err.message}`);
+          }
+        }
+        const newFlags = applyDecisionToFlags(decision, pipelineFlags);
+        Object.assign(pipelineFlags, newFlags);
+        recordDecision(tracker, decision, {
+          taskType: triageResult.stageResult?.taskType,
+          level: triageResult.stageResult?.level,
+        });
+        emitProgress(emitter, makeEvent("brain:decision", { ...eventBase, stage: "brain" }, {
+          status: "ok",
+          message: `Brain routing: ${decision.rolesOn.length} role(s) active — ${decision.rationale}`,
+          detail: {
+            rolesOn: decision.rolesOn,
+            rolesOff: decision.rolesOff,
+            confidence: decision.confidence,
+            consultSolomon: decision.consultSolomon,
+            appliedOverrides: decision.appliedOverrides,
+          },
+        }));
+      } else {
+        logger.warn(`Brain decisor skipped: ${limitStatus.detail}`);
+        emitProgress(emitter, makeEvent("brain:decision", { ...eventBase, stage: "brain" }, {
+          status: "warn",
+          message: `Brain decisor skipped: ${limitStatus.status}`,
+          detail: limitStatus,
+        }));
+      }
+    } catch (err) {
+      logger.warn(`Brain decisor failed (non-blocking): ${err.message}`);
+    }
+  }
+
+  // --- Persist inline domain if provided via --domain flag ---
+  if (flags?.domain) {
+    try {
+      await persistInlineDomain(flags.domain, config.projectDir || process.cwd());
+    } catch (err) {
+      logger.warn(`Failed to persist inline domain: ${err.message}`);
+    }
+  }
+
+  // --- Domain Curator (after triage + skill auto-install, before planning phases) ---
+  const domainHints = triageResult.stageResult?.domainHints || [];
+  if (domainHints.length > 0 || config.projectDir) {
+    try {
+      const { domainContext, stageResult: dcStageResult } = await runDomainCuratorStage({
+        config, logger, emitter, eventBase, session, trackBudget,
+        domainHints, askQuestion
+      });
+      stageResults.domainCurator = dcStageResult;
+      if (domainContext) {
+        config = { ...config, domainContext };
+      }
+    } catch (err) {
+      logger.warn(`Domain Curator failed (non-blocking): ${err.message}`);
+    }
+  }
+
+  // --- HU Reviewer auto-activation from triage (post-triage, no huFile needed) ---
+  const triageRoles = new Set(triageResult.stageResult?.roles || []);
+  if (triageRoles.has("hu-reviewer") && !stageResults.huReviewer) {
+    pipelineFlags.huReviewerEnabled = true;
+    // Feed PG card structured data to hu-reviewer when available
+    let pgStories = null;
+    if (pgTaskId && pgProject && session.pg_card) {
+      const { buildHuStoriesFromPgCard } = await import("../planning-game/pipeline-adapter.js");
+      pgStories = buildHuStoriesFromPgCard(session.pg_card);
+    }
+    const huResult = await runHuReviewerStage({ config, logger, emitter, eventBase, session, coderRole, trackBudget, huFile: null, askQuestion, pgStories });
+    stageResults.huReviewer = huResult.stageResult;
+  }
+
+  // --- Auto-simplify pipeline for simple tasks (before explicit flag overrides) ---
+  const simplified = applyAutoSimplify({
+    pipelineFlags,
+    triageLevel: triageResult.stageResult?.level || null,
+    config, flags, logger, emitter, eventBase
+  });
+  if (simplified) stageResults.triage.autoSimplified = true;
+
+  const { handlePgDecomposition } = await import("../planning-game/pipeline-adapter.js");
+  await handlePgDecomposition({ triageResult, pgTaskId, pgProject, config, askQuestion, emitter, eventBase, session, stageResults, logger });
+
+  applyFlagOverrides(pipelineFlags, flags);
+
+  // --- Auto-detect TDD applicability when methodology not explicitly set ---
+  if (!flags.methodology) {
+    const projectDir = config.projectDir || process.cwd();
+    const detection = await detectTestFramework(projectDir);
+    if (!detection.hasTests) {
+      config = { ...config, development: { ...config.development, methodology: "standard", require_test_changes: false } };
+      logger.info("No test framework detected — using standard methodology");
+    } else {
+      config = { ...config, development: { ...config.development, methodology: "tdd", require_test_changes: true } };
+      logger.info(`Test framework detected (${detection.framework}) — using TDD methodology`);
+    }
+    emitProgress(emitter, makeEvent("tdd:auto-detect", { ...eventBase, stage: "pre-loop" }, {
+      message: detection.hasTests
+        ? `TDD auto-detected: ${detection.framework}`
+        : "TDD skipped: no test framework found",
+      detail: detection
+    }));
+  }
+
+  let updatedConfig = resolvePipelinePolicies({ flags, config, stageResults, emitter, eventBase, session, pipelineFlags });
+
+  // --- Preflight environment checks ---
+  const preflightResult = await runPreflightChecks({
+    config: updatedConfig, logger, emitter, eventBase,
+    resolvedPolicies: session.resolved_policies,
+    securityEnabled: pipelineFlags.securityEnabled
+  });
+  session.preflight = preflightResult;
+  await saveSession(session);
+
+  // Hard fail if blocking checks failed (SonarQube enabled but not available)
+  if (!preflightResult.ok) {
+    const errorLines = (preflightResult.errors || [])
+      .map(e => `  - ${e.message}\n    Fix: ${e.fix}`)
+      .join("\n");
+    throw new Error(
+      `Preflight FAILED — environment changed during session. Fix the issues and retry:\n${errorLines}`
+    );
+  }
+
+  if (preflightResult.configOverrides.securityDisabled) {
+    pipelineFlags.securityEnabled = false;
+  }
+
+  // --- Plan injection: skip researcher/architect/planner if a persisted plan is loaded ---
+  if (flags.plan) {
+    try {
+      const { loadPlan, savePlan: savePlanToDisk } = await import("../plan/plan-store.js");
+      const { isPlanV2 } = await import("../plan/plan-schema.js");
+      const { planToHuBatch, syncResultsToPlan } = await import("../plan/plan-executor.js");
+      const projectDir = updatedConfig.projectDir || process.cwd();
+      const loadedPlan = await loadPlan(projectDir, flags.plan);
+      if (loadedPlan) {
+        logger.info(`Loaded persisted plan: ${flags.plan} (v${loadedPlan.version || 1})`);
+        emitProgress(emitter, makeEvent("plan:loaded", { ...eventBase, stage: "plan" }, {
+          message: `Plan loaded: ${flags.plan}`,
+          detail: { planId: flags.plan, task: loadedPlan.task, version: loadedPlan.version, huCount: loadedPlan.hus?.length || 0 }
+        }));
+        stageResults.researcher = { ok: true, summary: "Loaded from persisted plan", fromPlan: flags.plan };
+        stageResults.architect = { ok: true, summary: "Loaded from persisted plan", fromPlan: flags.plan };
+        stageResults.planner = { ok: true, summary: "Loaded from persisted plan", fromPlan: flags.plan };
+
+        // v2 plan with HUs → inject as huReviewer so sub-pipeline picks them up
+        if (isPlanV2(loadedPlan) && loadedPlan.hus?.length > 0) {
+          const huBatch = planToHuBatch(loadedPlan);
+          if (huBatch.ok) {
+            stageResults.huReviewer = huBatch;
+            // Store plan reference so we can sync results back after execution
+            session._planRef = { planId: loadedPlan.planId, projectDir };
+            session._syncResultsToPlan = async (subResult) => {
+              syncResultsToPlan(loadedPlan, subResult);
+              await savePlanToDisk(projectDir, loadedPlan);
+            };
+            logger.info(`Plan ${flags.plan}: ${huBatch.certified} HUs ready for execution`);
+            emitProgress(emitter, makeEvent("plan:hus-loaded", { ...eventBase, stage: "plan" }, {
+              message: `${huBatch.certified} HUs loaded from plan`,
+              detail: { planId: flags.plan, total: huBatch.total, certified: huBatch.certified }
+            }));
+            // Mark plan as running
+            loadedPlan.status = "running";
+            await savePlanToDisk(projectDir, loadedPlan);
+          }
+        }
+
+        // Inject contexts
+        const ctx = loadedPlan.context || {};
+        session.research_context = ctx.researchContext || loadedPlan.researchContext || null;
+        session.architect_context = ctx.architectContext || loadedPlan.architectContext || null;
+        session.loaded_plan = loadedPlan.approach || loadedPlan.plan || null;
+        await saveSession(session);
+
+        // Build planned task from plan (for non-HU or fallback path)
+        let plannedTask = task;
+        const plan = loadedPlan.plan || (loadedPlan.approach ? { approach: loadedPlan.approach } : null);
+        if (plan && typeof plan === "object" && plan.steps) {
+          const stepList = plan.steps.map((s, idx) => `${idx + 1}. ${s.description || s}`).join("\n");
+          plannedTask = `${task}\n\n## Implementation Plan\n${plan.approach || ""}\n\n## Steps\n${stepList}`;
+        } else if (typeof plan === "string") {
+          plannedTask = `${task}\n\n## Implementation Plan\n${plan}`;
+        } else if (loadedPlan.approach) {
+          plannedTask = `${task}\n\n## Approach\n${loadedPlan.approach}`;
+        }
+        return { plannedTask, updatedConfig };
+      }
+      logger.warn(`Plan ${flags.plan} not found — falling back to normal pipeline`);
+    } catch (err) {
+      logger.warn(`Plan loading failed: ${err.message} — falling back to normal pipeline`);
+    }
+  }
+
+  // --- Researcher → Planner ---
+  const { plannedTask } = await runPlanningPhases({ config: updatedConfig, logger, emitter, eventBase, session, stageResults, pipelineFlags, coderRole, trackBudget, task, askQuestion, brainCtx });
+
+  // --- Update .gitignore with stack-specific entries based on planner/architect output ---
+  const projectDir = updatedConfig.projectDir || process.cwd();
+  await updateGitignoreForStack(projectDir, { stageResults, task, logger });
+
+  // --- Auto-HU: when triage recommended decomposition and no manual huFile, generate HU batch ---
+  await maybeGenerateAutoHuBatch({
+    flags, stageResults, task, plannedTask, logger, emitter, eventBase, projectDir, session
+  });
+
+  // --- Auto-install skills based on task + planner output + project detection ---
+  // Runs AFTER triage and planner so that the planned task text (which includes
+  // planner output like implementation steps) is available for keyword detection.
+  // This ensures greenfield projects with no package.json still get correct skills.
+  //
+  // Detection runs UNCONDITIONALLY — even when the openskills CLI is missing —
+  // so the session report can recommend which skills would have been used. The
+  // actual install is skipped inside autoInstallSkills when unavailable.
+  const skillProjectDir = updatedConfig.projectDir || process.cwd();
+  try {
+    const skillsMode = resolveSkillsMode(updatedConfig, flags);
+    if (skillsMode === "none") {
+      // User explicitly opted out. Skip detection entirely.
+      return { plannedTask, updatedConfig };
+    }
+    let neededSkills = skillsMode === "semantic"
+      ? []
+      : await detectNeededSkills(plannedTask, skillProjectDir);
+    // Semantic mode augments regex detection with a classifier call.
+    // "auto" = regex always + semantic when budget allows (v1: always when mode=auto and classifier reachable).
+    if (skillsMode === "auto" || skillsMode === "semantic") {
+      const extra = await refineSkillsSemantically({
+        task: plannedTask,
+        alreadyDetected: neededSkills,
+        config: updatedConfig,
+        logger,
+      });
+      if (extra.length > 0) {
+        logger.info(`Semantic skill detection added: ${extra.join(", ")}`);
+        neededSkills = Array.from(new Set([...neededSkills, ...extra]));
+      }
+    }
+    if (neededSkills.length > 0) {
+      const skillResult = await autoInstallSkills(neededSkills, skillProjectDir);
+      if (skillResult.installed.length > 0) {
+        session.autoInstalledSkills = skillResult.installed;
+      }
+      if (skillResult.osAvailable === false && skillResult.wouldHaveUsed.length > 0) {
+        session.skillsRecommended = skillResult.wouldHaveUsed;
+        logger.warn(`OpenSkills CLI not available — would have used: ${skillResult.wouldHaveUsed.join(", ")}. Install openskills globally to auto-inject them next time.`);
+        emitProgress(emitter, makeEvent("skills:unavailable", { ...eventBase, stage: "skills" }, {
+          message: `OpenSkills CLI not available — would have used: ${skillResult.wouldHaveUsed.join(", ")}`,
+          detail: {
+            wouldHaveUsed: skillResult.wouldHaveUsed,
+            hint: "Install openskills globally: npm install -g openskills",
+          },
+        }));
+      }
+      emitProgress(emitter, makeEvent("skills:auto-install", { ...eventBase, stage: "skills" }, {
+        message: skillResult.installed.length > 0
+          ? `Auto-installed ${skillResult.installed.length} skill(s): ${skillResult.installed.join(", ")}`
+          : `Skills detected (${neededSkills.join(", ")}) — all already installed or unavailable`,
+        detail: skillResult,
+      }));
+    }
+  } catch (err) {
+    logger.warn(`Skill auto-install failed (non-blocking): ${err.message}`);
+  }
+
+  return { plannedTask, updatedConfig };
+}
+
+/**
+ * Auto-generate HU batch from triage decomposition when no manual huFile is present.
+ * Runs after researcher/architect/planner so that context is available for better HUs.
+ * Sets stageResults.huReviewer so needsSubPipeline picks it up later.
+ */
+async function maybeGenerateAutoHuBatch({ flags, stageResults, task, plannedTask, logger, emitter, eventBase, projectDir, session }) {
+  // Skip if user passed a manual hu-file
+  if (flags?.huFile) return;
+  // Skip if hu-reviewer already produced a batch (manual enable + PG stories)
+  if (stageResults.huReviewer) return;
+  // Need triage decomposition recommendation
+  const shouldDecompose = stageResults.triage?.shouldDecompose;
+  const subtasks = stageResults.triage?.subtasks;
+  if (!shouldDecompose || !Array.isArray(subtasks) || subtasks.length < 2) return;
+
+  const { generateHuBatch } = await import("../hu/auto-generator.js");
+
+  // Detect if project is new: empty dir or only .git/.karajan/.gitignore
+  let isNewProject = false;
+  try {
+    const fs = await import("node:fs/promises");
+    const entries = await fs.readdir(projectDir);
+    const relevant = entries.filter(e => !e.startsWith(".git") && e !== ".karajan" && e !== ".gitignore");
+    isNewProject = relevant.length === 0;
+  } catch { /* ignore */ }
+
+  // Extract stack hints from planner + architect output
+  const stackHints = [];
+  const combined = `${stageResults.planner?.plan || ""} ${stageResults.architect?.architecture ? JSON.stringify(stageResults.architect.architecture) : ""} ${task}`.toLowerCase();
+  const stackKeywords = ["express", "vite", "vitest", "jest", "next", "astro", "react", "vue", "svelte", "fastapi", "django", "spring", "gin", "nestjs", "monorepo", "workspaces"];
+  for (const kw of stackKeywords) {
+    if (combined.includes(kw)) stackHints.push(kw);
+  }
+
+  const batch = generateHuBatch({
+    originalTask: task,
+    subtasks,
+    stackHints,
+    isNewProject,
+    researcherContext: stageResults.researcher?.summary || null,
+    architectContext: stageResults.architect?.architecture ? JSON.stringify(stageResults.architect.architecture) : null
+  });
+
+  // Persist batch to HU store so hu-sub-pipeline can update story status via saveHuBatch.
+  // Use session.id as batchSessionId.
+  const batchSessionId = `auto-${session.id}`;
+  try {
+    const fs = await import("node:fs/promises");
+    const path = await import("node:path");
+    const { getKarajanHome } = await import("../utils/paths.js");
+    const huDir = path.join(getKarajanHome(), "hu-stories", batchSessionId);
+    await fs.mkdir(huDir, { recursive: true });
+    const persistBatch = {
+      session_id: batchSessionId,
+      project_id: batchSessionId,
+      project_name: batch.projectName,
+      created_at: new Date().toISOString(),
+      updated_at: new Date().toISOString(),
+      stories: batch.stories
+    };
+    await fs.writeFile(path.join(huDir, "batch.json"), JSON.stringify(persistBatch, null, 2));
+  } catch (err) {
+    logger.warn(`Auto-HU: failed to persist batch (${err.message}) — sub-pipeline will use in-memory fallback`);
+  }
+
+  // Wrap in format compatible with needsSubPipeline + runHuSubPipeline
+  stageResults.huReviewer = {
+    ok: true,
+    stories: batch.stories,
+    total: batch.total,
+    certified: batch.certified,
+    batchSessionId,
+    auto_generated: true,
+    source: batch.source
+  };
+
+  logger.info(`Auto-HU: generated ${batch.total} stories (${batch.source.triage_subtasks} subtasks${isNewProject ? ", new project" : ""}${stackHints.length ? `, stack: ${stackHints.join(",")}` : ""})`);
+  emitProgress(emitter, makeEvent("hu:auto-generated", { ...eventBase, stage: "hu-auto-gen" }, {
+    message: `Auto-generated ${batch.total} HU(s) from triage decomposition`,
+    detail: { total: batch.total, subtasks: batch.source.triage_subtasks, isNewProject, stackHints, projectName: batch.projectName }
+  }));
+
+  // Auto-start the board so the user can see the generated HUs.
+  // Always fires when auto-HU runs, independent of hu_board.auto_start flag.
+  // Never during vitest — would race the PID file and leave a detached
+  // node process around after the suite (TSK-0273).
+  if (process.env.VITEST || process.env.NODE_ENV === "test") return;
+  try {
+    const { startBoard, renderBoardBanner } = await import("../commands/board.js");
+    const desiredPort = session.config_snapshot?.hu_board?.port ?? 4000;
+    const boardResult = await startBoard(desiredPort);
+    const url = boardResult.url;
+    const status = boardResult.alreadyRunning ? "already running" : "started";
+    const projectName = batch.projectName || "Auto-generated HUs";
+    console.log(renderBoardBanner({ url, status, projectName }));
+    logger.info(`HU Board ${status} at ${url} (project: ${projectName})`);
+  } catch (err) {
+    logger.warn(`HU Board auto-start failed (non-blocking): ${err.message}`);
+  }
+}
+
+async function runCoderAndRefactorerStages({ coderRoleInstance, coderRole, refactorerRole, pipelineFlags, config, logger, emitter, eventBase, session, plannedTask, trackBudget, i, brainCtx }) {
+  const coderResult = await runCoderStage({ coderRoleInstance, coderRole, config, logger, emitter, eventBase, session, plannedTask, trackBudget, iteration: i, brainCtx });
+  if (coderResult?.action === "pause") return { action: "return", result: coderResult.result };
+  const coderStandby = await handleStandbyResult({ stageResult: coderResult, session, emitter, eventBase, i, stage: "coder", logger, config });
+  if (coderStandby.handled) {
+    return coderStandby.action === "return"
+      ? { action: "return", result: coderStandby.result }
+      : { action: "retry" };
+  }
+
+  if (pipelineFlags.refactorerEnabled) {
+    const refResult = await runRefactorerStage({ refactorerRole, config, logger, emitter, eventBase, session, plannedTask, trackBudget, iteration: i });
+    if (refResult?.action === "pause") return { action: "return", result: refResult.result };
+    const refStandby = await handleStandbyResult({ stageResult: refResult, session, emitter, eventBase, i, stage: "refactorer", logger, config });
+    if (refStandby.handled) {
+      return refStandby.action === "return"
+        ? { action: "return", result: refStandby.result }
+        : { action: "retry" };
+    }
+  }
+
+  return { action: "ok" };
+}
+
+async function runGuardStages({ config, logger, emitter, eventBase, session, iteration }) {
+  const outputEnabled = config.guards?.output?.enabled !== false;
+  const perfEnabled = config.guards?.perf?.enabled !== false;
+
+  if (!outputEnabled && !perfEnabled) return { action: "ok" };
+
+  const baseBranch = config.base_branch || "main";
+  let diff;
+  try {
+    const { generateDiff: genDiff, computeBaseRef: compBase } = await import("../review/diff-generator.js");
+    const baseRef = await compBase({ baseBranch });
+    diff = await genDiff({ baseRef });
+  } catch {
+    logger.warn("Guards: could not generate diff, skipping");
+    return { action: "ok" };
+  }
+
+  if (!diff) return { action: "ok" };
+
+  if (outputEnabled) {
+    const outputResult = scanDiff(diff, config);
+    if (outputResult.violations.length > 0) {
+      const critical = outputResult.violations.filter(v => v.severity === "critical");
+      const warnings = outputResult.violations.filter(v => v.severity === "warning");
+      emitProgress(emitter, makeEvent("guard:output", { ...eventBase, stage: "guard" }, {
+        message: `Output guard: ${critical.length} critical, ${warnings.length} warnings`,
+        detail: { violations: outputResult.violations, executorType: "local" }
+      }));
+      logger.info(`Output guard: ${outputResult.violations.length} violation(s) found`);
+      for (const v of outputResult.violations) {
+        logger.info(`  [${v.severity}] ${v.file}:${v.line} — ${v.message}`);
+      }
+      await addCheckpoint(session, { stage: "guard-output", iteration, pass: outputResult.pass, violations: outputResult.violations.length });
+
+      if (!outputResult.pass && config.guards.output.on_violation === "block") {
+        await markSessionStatus(session, "failed");
+        emitProgress(emitter, makeEvent("guard:blocked", { ...eventBase, stage: "guard" }, {
+          message: "Output guard blocked: critical violations detected",
+          detail: { violations: critical }
+        }));
+        return {
+          action: "return",
+          result: { approved: false, sessionId: session.id, reason: "guard_blocked", violations: critical }
+        };
+      }
+    }
+  }
+
+  if (perfEnabled) {
+    const perfResult = scanPerfDiff(diff, config);
+    if (!perfResult.skipped && perfResult.violations.length > 0) {
+      emitProgress(emitter, makeEvent("guard:perf", { ...eventBase, stage: "guard" }, {
+        message: `Perf guard: ${perfResult.violations.length} issue(s)`,
+        detail: { violations: perfResult.violations, executorType: "local" }
+      }));
+      logger.info(`Perf guard: ${perfResult.violations.length} issue(s) found`);
+      for (const v of perfResult.violations) {
+        logger.info(`  [${v.severity}] ${v.file}:${v.line} — ${v.message}`);
+      }
+      await addCheckpoint(session, { stage: "guard-perf", iteration, pass: perfResult.pass, violations: perfResult.violations.length });
+    }
+  }
+
+  return { action: "ok" };
+}
+
+async function runQualityGateStages({ config, logger, emitter, eventBase, session, trackBudget, i, askQuestion, repeatDetector, budgetSummary, sonarState, task, stageResults, coderRole, pipelineFlags, brainCtx }) {
+  const tddResult = await runTddCheckStage({ config, logger, emitter, eventBase, session, trackBudget, iteration: i, askQuestion, task, brainCtx });
+  if (tddResult.action === "pause") return { action: "return", result: tddResult.result };
+  if (tddResult.action === "continue") return { action: "continue" };
+
+  const skipSonarForTaskType = new Set(["infra", "doc", "no-code"]);
+  const effectiveTaskType = session.resolved_policies?.taskType || null;
+  if (config.sonarqube.enabled && !skipSonarForTaskType.has(effectiveTaskType)) {
+    const sonarResult = await runSonarStage({
+      config, logger, emitter, eventBase, session, trackBudget, iteration: i,
+      repeatDetector, budgetSummary, sonarState, askQuestion, task, brainCtx
+    });
+    if (sonarResult.action === "stalled" || sonarResult.action === "pause") return { action: "return", result: sonarResult.result };
+    if (sonarResult.action === "continue") return { action: "continue" };
+    if (sonarResult.stageResult) {
+      stageResults.sonar = sonarResult.stageResult;
+      await tryCiComment({ config, session, logger, agent: "Sonar", body: `SonarQube scan: ${sonarResult.stageResult.summary || "completed"}` });
+    }
+  }
+
+  if (config.sonarcloud?.enabled) {
+    const cloudResult = await runSonarCloudStage({
+      config, logger, emitter, eventBase, session, trackBudget, iteration: i
+    });
+    if (cloudResult.stageResult) {
+      stageResults.sonarcloud = cloudResult.stageResult;
+    }
+  }
+
+  if (pipelineFlags?.impeccableEnabled) {
+    const diff = await generateDiff({ baseRef: session.session_start_sha });
+    const impeccableMode = pipelineFlags?.impeccableMode || "audit";
+    const impeccableResult = await runImpeccableStage({
+      config, logger, emitter, eventBase, session, coderRole, trackBudget,
+      iteration: i, task, diff, mode: impeccableMode
+    });
+    if (impeccableResult.stageResult) {
+      stageResults.impeccable = impeccableResult.stageResult;
+    }
+  }
+
+  return { action: "ok" };
+}
+
+async function runReviewerGateStage({ pipelineFlags, reviewerRole, config, logger, emitter, eventBase, session, trackBudget, i, reviewRules, task, repeatDetector, budgetSummary, askQuestion, brainCtx }) {
+  if (!pipelineFlags.reviewerEnabled) {
+    return {
+      action: "ok",
+      review: { approved: true, blocking_issues: [], non_blocking_suggestions: [], summary: "Reviewer disabled by pipeline", confidence: 1 }
+    };
+  }
+
+  const reviewerResult = await runReviewerStage({
+    reviewerRole, config, logger, emitter, eventBase, session, trackBudget,
+    iteration: i, reviewRules, task, repeatDetector, budgetSummary, askQuestion, brainCtx
+  });
+  if (reviewerResult.action === "pause") return { action: "return", result: reviewerResult.result };
+  const revStandby = await handleStandbyResult({ stageResult: reviewerResult, session, emitter, eventBase, i, stage: "reviewer", logger, config, askQuestion });
+  if (revStandby.handled) {
+    if (revStandby.action === "return") return { action: "return", result: revStandby.result };
+    if (revStandby.action === "skip") {
+      // Solomon said skip review — treat as approved
+      return { action: "ok", review: { approved: true, blocking_issues: [], non_blocking_suggestions: [], summary: "Review skipped (agent rate-limited, Solomon approved)", confidence: 0.7 } };
+    }
+    if (revStandby.action === "retry_reviewer_only") {
+      // Retry just the reviewer — use alternative agent if Solomon recommended one
+      let retryReviewerRole = reviewerRole;
+      const alt = session._alternative_agent;
+      if (alt?.stage === "reviewer" && alt?.provider) {
+        const { createAgent } = await import("../agents/index.js");
+        retryReviewerRole = { provider: alt.provider, model: null };
+        logger.info(`Retrying reviewer with alternative agent: ${alt.provider}`);
+        delete session._alternative_agent;
+      }
+      return runReviewerGateStage({ pipelineFlags: { reviewerEnabled: true }, reviewerRole: retryReviewerRole, config, logger, emitter, eventBase, session, trackBudget, i, reviewRules, task, repeatDetector, budgetSummary, askQuestion });
+    }
+    return { action: "retry" };
+  }
+  if (reviewerResult.stalled) return { action: "return", result: reviewerResult.stalledResult };
+  return { action: "ok", review: reviewerResult.review };
+}
+
+async function handleApprovedReview({ config, session, emitter, eventBase, coderRole, trackBudget, i, task, stageResults, pipelineFlags, askQuestion, logger, gitCtx, budgetSummary, pgCard, pgProject, review, rtkTracker, brainCtx }) {
+  session.reviewer_retry_count = 0;
+  const postLoopResult = await handlePostLoopStages({
+    config, session, emitter, eventBase, coderRole, trackBudget, i, task, stageResults,
+    ciEnabled: Boolean(config.ci?.enabled), testerEnabled: pipelineFlags.testerEnabled, securityEnabled: pipelineFlags.securityEnabled, askQuestion, logger, brainCtx
+  });
+  if (postLoopResult.action === "return") return { action: "return", result: postLoopResult.result };
+  if (postLoopResult.action === "continue") return { action: "continue" };
+
+  const result = await finalizeApprovedSession({ config, gitCtx, task, logger, session, stageResults, emitter, eventBase, budgetSummary, pgCard, pgProject, review, i, rtkTracker });
+  return { action: "return", result };
+}
+
+async function handleMaxIterationsReached({ session, budgetSummary, emitter, eventBase, config, stageResults, logger, askQuestion, task, rtkTracker, brainCtx }) {
+  const budget = budgetSummary();
+
+  // Brain-owned decision: max_iterations is guidance, not a hard rule.
+  // Brain evaluates the feedback queue state to decide extend / finalize / escalate.
+  // Solomon is only consulted if Brain cannot decide on its own.
+  if (brainCtx?.enabled) {
+    const entries = brainCtx.feedbackQueue?.entries || [];
+    const pending = entries.map(e => ({ source: e.source, category: e.category, severity: e.severity, description: e.description }));
+    const hasSecurity = entries.some(e => e.category === "security" || e.source === "security");
+    const hasCorrectness = entries.some(e => ["correctness", "tests"].includes(e.category));
+    const hasStyleOnly = entries.length > 0 && !hasSecurity && !hasCorrectness;
+
+    if (hasSecurity) {
+      // Brain: security issues unresolved → cannot finalize, escalate
+      logger.warn(`Brain: max_iterations reached with ${entries.filter(e => e.category === "security" || e.source === "security").length} unresolved security issue(s) — cannot finalize`);
+      return { paused: true, sessionId: session.id, question: "Brain: unresolved security issues at max_iterations. Review manually or extend pipeline.", context: "brain_security_block", pending };
+    }
+
+    if (hasCorrectness) {
+      // Brain: correctness/test issues pending. Cap at MAX_EXTENSIONS to avoid infinite extensions.
+      const MAX_EXTENSIONS = 2;
+      if (brainCtx.extensionCount >= MAX_EXTENSIONS) {
+        logger.warn(`Brain: ${brainCtx.extensionCount} extensions exhausted with correctness issues still pending — escalating to human`);
+        return { paused: true, sessionId: session.id, question: `Brain exhausted ${MAX_EXTENSIONS} extensions with correctness/tests still pending. Manual intervention needed.`, context: "brain_extension_cap", pending };
+      }
+      brainCtx.extensionCount += 1;
+      logger.info(`Brain: max_iterations reached with ${entries.filter(e => ["correctness", "tests"].includes(e.category)).length} correctness issue(s) pending — extending iterations (extension ${brainCtx.extensionCount}/${MAX_EXTENSIONS})`);
+      session.reviewer_retry_count = 0;
+      await saveSession(session);
+      return { approved: false, sessionId: session.id, reason: "max_iterations_extended", extraIterations: Math.ceil(config.max_iterations / 2) };
+    }
+
+    if (entries.length === 0) {
+      // Brain: no pending feedback → last reviewer approved, finalize
+      logger.info("Brain: max_iterations reached with clean feedback queue — finalizing as approved");
+      return { approved: true, sessionId: session.id, reason: "brain_approved" };
+    }
+
+    // hasStyleOnly: genuine dilemma → Brain consults Solomon
+    logger.info(`Brain: max_iterations with ${entries.length} style-only issue(s) — consulting Solomon on dilemma`);
+    const { invokeSolomon: invokeSolomonAI } = await import("./solomon-escalation.js");
+    const solomonResult = await invokeSolomonAI({
+      config, logger, emitter, eventBase, stage: "max_iterations", askQuestion, session,
+      iteration: config.max_iterations,
+      conflict: {
+        stage: "brain-max-iterations",
+        task,
+        iterationCount: config.max_iterations,
+        maxIterations: config.max_iterations,
+        budget_usd: budget?.total_cost_usd || 0,
+        dilemma: `Max iterations reached with ${entries.length} style-only issue(s) pending. Accept as-is or request more work?`,
+        pendingIssues: pending,
+        history: [{ agent: "pipeline", feedback: session.last_reviewer_feedback || "Max iterations reached" }]
+      }
+    });
+    // Brain applies Solomon's decision
+    if (solomonResult.action === "approve") {
+      logger.info("Brain: Solomon advised approve for style-only pending — finalizing");
+      return { approved: true, sessionId: session.id, reason: "brain_solomon_approved" };
+    }
+    if (solomonResult.action === "continue") {
+      return { approved: false, sessionId: session.id, reason: "max_iterations_extended", extraIterations: solomonResult.extraIterations || Math.ceil(config.max_iterations / 2) };
+    }
+    if (solomonResult.action === "pause") {
+      return { paused: true, sessionId: session.id, question: solomonResult.question, context: "brain_solomon_dilemma" };
+    }
+    // Fallback: escalate to human
+    return { paused: true, sessionId: session.id, question: `Brain+Solomon cannot resolve: ${entries.length} pending issue(s) at max_iterations`, context: "max_iterations" };
+  }
+
+  // Legacy path (Brain disabled): original Solomon-driven flow
+  const solomonResult = await invokeSolomon({
+    config, logger, emitter, eventBase, stage: "max_iterations", askQuestion, session,
+    iteration: config.max_iterations,
+    conflict: {
+      stage: "max_iterations",
+      task,
+      iterationCount: config.max_iterations,
+      maxIterations: config.max_iterations,
+      budget_usd: budget?.total_cost_usd || 0,
+      history: [{ agent: "pipeline", feedback: session.last_reviewer_feedback || "Max iterations reached without reviewer approval" }]
+    }
+  });
+
+  if (solomonResult.action === "approve") {
+    logger.info("Solomon approved coder's work at max_iterations checkpoint");
+    return { approved: true, sessionId: session.id, reason: "solomon_approved" };
+  }
+
+  if (solomonResult.action === "continue") {
+    if (solomonResult.humanGuidance) {
+      session.last_reviewer_feedback = `Solomon guidance: ${solomonResult.humanGuidance}`;
+    }
+    session.reviewer_retry_count = 0;
+    await saveSession(session);
+    const extraIterations = solomonResult.extraIterations || config.max_iterations;
+    return { approved: false, sessionId: session.id, reason: "max_iterations_extended", humanGuidance: solomonResult.humanGuidance, extraIterations };
+  }
+
+  if (solomonResult.action === "pause") {
+    return { paused: true, sessionId: session.id, question: solomonResult.question, context: "max_iterations" };
+  }
+
+  if (solomonResult.action === "subtask") {
+    return { paused: true, sessionId: session.id, subtask: solomonResult.subtask, context: "max_iterations_subtask" };
+  }
+
+  // Solomon couldn't resolve — fail
+  session.budget = budgetSummary();
+  const rtkSavings = rtkTracker?.hasData() ? rtkTracker.summary() : undefined;
+  if (rtkSavings) session.rtk_savings = rtkSavings;
+  await markSessionStatus(session, "failed");
+  const failDetail = { approved: false, reason: "max_iterations", iterations: config.max_iterations, stages: stageResults, budget: budgetSummary() };
+  if (rtkSavings) failDetail.rtk_savings = rtkSavings;
+  emitProgress(
+    emitter,
+    makeEvent("session:end", { ...eventBase, stage: "done" }, {
+      status: "fail",
+      message: "Max iterations reached (Solomon could not resolve)",
+      detail: failDetail
+    })
+  );
+  return { approved: false, sessionId: session.id, reason: "max_iterations" };
+}
+
+async function tryAutoStartBoard(config, logger, emitter, eventBase) {
+  // TSK-0273: gate on hu_board.auto_start alone. The previous double-gate
+  // (enabled && auto_start) was confusing because `enabled` was never the
+  // user-facing switch — only auto_start was. This matches the AC: "kj run
+  // auto-inicia el board server si hu_board.auto_start es true".
+  if (!config.hu_board?.auto_start) return;
+  // Never auto-start during vitest: races the PID file and starts a detached
+  // process that outlives the test run.
+  if (process.env.VITEST || process.env.NODE_ENV === "test") return;
+
+  try {
+    const { startBoard, renderBoardBanner } = await import("../commands/board.js");
+    const boardPort = config.hu_board.port || 4000;
+    const boardResult = await startBoard(boardPort);
+    const status = boardResult.alreadyRunning ? "already running" : "started";
+    // Highlighted URL box — visible regardless of log level (matches the
+    // post-planner auto-HU banner so the user gets the same signal either
+    // way the board got started).
+    console.log(renderBoardBanner({ url: boardResult.url, status }));
+    logger.info(`HU Board ${status} at ${boardResult.url}`);
+    emitProgress(emitter, makeEvent("board:started", eventBase, {
+      message: `HU Board running at ${boardResult.url}`,
+      detail: { pid: boardResult.pid, port: boardPort, url: boardResult.url }
+    }));
+  } catch (err) {
+    logger.warn(`HU Board auto-start failed (non-blocking): ${err.message}`);
+  }
+}
+
+async function initFlowContext({ task, config, logger, emitter, askQuestion, pgTaskId, pgProject, flags }) {
+  // Auto-init .karajan/ if missing (copies coder-rules, review-rules, role templates)
+  const initProjectDir = config.projectDir || process.cwd();
+  await autoInit(initProjectDir, logger);
+
+  // Smart role assignment: detect installed AIs and assign to roles
+  // Only runs if: (a) no roles configured AND (b) not in test environment
+  const needsAssignment = !config.roles?.coder?.provider && !config.coder && process.env.NODE_ENV !== "test" && !process.env.VITEST;
+  if (needsAssignment) {
+    try {
+      const { autoAssignRoles, applyRoleAssignments } = await import("../utils/role-assigner.js");
+      const { assignments } = await autoAssignRoles(logger);
+      if (assignments) config = applyRoleAssignments({ ...config }, assignments);
+    } catch { /* non-blocking — defaults will be used */ }
+  }
+
+  // Scope all git diffs to projectDir (prevents leaking unrelated branch changes)
+  // When running from a subdirectory of a git repo, use relative path as scope
+  let diffScope = config.projectDir || null;
+  if (!diffScope) {
+    try {
+      const { execSync } = await import("node:child_process");
+      const repoRoot = execSync("git rev-parse --show-toplevel", { encoding: "utf8" }).trim();
+      const cwd = process.cwd();
+      if (cwd !== repoRoot && cwd.startsWith(repoRoot)) {
+        diffScope = cwd.slice(repoRoot.length + 1);
+        logger.info(`Running from subdirectory — diff scoped to ${diffScope}/`);
+      }
+    } catch { /* git not available */ }
+  }
+  setDiffProjectDir(diffScope);
+
+  // Auto-detect Chrome DevTools MCP
+  const { detectDevToolsMcp } = await import("../webperf/devtools-detect.js");
+  const devToolsAvailable = await detectDevToolsMcp(logger);
+  if (devToolsAvailable) {
+    config = { ...config, webperf: { ...config.webperf, devtools_mcp: true } };
+  }
+
+  const ctx = new PipelineContext({ config, session: null, logger, emitter, task, flags });
+  ctx.askQuestion = askQuestion;
+  ctx.pgTaskId = pgTaskId;
+  ctx.pgProject = pgProject;
+
+  ctx.coderRole = resolveRole(config, "coder");
+  ctx.reviewerRole = resolveRole(config, "reviewer");
+  ctx.refactorerRole = resolveRole(config, "refactorer");
+  ctx.pipelineFlags = resolvePipelineFlags(config);
+  ctx.repeatDetector = new RepeatDetector({ threshold: getRepeatThreshold(config) });
+  ctx.coderRoleInstance = new CoderRole({ config, logger, emitter, createAgentFn: createAgent, askHost: askQuestion });
+  ctx.startedAt = Date.now();
+  ctx.eventBase = { sessionId: null, iteration: 0, stage: null, startedAt: ctx.startedAt };
+  const { budgetTracker, budgetLimit, budgetSummary, trackBudget } = createBudgetManager({
+    config,
+    emitter,
+    eventBase: ctx.eventBase,
+    // Provides fresh compression data for "With KJ vs Without KJ" comparison
+    // (KJC-TSK-0274). Invoked on every budgetSummary() call.
+    getCompressionStats: () => ({
+      rtkSavings: ctx.rtkTracker?.hasData() ? ctx.rtkTracker.summary() : (ctx.session?.rtk_savings || null),
+      brainCtx: ctx.brainCtx || null,
+    }),
+  });
+  ctx.budgetTracker = budgetTracker;
+  ctx.budgetLimit = budgetLimit;
+  ctx.budgetSummary = budgetSummary;
+  ctx.trackBudget = trackBudget;
+
+  // --- RTK detection ---
+  const rtkResult = await detectRtk();
+  if (rtkResult.available) {
+    config = { ...config, rtk: { available: true, version: rtkResult.version } };
+    const rtkTracker = new RtkSavingsTracker();
+    const rtkRunner = createRtkRunner(true, rtkTracker);
+    setDiffRunner(rtkRunner);
+    setGitRunner(rtkRunner);
+    ctx.rtkTracker = rtkTracker;
+    logger.info(`RTK detected (${rtkResult.version}) — wrapping internal git/diff commands with rtk`);
+    emitProgress(emitter, makeEvent("rtk:detected", ctx.eventBase, {
+      message: "RTK detected — internal commands wrapped for token optimization",
+      detail: { version: rtkResult.version, executorType: "local" }
+    }));
+  }
+
+  // --- HU Board auto-start ---
+  await tryAutoStartBoard(config, logger, emitter, ctx.eventBase);
+
+  // --- Product Context ---
+  const ctxProjectDir = config.projectDir || process.cwd();
+  const { content: productContext, source: productContextSource } = await loadProductContext(ctxProjectDir);
+  if (productContext) {
+    config = { ...config, productContext };
+    logger.info(`Product context loaded from ${productContextSource}`);
+    emitProgress(emitter, makeEvent("context:loaded", ctx.eventBase, {
+      message: "Product context loaded",
+      detail: { source: productContextSource }
+    }));
+  }
+
+  ctx.session = await initializeSession({ task, config, flags, pgTaskId, pgProject });
+  ctx.eventBase.sessionId = ctx.session.id;
+
+  // Karajan Brain: initialize runtime context (opt-in via config.brain.enabled)
+  const { createBrainContext, isBrainEnabled } = await import("./brain-coordinator.js");
+  ctx.brainCtx = createBrainContext({ enabled: isBrainEnabled(config) });
+  if (ctx.brainCtx.enabled) {
+    logger.info("Karajan Brain enabled — feedback queue, verification, compression active");
+  }
+
+  const { initPgAdapter } = await import("../planning-game/pipeline-adapter.js");
+  const pgAdapterResult = await initPgAdapter({ session: ctx.session, config, logger, pgTaskId, pgProject });
+  ctx.pgCard = pgAdapterResult.pgCard;
+  ctx.session.pg_card = ctx.pgCard || null;
+
+  emitProgress(
+    emitter,
+    makeEvent("session:start", ctx.eventBase, {
+      message: "Session started",
+      detail: { task, coder: ctx.coderRole.provider, reviewer: ctx.reviewerRole.provider, maxIterations: config.max_iterations }
+    })
+  );
+
+  ctx.stageResults = {};
+  ctx.sonarState = { issuesInitial: null, issuesFinal: null };
+
+  const preLoopResult = await runPreLoopStages({ config, logger, emitter, eventBase: ctx.eventBase, session: ctx.session, flags, pipelineFlags: ctx.pipelineFlags, coderRole: ctx.coderRole, trackBudget: ctx.trackBudget, task, askQuestion, pgTaskId, pgProject, stageResults: ctx.stageResults, brainCtx: ctx.brainCtx });
+  ctx.plannedTask = preLoopResult.plannedTask;
+  ctx.config = preLoopResult.updatedConfig;
+
+  // --- Session Journal: persist pre-loop outputs + display plan summary ---
+  const reportDir = ctx.config.output?.report_dir || ".reviews";
+  try {
+    ctx.journalDir = await createJournalDir(reportDir, ctx.session.id);
+    const journalFiles = await writePreLoopJournal(ctx.journalDir, ctx.stageResults);
+    ctx.journalFiles = journalFiles;
+    ctx.journalIterations = [];
+    ctx.journalDecisions = [];
+
+    // Attach journal state to session so finalizeApprovedSession can access it
+    ctx.session._journalDir = ctx.journalDir;
+    ctx.session._journalFiles = journalFiles;
+    ctx.session._journalIterations = ctx.journalIterations;
+    ctx.session._journalDecisions = ctx.journalDecisions;
+    ctx.session._startedAt = ctx.startedAt;
+
+    // Display plan summary in console before iteration loop
+    const planSummary = buildPlanSummary({
+      pipelineFlags: ctx.pipelineFlags,
+      config: ctx.config,
+      stageResults: ctx.stageResults,
+      task
+    });
+    console.log(planSummary);
+  } catch (err) {
+    logger.warn(`Journal init failed (non-blocking): ${err.message}`);
+    ctx.journalDir = null;
+    ctx.journalFiles = [];
+    ctx.journalIterations = [];
+    ctx.journalDecisions = [];
+  }
+
+  ctx.gitCtx = await prepareGitAutomation({ config: ctx.config, task, logger, session: ctx.session });
+  const projectDir = ctx.config.projectDir || process.cwd();
+  ctx.reviewRules = (await resolveReviewProfile({ mode: ctx.config.review_mode, projectDir })).rules;
+  await ctx.coderRoleInstance.init();
+
+  return ctx;
+}
+
+async function runSingleIteration(ctx) {
+  // Use plannedTask (HU-scoped or planner-enriched) over the raw original task.
+  // When running per-HU sub-pipelines, plannedTask is the HU's text, not the full spec.
+  const { config, logger, emitter, eventBase, session, iteration: i } = ctx;
+  const task = ctx.plannedTask || ctx.task;
+
+  const iterStart = Date.now();
+  const ciEnabled = Boolean(config.ci?.enabled) && ctx.gitCtx?.enabled;
+  logger.setContext({ iteration: i, stage: "iteration" });
+
+  const reviewerRetryCount = session.reviewer_retry_count || 0;
+  const maxReviewerRetries = config.session.max_reviewer_retries ?? config.session.fail_fast_repeats;
+  const iterLang = getLang(config);
+  const iterMsg = msg("pipeline_iteration", iterLang, { current: i, max: config.max_iterations });
+  emitProgress(emitter, makeEvent("iteration:start", { ...eventBase, stage: "iteration" }, {
+    message: iterMsg,
+    detail: { iteration: i, maxIterations: config.max_iterations, reviewerRetryCount, maxReviewerRetries }
+  }));
+  logger.info(iterMsg);
+
+  const crResult = await runCoderAndRefactorerStages({
+    coderRoleInstance: ctx.coderRoleInstance, coderRole: ctx.coderRole, refactorerRole: ctx.refactorerRole,
+    pipelineFlags: ctx.pipelineFlags, config, logger, emitter, eventBase, session,
+    plannedTask: ctx.plannedTask, trackBudget: ctx.trackBudget, i, brainCtx: ctx.brainCtx
+  });
+  if (crResult.action === "return" || crResult.action === "retry") return crResult;
+
+  const guardResult = await runGuardStages({ config, logger, emitter, eventBase, session, iteration: i });
+  if (guardResult.action === "return") return guardResult;
+
+  const qgResult = await runQualityGateStages({
+    config, logger, emitter, eventBase, session, trackBudget: ctx.trackBudget, i,
+    askQuestion: ctx.askQuestion, repeatDetector: ctx.repeatDetector, budgetSummary: ctx.budgetSummary,
+    sonarState: ctx.sonarState, task, stageResults: ctx.stageResults, coderRole: ctx.coderRole,
+    pipelineFlags: ctx.pipelineFlags, brainCtx: ctx.brainCtx
+  });
+  if (qgResult.action === "return" || qgResult.action === "continue") return qgResult;
+
+  await handleCiEarlyPrOrPush({
+    ciEnabled, config, session, emitter, eventBase, gitCtx: ctx.gitCtx, task, logger,
+    stageResults: ctx.stageResults, i
+  });
+
+  const revResult = await runReviewerGateStage({
+    pipelineFlags: ctx.pipelineFlags, reviewerRole: ctx.reviewerRole, config, logger, emitter, eventBase,
+    session, trackBudget: ctx.trackBudget, i, reviewRules: ctx.reviewRules, task,
+    repeatDetector: ctx.repeatDetector, budgetSummary: ctx.budgetSummary, askQuestion: ctx.askQuestion,
+    brainCtx: ctx.brainCtx
+  });
+  if (revResult.action === "return" || revResult.action === "retry") return revResult;
+  const review = revResult.review;
+
+  const iterDuration = Date.now() - iterStart;
+  emitProgress(emitter, makeEvent("iteration:end", { ...eventBase, stage: "iteration" }, {
+    message: `Iteration ${i} completed`, detail: { duration: iterDuration }
+  }));
+  session.standby_retry_count = 0;
+
+  // --- Journal: record iteration ---
+  // Uses the richer iteration-logger from TSK-0286: captures blocking-issue
+  // details, sonar counts and Solomon rulings in addition to summaries.
+  if (ctx.journalIterations) {
+    try {
+      const { recordIteration, extractIterationData } = await import("../session/journal/iteration-logger.js");
+      const iterationData = extractIterationData({
+        iteration: i,
+        durationMs: iterDuration,
+        stageResults: ctx.stageResults,
+        session: ctx.session,
+      });
+      // Derive a richer reviewer summary if none was produced.
+      if (!iterationData.reviewerSummary && review) {
+        iterationData.reviewerSummary = review.approved
+          ? `Approved: ${review.raw_summary || ""}`
+          : `Rejected: ${(review?.blocking_issues || []).length} blocking issue(s)`;
+      }
+      if (!iterationData.blockingIssues || iterationData.blockingIssues.length === 0) {
+        iterationData.blockingIssues = review?.blocking_issues || [];
+      }
+      recordIteration(ctx.session, iterationData);
+    } catch (err) {
+      logger.warn(`Iteration journal record failed (non-blocking): ${err.message}`);
+    }
+  }
+
+  const solomonResult = await handleSolomonCheck({
+    config, session, emitter, eventBase, logger, task, i, askQuestion: ctx.askQuestion,
+    ciEnabled, blockingIssues: review?.blocking_issues, brainCtx: ctx.brainCtx
+  });
+  if (solomonResult.action === "pause") return { action: "return", result: solomonResult.result };
+
+  await handleCiReviewDispatch({ ciEnabled, config, session, review, i, logger });
+
+  if (review.approved) {
+    const approvedResult = await handleApprovedReview({
+      config, session, emitter, eventBase, coderRole: ctx.coderRole, trackBudget: ctx.trackBudget, i, task,
+      stageResults: ctx.stageResults, pipelineFlags: ctx.pipelineFlags, askQuestion: ctx.askQuestion, logger,
+      gitCtx: ctx.gitCtx, budgetSummary: ctx.budgetSummary, pgCard: ctx.pgCard, pgProject: ctx.pgProject, review,
+      rtkTracker: ctx.rtkTracker, brainCtx: ctx.brainCtx
+    });
+    if (approvedResult.action === "return" || approvedResult.action === "continue") return approvedResult;
+  }
+
+  // Solomon already evaluated the rejection in runReviewerStage -> handleReviewerRejection
+  // Only use retry counter as fallback if Solomon is disabled
+  if (!config.pipeline?.solomon?.enabled) {
+    const retryResult = await handleReviewerRetryAndSolomon({ config, session, emitter, eventBase, logger, review, task, i, askQuestion: ctx.askQuestion });
+    if (retryResult.action === "return") return retryResult;
+  } else {
+    // Solomon is enabled — feed back the blocking issues for the next coder iteration
+    session.last_reviewer_feedback = review.blocking_issues
+      .map((x) => {
+        const parts = [`[${x.severity || "high"}] ${x.id || "ISSUE"}: ${x.description || "Missing description"}`];
+        if (x.file) parts.push(`  File: ${x.file}${x.line ? `:${x.line}` : ""}`);
+        if (x.suggested_fix) parts.push(`  Fix: ${x.suggested_fix}`);
+        return parts.join("\n");
+      })
+      .join("\n\n");
+    await saveSession(session);
+  }
+
+  return { action: "next" };
+}
+
+async function writeHistoryRecord({ sessionId, task, result, logger }) {
+  try {
+    const { createHistoryRecord } = await import("../hu/store.js");
+    const approved = Boolean(result?.approved);
+    const summary = result?.review?.summary || result?.reason || null;
+    await createHistoryRecord(sessionId, {
+      task,
+      result: JSON.stringify(result),
+      approved,
+      summary
+    });
+  } catch (err) {
+    logger.warn(`HU history record failed (non-blocking): ${err.message}`);
+  }
+}
+
+/**
+ * Run the standard iteration loop for a given task using the pipeline context.
+ * Returns a result object with at least { approved: boolean }.
+ * Used both as the main loop and as the per-HU callback in sub-pipeline mode.
+ */
+async function runIterationLoop(ctx, { task: loopTask, askQuestion, emitter, logger }) {
+  ctx.plannedTask = loopTask;
+
+  const checkpointIntervalMs = (ctx.config.session.checkpoint_interval_minutes ?? 5) * 60 * 1000;
+  let lastCheckpointAt = Date.now();
+  let checkpointDisabled = false;
+  let lastCheckpointSnapshot = takeCheckpointSnapshot(ctx.session);
+
+  let i = 0;
+  while (i < ctx.config.max_iterations) {
+    i += 1;
+    const elapsedMinutes = (Date.now() - ctx.startedAt) / 60000;
+
+    const cpResult = await handleCheckpoint({
+      checkpointDisabled, askQuestion, lastCheckpointAt, checkpointIntervalMs, elapsedMinutes,
+      i, config: ctx.config, budgetTracker: ctx.budgetTracker, stageResults: ctx.stageResults, emitter, eventBase: ctx.eventBase, session: ctx.session, budgetSummary: ctx.budgetSummary, lastCheckpointSnapshot
+    });
+    if (cpResult.action === "stop") {
+      return cpResult.result;
+    }
+    checkpointDisabled = cpResult.checkpointDisabled;
+    lastCheckpointAt = cpResult.lastCheckpointAt;
+    if (cpResult.lastCheckpointSnapshot !== undefined) lastCheckpointSnapshot = cpResult.lastCheckpointSnapshot;
+
+    await checkSessionTimeout({ askQuestion, elapsedMinutes, config: ctx.config, session: ctx.session, emitter, eventBase: ctx.eventBase, i, budgetSummary: ctx.budgetSummary });
+    await checkBudgetExceeded({ budgetTracker: ctx.budgetTracker, config: ctx.config, session: ctx.session, emitter, eventBase: ctx.eventBase, i, budgetLimit: ctx.budgetLimit, budgetSummary: ctx.budgetSummary });
+
+    ctx.eventBase.iteration = i;
+    ctx.iteration = i;
+
+    let iterResult;
+    try {
+      iterResult = await runSingleIteration(ctx);
+    } catch (stageError) {
+      // ANY unhandled error in a stage = out of normal flow → Solomon decides
+      logger.warn(`Stage error caught — escalating to Solomon: ${stageError.message}`);
+      const solomonResult = await invokeSolomon({
+        config: ctx.config, logger, emitter, eventBase: ctx.eventBase,
+        stage: "stage_error", askQuestion, session: ctx.session, iteration: i,
+        conflict: {
+          stage: "stage_error",
+          task: loopTask,
+          iterationCount: i,
+          maxIterations: ctx.config.max_iterations,
+          budget_usd: ctx.budgetSummary()?.total_cost_usd || 0,
+          history: [{ agent: "pipeline", feedback: `Stage threw: ${stageError.message}` }]
+        }
+      });
+
+      if (solomonResult.action === "approve") {
+        logger.info("Solomon approved despite stage error");
+        return { approved: true, sessionId: ctx.session.id, reason: "solomon_approved_after_error" };
+      }
+      if (solomonResult.action === "continue") {
+        if (solomonResult.humanGuidance) {
+          ctx.session.last_reviewer_feedback = `Solomon guidance: ${solomonResult.humanGuidance}`;
+        }
+        continue; // next iteration
+      }
+      if (solomonResult.action === "pause") {
+        return { paused: true, sessionId: ctx.session.id, question: solomonResult.question, context: "stage_error" };
+      }
+      // Solomon couldn't resolve — fail
+      await markSessionStatus(ctx.session, "failed");
+      return { approved: false, sessionId: ctx.session.id, reason: "stage_error", error: stageError.message };
+    }
+
+    if (iterResult.action === "return") {
+      return iterResult.result;
+    }
+    if (iterResult.action === "retry") { i -= 1; }
+  }
+
+  // Solomon decides whether to extend iterations or stop
+  const maxIterResult = await handleMaxIterationsReached({ session: ctx.session, budgetSummary: ctx.budgetSummary, emitter, eventBase: ctx.eventBase, config: ctx.config, stageResults: ctx.stageResults, logger, askQuestion, task: loopTask, rtkTracker: ctx.rtkTracker, brainCtx: ctx.brainCtx });
+
+  // Solomon said "continue" — extend iterations and keep going
+  if (maxIterResult.reason === "max_iterations_extended") {
+    const extra = maxIterResult.extraIterations || ctx.config.max_iterations;
+    ctx.config.max_iterations += extra;
+    logger.info(`Solomon extended pipeline by ${extra} iterations (new max: ${ctx.config.max_iterations})`);
+
+    if (maxIterResult.humanGuidance) {
+      ctx.session.last_reviewer_feedback = `Solomon guidance: ${maxIterResult.humanGuidance}`;
+    }
+
+    // Continue the loop
+    while (i < ctx.config.max_iterations) {
+      i += 1;
+      const elapsedMinutes = (Date.now() - ctx.startedAt) / 60000;
+
+      const cpResult = await handleCheckpoint({
+        checkpointDisabled, askQuestion, lastCheckpointAt, checkpointIntervalMs, elapsedMinutes,
+        i, config: ctx.config, budgetTracker: ctx.budgetTracker, stageResults: ctx.stageResults, emitter, eventBase: ctx.eventBase, session: ctx.session, budgetSummary: ctx.budgetSummary, lastCheckpointSnapshot
+      });
+      if (cpResult.action === "stop") return cpResult.result;
+      checkpointDisabled = cpResult.checkpointDisabled;
+      lastCheckpointAt = cpResult.lastCheckpointAt;
+      if (cpResult.lastCheckpointSnapshot !== undefined) lastCheckpointSnapshot = cpResult.lastCheckpointSnapshot;
+
+      await checkSessionTimeout({ askQuestion, elapsedMinutes, config: ctx.config, session: ctx.session, emitter, eventBase: ctx.eventBase, i, budgetSummary: ctx.budgetSummary });
+      await checkBudgetExceeded({ budgetTracker: ctx.budgetTracker, config: ctx.config, session: ctx.session, emitter, eventBase: ctx.eventBase, i, budgetLimit: ctx.budgetLimit, budgetSummary: ctx.budgetSummary });
+
+      ctx.eventBase.iteration = i;
+      ctx.iteration = i;
+
+      const iterResult = await runSingleIteration(ctx);
+      if (iterResult.action === "return") return iterResult.result;
+      if (iterResult.action === "retry") { i -= 1; }
+    }
+
+    // Extended iterations also exhausted — final Solomon call
+    const finalResult = await handleMaxIterationsReached({ session: ctx.session, budgetSummary: ctx.budgetSummary, emitter, eventBase: ctx.eventBase, config: ctx.config, stageResults: ctx.stageResults, logger, askQuestion, task: loopTask, rtkTracker: ctx.rtkTracker, brainCtx: ctx.brainCtx });
+    return finalResult;
+  }
+
+  return maxIterResult;
+}
+
+export async function runFlow({ task, config, logger, flags = {}, emitter = null, askQuestion = null, pgTaskId = null, pgProject = null }) {
+  const pipelineFlags = resolvePipelineFlags(config);
+
+  if (flags.dryRun) {
+    return handleDryRun({ task, config, flags, emitter, pipelineFlags });
+  }
+
+  let ctx;
+  try {
+    ctx = await initFlowContext({ task, config, logger, emitter, askQuestion, pgTaskId, pgProject, flags });
+  } catch (initError) {
+    // Pre-loop stage failure → Solomon decides
+    logger.warn(`Init/pre-loop error — escalating to Solomon: ${initError.message}`);
+    const tempSession = { id: "init-error", task, status: "failed" };
+    const solomonResult = await invokeSolomon({
+      config, logger, emitter, eventBase: { sessionId: "init-error", iteration: 0, stage: "init", startedAt: Date.now() },
+      stage: "init_error", askQuestion, session: tempSession, iteration: 0,
+      conflict: {
+        stage: "init_error",
+        task,
+        iterationCount: 0,
+        maxIterations: config.max_iterations || 5,
+        history: [{ agent: "pipeline", feedback: `Initialization failed: ${initError.message}` }]
+      }
+    });
+    if (solomonResult.action === "pause") {
+      return { paused: true, sessionId: "init-error", question: solomonResult.question, context: "init_error" };
+    }
+    throw initError; // Solomon couldn't resolve — propagate
+  }
+
+  try {
+    // --- Analysis-only flow: skip coder/reviewer when coderRequired === false ---
+    if (ctx.pipelineFlags.coderRequired === false) {
+      logger.info("Analysis-only task — skipping coder/reviewer iteration loop");
+      emitProgress(emitter, makeEvent("pipeline:analysis-only", { ...ctx.eventBase, stage: "analysis" }, {
+        message: "Analysis-only task — running security and audit stages only",
+        detail: { taskType: ctx.session.resolved_policies?.taskType, coderRequired: false }
+      }));
+
+      const analysisStageResults = ctx.stageResults;
+      const postLoopDiff = await generateDiff({ baseRef: ctx.session.session_start_sha });
+
+      if (ctx.pipelineFlags.securityEnabled) {
+        const securityResult = await runSecurityStage({
+          config: ctx.config, logger, emitter, eventBase: ctx.eventBase, session: ctx.session,
+          coderRole: ctx.coderRole, trackBudget: ctx.trackBudget,
+          iteration: 1, task: ctx.plannedTask, diff: postLoopDiff, askQuestion, brainCtx: ctx.brainCtx
+        });
+        if (securityResult.stageResult) analysisStageResults.security = securityResult.stageResult;
+      }
+
+      const auditResult = await runFinalAuditStage({
+        config: ctx.config, logger, emitter, eventBase: ctx.eventBase, session: ctx.session,
+        coderRole: ctx.coderRole, trackBudget: ctx.trackBudget,
+        iteration: 1, task: ctx.plannedTask, diff: postLoopDiff
+      });
+      if (auditResult.stageResult) analysisStageResults.audit = auditResult.stageResult;
+
+      ctx.session.budget = ctx.budgetSummary();
+      await markSessionStatus(ctx.session, "approved");
+
+      const analysisResult = {
+        approved: true,
+        sessionId: ctx.session.id,
+        analysisOnly: true,
+        stages: analysisStageResults,
+        budget: ctx.budgetSummary()
+      };
+      await writeHistoryRecord({ sessionId: ctx.session.id, task, result: analysisResult, logger });
+
+      emitProgress(emitter, makeEvent("session:end", { ...ctx.eventBase, stage: "done" }, {
+        message: "Analysis-only session completed",
+        detail: analysisResult
+      }));
+
+      return analysisResult;
+    }
+
+    // --- HU Sub-Pipeline: run each certified HU as an independent iteration loop ---
+    if (needsSubPipeline(ctx.stageResults.huReviewer)) {
+      logger.info(`HU sub-pipeline: ${ctx.stageResults.huReviewer.certified} certified stories — running each as a sub-pipeline`);
+      emitProgress(emitter, makeEvent("hu:sub-pipeline:start", { ...ctx.eventBase, stage: "hu-sub-pipeline" }, {
+        message: `Running ${ctx.stageResults.huReviewer.certified} HUs as sub-pipelines`,
+        detail: { total: ctx.stageResults.huReviewer.total, certified: ctx.stageResults.huReviewer.certified }
+      }));
+
+      // Per-HU pipeline: focused max_iterations, fresh Brain state, own git branch.
+      const originalMaxIterations = ctx.config.max_iterations;
+      const huMaxIterations = ctx.config.hu_max_iterations ?? 3;
+      const huBranches = new Map();
+      const { prepareHuBranch, finalizeHuCommit } = await import("../git/hu-automation.js");
+      const subPipelineResult = await runHuSubPipeline({
+        huReviewerResult: ctx.stageResults.huReviewer,
+        runIterationFn: async (huTask, story) => {
+          ctx.config.max_iterations = huMaxIterations;
+          if (ctx.brainCtx?.enabled) {
+            ctx.brainCtx.extensionCount = 0;
+            const { createBrainContext } = await import("./brain-coordinator.js");
+            const fresh = createBrainContext({ enabled: true });
+            ctx.brainCtx.feedbackQueue = fresh.feedbackQueue;
+            ctx.brainCtx.verificationTracker = fresh.verificationTracker;
+          }
+          // Apply per-HU policies based on task_type (infra skips reviewer/sonar/tdd)
+          const { applyPolicies } = await import("../guards/policy-resolver.js");
+          const huPolicies = applyPolicies({ taskType: story.task_type, policies: ctx.config.policies });
+          const savedFlags = { ...ctx.pipelineFlags };
+          if (!huPolicies.reviewer) ctx.pipelineFlags.reviewerEnabled = false;
+          if (!huPolicies.tdd) ctx.config.development = { ...ctx.config.development, methodology: "standard", require_test_changes: false };
+          if (!huPolicies.sonar) ctx.config.sonarqube = { ...ctx.config.sonarqube, enabled: false };
+          if (!huPolicies.testsRequired) ctx.pipelineFlags.testerEnabled = false;
+          logger.info(`HU ${story.id} (${story.task_type}): policies → reviewer=${huPolicies.reviewer}, tdd=${huPolicies.tdd}, sonar=${huPolicies.sonar}, tests=${huPolicies.testsRequired}`);
+
+          const branchName = await prepareHuBranch({ story, huBranches, config: ctx.config, logger });
+          const projectDir = ctx.config.projectDir || process.cwd();
+
+          // If HU has acceptance_tests, Brain runs them as the gate instead of
+          // the standard reviewer/tester pipeline. This is the radical fix:
+          // concrete executable tests replace subjective reviewer opinions.
+          if (story.acceptance_tests?.length > 0) {
+            const { runAcceptanceTests, buildDiagnosticPrompt } = await import("../hu/acceptance-runner.js");
+
+            for (let attempt = 1; attempt <= ctx.config.max_iterations; attempt++) {
+              logger.info(`HU ${story.id}: coder iteration ${attempt}/${ctx.config.max_iterations}`);
+              emitProgress(emitter, makeEvent("iteration:start", { ...ctx.eventBase, stage: "iteration" }, {
+                message: `Iteration ${attempt}/${ctx.config.max_iterations}`,
+                detail: { iteration: attempt, maxIterations: ctx.config.max_iterations }
+              }));
+
+              // Coder runs with the HU task + any diagnostic feedback from previous attempt
+              const coderResult = await runCoderStage({
+                coderRoleInstance: ctx.coderRoleInstance, coderRole: ctx.coderRole,
+                config: ctx.config, logger, emitter, eventBase: ctx.eventBase,
+                session: ctx.session, plannedTask: ctx.plannedTask,
+                trackBudget: ctx.trackBudget, iteration: attempt, brainCtx: ctx.brainCtx
+              });
+              if (coderResult?.action === "standby" || coderResult?.action === "pause") {
+                return coderResult?.result || { approved: false, reason: "coder_failed" };
+              }
+
+              // Sonar quality gate (sw task_type only, when policies say sonar=true)
+              if (huPolicies.sonar && ctx.config.sonarqube?.enabled) {
+                try {
+                  const sonarResult = await runSonarStage({
+                    config: ctx.config, logger, emitter, eventBase: ctx.eventBase,
+                    session: ctx.session, trackBudget: ctx.trackBudget, iteration: attempt,
+                    askQuestion, brainCtx: ctx.brainCtx
+                  });
+                  if (sonarResult?.action === "continue") {
+                    // Sonar failed — add to feedback for next coder attempt
+                    ctx.plannedTask = `${huTask}\n\n--- SONAR FAILURE ---\n${ctx.session.last_reviewer_feedback}`;
+                    continue;
+                  }
+                } catch (err) {
+                  logger.warn(`HU ${story.id}: sonar stage failed (non-blocking): ${err.message}`);
+                }
+              }
+
+              // Brain runs acceptance tests
+              logger.info(`HU ${story.id}: running ${story.acceptance_tests.length} acceptance tests`);
+              emitProgress(emitter, makeEvent("hu:acceptance-start", { ...ctx.eventBase, stage: "acceptance" }, {
+                message: `Running ${story.acceptance_tests.length} acceptance tests`,
+                detail: { huId: story.id, testCount: story.acceptance_tests.length }
+              }));
+
+              const testResult = await runAcceptanceTests(story.acceptance_tests, projectDir);
+              emitProgress(emitter, makeEvent("hu:acceptance-end", { ...ctx.eventBase, stage: "acceptance" }, {
+                status: testResult.allPassed ? "ok" : "fail",
+                message: testResult.summary,
+                detail: { allPassed: testResult.allPassed, results: testResult.results.map(r => ({ cmd: r.cmd, passed: r.passed })) }
+              }));
+
+              if (testResult.allPassed) {
+                logger.info(`HU ${story.id}: all acceptance tests PASSED — approved`);
+                await finalizeHuCommit({ story, branchName, config: ctx.config, logger });
+                return { approved: true, sessionId: ctx.session.id, reason: "acceptance_tests_passed" };
+              }
+
+              // Brain diagnoses failures and sends concrete fix to coder
+              const failed = testResult.results.filter(r => !r.passed);
+              const diagnostic = buildDiagnosticPrompt(failed);
+              logger.warn(`HU ${story.id}: ${failed.length} acceptance test(s) FAILED — sending diagnostic to coder`);
+              ctx.session.last_reviewer_feedback = diagnostic;
+              ctx.plannedTask = `${huTask}\n\n--- ACCEPTANCE TEST FAILURES ---\n${diagnostic}`;
+            }
+
+            // All iterations exhausted
+            logger.warn(`HU ${story.id}: max iterations reached with acceptance tests still failing`);
+            return { approved: false, sessionId: ctx.session.id, reason: "acceptance_tests_failed" };
+          }
+
+          // Fallback: no acceptance_tests → standard pipeline (reviewer/tester)
+          try {
+            const result = await runIterationLoop(ctx, { task: huTask, askQuestion, emitter, logger });
+            if (result?.approved) {
+              await finalizeHuCommit({ story, branchName, config: ctx.config, logger });
+            }
+            return result;
+          } finally {
+            ctx.config.max_iterations = originalMaxIterations;
+            Object.assign(ctx.pipelineFlags, savedFlags);
+          }
+        },
+        emitter,
+        eventBase: ctx.eventBase,
+        logger,
+        config: ctx.config
+      });
+
+      emitProgress(emitter, makeEvent("hu:sub-pipeline:end", { ...ctx.eventBase, stage: "hu-sub-pipeline" }, {
+        status: subPipelineResult.approved ? "ok" : "fail",
+        message: subPipelineResult.approved
+          ? "All HUs completed successfully"
+          : `Sub-pipeline finished with failures (${subPipelineResult.blockedIds.length} blocked)`,
+        detail: { results: subPipelineResult.results, blockedIds: subPipelineResult.blockedIds }
+      }));
+
+      // Sync results back to plan file if this was a plan-based run
+      if (ctx.session._syncResultsToPlan) {
+        try {
+          await ctx.session._syncResultsToPlan(subPipelineResult);
+          logger.info("Plan updated with execution results");
+        } catch (err) {
+          logger.warn(`Failed to sync results to plan: ${err.message}`);
+        }
+      }
+
+      const finalResult = {
+        approved: subPipelineResult.approved,
+        sessionId: ctx.session.id,
+        huResults: subPipelineResult.results,
+        blockedIds: subPipelineResult.blockedIds,
+        planId: ctx.session._planRef?.planId || null
+      };
+      await writeHistoryRecord({ sessionId: ctx.session.id, task, result: finalResult, logger });
+      return finalResult;
+    }
+
+    // --- Standard single-task pipeline (1 HU or no HU reviewer) ---
+    const result = await runIterationLoop(ctx, { task: ctx.plannedTask, askQuestion, emitter, logger });
+    await writeHistoryRecord({ sessionId: ctx.session.id, task, result, logger });
+    return result;
+  } finally {
+    // --- Cleanup auto-installed skills ---
+    const autoSkills = ctx.session?.autoInstalledSkills;
+    if (autoSkills && autoSkills.length > 0) {
+      const cleanProjectDir = ctx.config?.projectDir || process.cwd();
+      try {
+        const cleanupResult = await cleanupAutoInstalledSkills(autoSkills, cleanProjectDir);
+        if (cleanupResult.removed.length > 0) {
+          logger.info(`Cleaned up ${cleanupResult.removed.length} auto-installed skill(s): ${cleanupResult.removed.join(", ")}`);
+        }
+      } catch (err) {
+        logger.warn(`Skill cleanup failed (non-blocking): ${err.message}`);
+      }
+    }
+
+    // --- Telemetry: anonymous pipeline_complete event (non-blocking) ---
+    try {
+      const { sendTelemetryEvent } = await import("../utils/telemetry.js");
+      const durationS = Math.round((Date.now() - ctx.startedAt) / 1000);
+      const sessionStatus = ctx.session?.status || "unknown";
+      sendTelemetryEvent("pipeline_complete", {
+        mode: config.review_mode,
+        agent: ctx.coderRole?.provider || config.coder,
+        duration_s: durationS,
+        success: sessionStatus === "approved",
+        taskType: ctx.session?.resolved_policies?.taskType || null
+      }, config).catch(() => {});
+    } catch { /* non-blocking */ }
+  }
+}
+
+export async function resumeFlow({ sessionId, answer, config, logger, flags = {}, emitter = null, askQuestion = null }) {
+  const session = answer
+    ? await resumeSessionWithAnswer(sessionId, answer)
+    : await loadSession(sessionId);
+
+  if (session.status === "paused" && !answer) {
+    logger.info(`Session ${sessionId} is paused. Provide --answer to resume.`);
+    return session;
+  }
+
+  // Allow resuming "stopped" sessions (checkpoint stop) and "failed" sessions
+  const resumableStatuses = new Set(["running", "stopped", "failed"]);
+  if (!resumableStatuses.has(session.status)) {
+    logger.info(`Session ${sessionId} has status ${session.status} — not resumable`);
+    return session;
+  }
+
+  // Mark as running again for stopped/failed sessions
+  if (session.status !== "running") {
+    logger.info(`Resuming ${session.status} session ${sessionId}`);
+    session.status = "running";
+    await saveSession(session);
+  }
+
+  // Session was paused and now resumed with answer - re-run the flow
+  const task = session.task;
+  const sessionConfig = config || session.config_snapshot;
+  if (!sessionConfig) {
+    throw new Error("No config available to resume session");
+  }
+
+  logger.info(`Resuming session ${sessionId} with answer: ${answer}`);
+
+  // Inject the answer as additional feedback for the coder
+  if (session.paused_state?.context?.lastFeedback) {
+    session.last_reviewer_feedback = `Previous feedback: ${session.paused_state.context.lastFeedback}\nUser guidance: ${answer}`;
+  }
+  session.repeated_issue_count = 0;
+  session.sonar_retry_count = 0;
+  session.reviewer_retry_count = 0;
+  session.standby_retry_count = 0;
+  session.tester_retry_count = 0;
+  session.security_retry_count = 0;
+  session.last_sonar_issue_signature = null;
+  session.sonar_repeat_count = 0;
+  session.last_reviewer_issue_signature = null;
+  session.reviewer_repeat_count = 0;
+  await saveSession(session);
+
+  // Re-run the flow with the existing session context
+  try {
+    return await runFlow({ task, config: sessionConfig, logger, flags, emitter, askQuestion });
+  } catch (err) {
+    await markSessionStatus(session, "failed");
+    throw err;
+  }
+}