@chllming/wave-orchestration 0.6.3 → 0.7.1

package/scripts/wave-orchestrator/evals.mjs

@@ -351,6 +351,12 @@ export function loadBenchmarkCatalog(options = {}) {
         id: benchmarkId,
         title: cleanText(benchmark.title) || benchmarkId,
         summary: cleanText(benchmark.summary) || null,
+        localCases: normalizeStringArray(
+          benchmark.localCases,
+          `families.${familyId}.benchmarks.${benchmarkId}.localCases`,
+        ).map((entry, index) =>
+          normalizeEvalTargetId(entry, `families.${familyId}.benchmarks.${benchmarkId}.localCases[${index}]`),
+        ),
         goal: cleanText(benchmark.goal) || null,
         failureModes: normalizeStringArray(
           benchmark.failureModes,
@@ -380,6 +386,12 @@ export function loadBenchmarkCatalog(options = {}) {
       id: familyId,
       title: cleanText(rawFamily.title) || familyId,
       summary: cleanText(rawFamily.summary) || null,
+      localCases: normalizeStringArray(
+        rawFamily.localCases,
+        `families.${familyId}.localCases`,
+      ).map((entry, index) =>
+        normalizeEvalTargetId(entry, `families.${familyId}.localCases[${index}]`),
+      ),
       category: cleanText(rawFamily.category) || null,
       coordinationModel: cleanText(rawFamily.coordinationModel) || null,
       primaryMetric: familyPrimaryMetric,
@@ -395,6 +407,17 @@ export function loadBenchmarkCatalog(options = {}) {
     absolutePath,
     families,
     benchmarkIndex,
+    localCaseIndex: Object.fromEntries(
+      Object.values(families).flatMap((family) => [
+        ...(family.localCases || []).map((caseId) => [caseId, { familyId: family.id, benchmarkId: null }]),
+        ...Object.values(family.benchmarks).flatMap((benchmark) =>
+          (benchmark.localCases || []).map((caseId) => [
+            caseId,
+            { familyId: family.id, benchmarkId: benchmark.id },
+          ]),
+        ),
+      ]),
+    ),
   };
 }
 

package/scripts/wave-orchestrator/executors.mjs

@@ -210,7 +210,8 @@ function buildClaudeLaunchSpec({ agent, promptPath, logPath, overlayDir }) {
     systemPromptPath,
     `${renderHarnessSystemPrompt(agent, "claude")}${skillText ? `\n\n${skillText}` : ""}\n`,
   );
-  const tokens = [executor.claude.command, "-p", "--no-session-persistence"];
+  const command = executor?.claude?.command || "claude";
+  const tokens = [command, "-p", "--no-session-persistence"];
   const settingsPath = buildClaudeSettingsPath(executor, overlayDir);
   appendSingleValueFlag(tokens, "--output-format", executor.claude.outputFormat || "text");
   appendSingleValueFlag(tokens, "--model", executor.claude.model || executor.model);
@@ -234,7 +235,7 @@ function buildClaudeLaunchSpec({ agent, promptPath, logPath, overlayDir }) {
   );
   return {
     executorId: "claude",
-    command: executor.claude.command,
+    command,
     useRateLimitRetries: true,
     invocationLines: [
       `task_prompt=$(cat ${shellQuote(promptPath)})`,

package/scripts/wave-orchestrator/feedback.mjs

@@ -21,6 +21,7 @@ import {
   truncate,
   writeJsonAtomic,
 } from "./shared.mjs";
+import { safeQueueWaveControlEvent } from "./wave-control-client.mjs";
 
 function sanitizeToken(value) {
   const token = String(value || "")
@@ -59,6 +60,7 @@ export function createFeedbackRequest({
   orchestratorId,
   question,
   context,
+  recordTelemetry = false,
 }) {
   ensureDirectory(feedbackRequestsDir);
   const requestId = buildRequestId({ lane, wave, agentId });
@@ -83,6 +85,31 @@ export function createFeedbackRequest({
     }
     writeJsonAtomic(filePath, payload);
   });
+  if (recordTelemetry) {
+    try {
+      const lanePaths = buildLanePaths(lane);
+      safeQueueWaveControlEvent(lanePaths, {
+        category: "feedback",
+        entityType: "human_input",
+        entityId: requestId,
+        action: "requested",
+        source: "orchestrator",
+        actor: orchestratorId || "orchestrator",
+        recordedAt: now,
+        identity: {
+          lane,
+          wave,
+          agentId,
+          runKind: lanePaths.runKind,
+          runId: lanePaths.runId,
+        },
+        tags: ["feedback", "human-input"],
+        data: payload,
+      });
+    } catch {
+      // Best-effort telemetry only.
+    }
+  }
   return { requestId, filePath, payload };
 }
 
@@ -93,6 +120,7 @@ export function answerFeedbackRequest({
   response,
   operator = "human-operator",
   force = false,
+  recordTelemetry = false,
 }) {
   const lockPath = path.join(feedbackStateDir, "requests.lock");
   let answeredPayload = null;
@@ -117,6 +145,31 @@ export function answerFeedbackRequest({
     };
     writeJsonAtomic(filePath, answeredPayload);
   });
+  if (recordTelemetry) {
+    try {
+      const lanePaths = buildLanePaths(answeredPayload?.lane || DEFAULT_WAVE_LANE);
+      safeQueueWaveControlEvent(lanePaths, {
+        category: "feedback",
+        entityType: "human_input",
+        entityId: answeredPayload.id,
+        action: "answered",
+        source: "operator",
+        actor: operator,
+        recordedAt: answeredPayload?.updatedAt || toIsoTimestamp(),
+        identity: {
+          lane: answeredPayload?.lane || DEFAULT_WAVE_LANE,
+          wave: answeredPayload?.wave ?? null,
+          agentId: answeredPayload?.agentId || null,
+          runKind: lanePaths.runKind,
+          runId: lanePaths.runId,
+        },
+        tags: ["feedback", "human-input", "answered"],
+        data: answeredPayload,
+      });
+    } catch {
+      // Best-effort telemetry only.
+    }
+  }
   return answeredPayload;
 }
 
@@ -292,6 +345,7 @@ export async function runFeedbackCli(argv) {
       orchestratorId: options.orchestratorId,
       question: options.question,
       context: options.context,
+      recordTelemetry: true,
     });
     console.log(`[wave-human-feedback] created ${result.requestId}`);
     console.log(`file: ${path.relative(REPO_ROOT, result.filePath)}`);
@@ -316,6 +370,7 @@ export async function runFeedbackCli(argv) {
       response: options.response,
       operator: options.operator,
       force: options.force,
+      recordTelemetry: true,
     });
     console.log(`[wave-human-feedback] answered ${options.id}`);
     return;

package/scripts/wave-orchestrator/install.mjs

@@ -5,6 +5,9 @@ import {
   applyContext7SelectionsToWave,
   loadContext7BundleIndex,
 } from "./context7.mjs";
+import {
+  PLANNER_CONTEXT7_TEMPLATE_PATHS,
+} from "./planner-context.mjs";
 import { buildLanePaths, ensureDirectory, PACKAGE_ROOT, readJsonOrNull, REPO_ROOT, writeJsonAtomic } from "./shared.mjs";
 import { fetchLatestPackageVersion } from "./package-update-notice.mjs";
 import {
@@ -29,6 +32,7 @@ export const STARTER_TEMPLATE_PATHS = [
   "docs/agents/wave-cont-qa-role.md",
   "docs/agents/wave-cont-eval-role.md",
   "docs/agents/wave-integration-role.md",
+  "docs/agents/wave-planner-role.md",
   "docs/agents/wave-security-role.md",
   "docs/concepts/context7-vs-skills.md",
   "docs/concepts/operating-modes.md",
@@ -36,23 +40,48 @@ export const STARTER_TEMPLATE_PATHS = [
   "docs/concepts/what-is-a-wave.md",
   "docs/context7/bundles.json",
   "docs/evals/benchmark-catalog.json",
+  "docs/evals/external-benchmarks.json",
+  "docs/evals/external-command-config.sample.json",
+  "docs/evals/external-command-config.swe-bench-pro.json",
+  "docs/evals/wave-benchmark-program.md",
+  "docs/evals/pilots/README.md",
+  "docs/evals/pilots/swe-bench-pro-public-pilot.json",
+  "docs/evals/pilots/swe-bench-pro-public-full-wave-review-10.json",
+  "docs/evals/arm-templates/README.md",
+  "docs/evals/arm-templates/single-agent.json",
+  "docs/evals/arm-templates/full-wave.json",
+  "docs/evals/cases/README.md",
+  "docs/evals/cases/wave-hidden-profile-private-evidence.json",
+  "docs/evals/cases/wave-premature-closure-guard.json",
+  "docs/evals/cases/wave-silo-cross-agent-state.json",
+  "docs/evals/cases/wave-blackboard-inbox-targeting.json",
+  "docs/evals/cases/wave-contradiction-conflict.json",
+  "docs/evals/cases/wave-simultaneous-lockstep.json",
+  "docs/evals/cases/wave-expert-routing-preservation.json",
   "docs/guides/planner.md",
   "docs/guides/terminal-surfaces.md",
   "docs/plans/component-cutover-matrix.json",
   "docs/plans/component-cutover-matrix.md",
   "docs/plans/context7-wave-orchestrator.md",
   "docs/plans/current-state.md",
+  "docs/plans/examples/wave-example-live-proof.md",
   "docs/plans/master-plan.md",
   "docs/plans/migration.md",
   "docs/plans/wave-orchestrator.md",
   "docs/plans/waves/wave-0.md",
+  "docs/reference/live-proof-waves.md",
   "docs/reference/repository-guidance.md",
+  "docs/reference/sample-waves.md",
   "docs/reference/skills.md",
+  "docs/reference/wave-planning-lessons.md",
   "docs/reference/runtime-config/README.md",
   "docs/reference/runtime-config/codex.md",
   "docs/reference/runtime-config/claude.md",
   "docs/reference/runtime-config/opencode.md",
+  "docs/research/coordination-failure-review.md",
   "docs/research/agent-context-sources.md",
+  "docs/plans/examples/wave-benchmark-improvement.md",
+  ...PLANNER_CONTEXT7_TEMPLATE_PATHS,
 ];
 const REQUIRED_GITIGNORE_ENTRIES = [
   ".tmp/",
@@ -63,6 +92,33 @@ const REQUIRED_GITIGNORE_ENTRIES = [
   "docs/research/papers/",
   "docs/research/articles/",
 ];
+const PLANNER_MIGRATION_REQUIRED_SURFACES = [
+  {
+    id: "planner-role",
+    label: "docs/agents/wave-planner-role.md",
+    path: "docs/agents/wave-planner-role.md",
+    kind: "file",
+  },
+  {
+    id: "planner-skill",
+    label: "skills/role-planner/",
+    path: "skills/role-planner",
+    kind: "dir",
+  },
+  {
+    id: "planner-context7",
+    label: "docs/context7/planner-agent/",
+    path: "docs/context7/planner-agent",
+    kind: "dir",
+  },
+  {
+    id: "planner-lessons",
+    label: "docs/reference/wave-planning-lessons.md",
+    path: "docs/reference/wave-planning-lessons.md",
+    kind: "file",
+  },
+];
+const PLANNER_REQUIRED_BUNDLE_ID = "planner-agentic";
 
 function collectDeclaredDeployKinds(waves = []) {
   return Array.from(
@@ -193,6 +249,8 @@ function slugifyVersion(value) {
 }
 
 function formatUpgradeReport(report) {
+  const plannerMigrationErrors = report.doctor.errors.filter((issue) => isPlannerMigrationIssue(issue));
+  const otherDoctorErrors = report.doctor.errors.filter((issue) => !isPlannerMigrationIssue(issue));
   return [
     `# Wave Upgrade Report`,
     "",
@@ -209,12 +267,27 @@ function formatUpgradeReport(report) {
     "",
     "- No repo-owned plans, waves, role prompts, or config files were overwritten.",
     "- New runtime behavior comes from the installed package version.",
+    ...(report.initMode === "adopt-existing" && plannerMigrationErrors.length > 0
+      ? [
+          "",
+          "## Adopted Repo Follow-Up",
+          "",
+          "- This workspace was adopted from an existing repo-owned Wave surface.",
+          "- `wave upgrade` does not copy new planner starter docs, skills, or Context7 bundle entries into adopted repos.",
+          ...plannerMigrationErrors.map((issue) => `- Error: ${issue}`),
+        ]
+      : []),
+    ...(report.initMode === "adopt-existing" && plannerMigrationErrors.length > 0
+      ? [
+          "- After syncing that planner surface, rerun `pnpm exec wave doctor` before relying on `wave draft --agentic` or planner-aware validation.",
+        ]
+      : []),
     ...(report.doctor.errors.length > 0 || report.doctor.warnings.length > 0
       ? [
           "",
           "## Follow-Up",
           "",
-          ...report.doctor.errors.map((issue) => `- Error: ${issue}`),
+          ...otherDoctorErrors.map((issue) => `- Error: ${issue}`),
           ...report.doctor.warnings.map((issue) => `- Warning: ${issue}`),
         ]
       : []),
@@ -233,6 +306,61 @@ function gitignoreWarnings() {
   );
 }
 
+function plannerRequiredPaths() {
+  return Array.from(
+    new Set(
+      [
+        "docs/agents/wave-planner-role.md",
+        "docs/reference/wave-planning-lessons.md",
+        "skills/role-planner/SKILL.md",
+        ...PLANNER_CONTEXT7_TEMPLATE_PATHS,
+      ].filter(Boolean),
+    ),
+  ).sort();
+}
+
+function isPlannerMigrationIssue(issue) {
+  return String(issue || "").startsWith("Planner starter surface is incomplete");
+}
+
+function missingPlannerMigrationSurfaceLabels() {
+  const missing = [];
+  for (const surface of PLANNER_MIGRATION_REQUIRED_SURFACES) {
+    const targetPath = path.join(REPO_ROOT, surface.path);
+    if (!fs.existsSync(targetPath)) {
+      missing.push(surface.label);
+      continue;
+    }
+    if (surface.kind === "dir") {
+      try {
+        if (fs.readdirSync(targetPath).length === 0) {
+          missing.push(surface.label);
+        }
+      } catch {
+        missing.push(surface.label);
+      }
+    }
+  }
+  return missing;
+}
+
+function plannerMigrationIssue(config, context7BundleIndex) {
+  const missing = missingPlannerMigrationSurfaceLabels();
+  const bundleId = String(config?.planner?.agentic?.context7Bundle || "").trim();
+  const bundleEntryMissing =
+    bundleId === "" || bundleId === PLANNER_REQUIRED_BUNDLE_ID
+      ? !context7BundleIndex?.bundles?.[PLANNER_REQUIRED_BUNDLE_ID]
+      : false;
+  if (missing.length === 0 && !bundleEntryMissing) {
+    return null;
+  }
+  const remediationItems = missing.slice();
+  if (bundleEntryMissing) {
+    remediationItems.push(`docs/context7/bundles.json#${PLANNER_REQUIRED_BUNDLE_ID}`);
+  }
+  return `Planner starter surface is incomplete for 0.7.x workspaces. Sync ${remediationItems.join(", ")} from the packaged release, then rerun \`pnpm exec wave doctor\`.`;
+}
+
 export function runDoctor() {
   const errors = [];
   const warnings = [];
@@ -277,9 +405,29 @@ export function runDoctor() {
           errors.push(`Missing required Wave file: ${relPath}`);
         }
       }
+      const context7BundleIndex = loadContext7BundleIndex(lanePaths.context7BundleIndexPath);
+      const plannerMigration = plannerMigrationIssue(config, context7BundleIndex);
+      if (plannerMigration) {
+        errors.push(plannerMigration);
+      }
+      const plannerPaths = plannerRequiredPaths();
+      for (const relPath of plannerPaths) {
+        if (!fs.existsSync(path.join(REPO_ROOT, relPath)) && !plannerMigration) {
+          errors.push(`Missing planner file: ${relPath}`);
+        }
+      }
+      const plannerBundleId = String(config.planner?.agentic?.context7Bundle || "").trim();
+      if (
+        plannerBundleId &&
+        !context7BundleIndex.bundles[plannerBundleId] &&
+        !(plannerMigration && plannerBundleId === PLANNER_REQUIRED_BUNDLE_ID)
+      ) {
+        errors.push(
+          `planner.agentic.context7Bundle references unknown bundle "${plannerBundleId}".`,
+        );
+      }
       let parsedWaves = [];
       if (fs.existsSync(lanePaths.wavesDir)) {
-        const context7BundleIndex = loadContext7BundleIndex(lanePaths.context7BundleIndexPath);
         parsedWaves = parseWaveFiles(lanePaths.wavesDir, { laneProfile: lanePaths.laneProfile })
           .map((wave) =>
             applyExecutorSelectionsToWave(wave, {
@@ -430,6 +578,7 @@ export function upgradeWorkspace() {
     previousVersion,
     currentVersion: metadata.version,
     generatedAt,
+    initMode: existingState.initMode || null,
     releases,
     doctor,
   };

package/scripts/wave-orchestrator/launcher-closure.mjs

@@ -201,7 +201,7 @@ export async function runClosureSweepPhase({
           refreshWaveDashboardAgentStates(dashboardState, [runInfo], pendingAgentIds, (event) =>
             recordCombinedEvent(event),
           );
-          monitorWaveHumanFeedbackFn({
+          const feedbackChanged = monitorWaveHumanFeedbackFn({
             lanePaths,
             waveNumber: wave.wave,
             agentRuns: [runInfo],
@@ -211,6 +211,9 @@ export async function runClosureSweepPhase({
             recordCombinedEvent,
             appendCoordination,
           });
+          if (feedbackChanged) {
+            refreshDerivedState?.(dashboardState?.attempt || 0);
+          }
           updateWaveDashboardMessageBoard(dashboardState, runInfo.messageBoardPath);
           flushDashboards();
         },

package/scripts/wave-orchestrator/launcher-runtime.mjs

@@ -69,6 +69,7 @@ export async function launchAgentSession(lanePaths, params, { runTmuxFn }) {
     sharedSummaryText,
     inboxPath,
     inboxText,
+    promptOverride = "",
     orchestratorId,
     agentRateLimitRetries,
     agentRateLimitBaseDelaySeconds,
@@ -101,27 +102,29 @@ export async function launchAgentSession(lanePaths, params, { runTmuxFn }) {
       artifacts: skillArtifacts,
     };
   }
-  const prompt = buildExecutionPrompt({
-    lane: lanePaths.lane,
-    wave,
-    agent,
-    orchestratorId,
-    messageBoardPath,
-    messageBoardSnapshot,
-    sharedSummaryPath,
-    sharedSummaryText,
-    inboxPath,
-    inboxText,
-    context7,
-    componentPromotions: resolvedWaveDefinition.componentPromotions,
-    evalTargets: resolvedWaveDefinition.evalTargets,
-    benchmarkCatalogPath: lanePaths.laneProfile?.paths?.benchmarkCatalogPath,
-    sharedPlanDocs: lanePaths.sharedPlanDocs,
-    contQaAgentId: lanePaths.contQaAgentId,
-    contEvalAgentId: lanePaths.contEvalAgentId,
-    integrationAgentId: lanePaths.integrationAgentId,
-    documentationAgentId: lanePaths.documentationAgentId,
-  });
+  const prompt =
+    String(promptOverride || "").trim() ||
+    buildExecutionPrompt({
+      lane: lanePaths.lane,
+      wave,
+      agent,
+      orchestratorId,
+      messageBoardPath,
+      messageBoardSnapshot,
+      sharedSummaryPath,
+      sharedSummaryText,
+      inboxPath,
+      inboxText,
+      context7,
+      componentPromotions: resolvedWaveDefinition.componentPromotions,
+      evalTargets: resolvedWaveDefinition.evalTargets,
+      benchmarkCatalogPath: lanePaths.laneProfile?.paths?.benchmarkCatalogPath,
+      sharedPlanDocs: lanePaths.sharedPlanDocs,
+      contQaAgentId: lanePaths.contQaAgentId,
+      contEvalAgentId: lanePaths.contEvalAgentId,
+      integrationAgentId: lanePaths.integrationAgentId,
+      documentationAgentId: lanePaths.documentationAgentId,
+    });
   const promptHash = hashAgentPromptFingerprint(agent);
   fs.writeFileSync(promptPath, `${prompt}\n`, "utf8");
   const launchSpec = buildExecutorLaunchSpec({
@@ -132,16 +135,16 @@ export async function launchAgentSession(lanePaths, params, { runTmuxFn }) {
     skillProjection: agent.skillsResolved,
   });
   const resolvedExecutorMode = launchSpec.executorId || agent.executorResolved?.id || "codex";
+  writeJsonAtomic(path.join(overlayDir, "launch-preview.json"), {
+    executorId: resolvedExecutorMode,
+    command: launchSpec.command,
+    env: launchSpec.env || {},
+    useRateLimitRetries: launchSpec.useRateLimitRetries === true,
+    invocationLines: launchSpec.invocationLines,
+    limits: launchSpec.limits || null,
+    skills: summarizeResolvedSkills(agent.skillsResolved),
+  });
   if (dryRun) {
-    writeJsonAtomic(path.join(overlayDir, "launch-preview.json"), {
-      executorId: resolvedExecutorMode,
-      command: launchSpec.command,
-      env: launchSpec.env || {},
-      useRateLimitRetries: launchSpec.useRateLimitRetries === true,
-      invocationLines: launchSpec.invocationLines,
-      limits: launchSpec.limits || null,
-      skills: summarizeResolvedSkills(agent.skillsResolved),
-    });
     return {
       promptHash,
       context7,