openpond 0.0.51 → 0.0.52

package/dist/chunks/{app-layer-DCUTQ4TM.js → app-layer-T26MAAGI.js}

@@ -5,9 +5,9 @@ import {
 import "./chunk-KHNHUA2H.js";
 import {
   CliUsageError
-} from "./chunk-PFNX5XOT.js";
+} from "./chunk-6QWTEC2D.js";
 import "./chunk-SMNFI2FU.js";
-import "./chunk-R4PRXXTC.js";
+import "./chunk-JSRCXDIL.js";
 import "./chunk-DBHSOTYZ.js";
 import "./chunk-KSRWKXRB.js";
 import "./chunk-MUMPWI4O.js";

package/dist/chunks/{app-server-runtime-Q3IGU5YB.js → app-server-runtime-2P2727X6.js}

@@ -2,13 +2,13 @@ import { createRequire as __openpondCreateRequire } from "node:module"; var requ
 import {
   APP_SERVER_COMPOSITION,
   createOpenPondAppServer
-} from "./chunk-UDTHREHB.js";
+} from "./chunk-KNKZSVGO.js";
 import "./chunk-NFNEENNR.js";
-import "./chunk-IGSEFSMA.js";
-import "./chunk-V3HADQL5.js";
-import "./chunk-J4LLKY6E.js";
-import "./chunk-3PYU3XCS.js";
-import "./chunk-35OYRVVH.js";
+import "./chunk-EIRW6Z6S.js";
+import "./chunk-FFD2YSZR.js";
+import "./chunk-C4DAHPVI.js";
+import "./chunk-55KT7K5P.js";
+import "./chunk-US7DOPAH.js";
 import "./chunk-WYDVQOBL.js";
 import "./chunk-KHNHUA2H.js";
 import "./chunk-UACZUFIA.js";

package/dist/chunks/{apps-JDRCDAPE.js → apps-7KLTYKHN.js}

@@ -12,12 +12,12 @@ import {
   resolveAppTarget
 } from "./chunk-FCVVOMIP.js";
 import "./chunk-KHNHUA2H.js";
-import "./chunk-PFNX5XOT.js";
+import "./chunk-6QWTEC2D.js";
 import {
   resolveBaseUrl,
   resolvePublicApiBaseUrl
 } from "./chunk-SMNFI2FU.js";
-import "./chunk-R4PRXXTC.js";
+import "./chunk-JSRCXDIL.js";
 import {
   createAgentFromPrompt,
   deployApp,

package/dist/chunks/{chunk-LYEXNSCO.js → chunk-3UPQUD6Y.js}

@@ -5,12 +5,12 @@ import {
 import {
   getBundledRuntimeVersion,
   openUrlWithSystemBrowser
-} from "./chunk-IGSEFSMA.js";
+} from "./chunk-EIRW6Z6S.js";
 import {
   DEFAULT_HOST,
   DEFAULT_PORT,
   parseListen
-} from "./chunk-J4LLKY6E.js";
+} from "./chunk-C4DAHPVI.js";
 
 // ../server/src/cli.ts
 import { existsSync } from "node:fs";

package/dist/chunks/{chunk-3PYU3XCS.js → chunk-55KT7K5P.js}

@@ -14,7 +14,7 @@ import {
   ToolDeclarationSchema,
   createHarnessAdvanceReceipt,
   validateConnectedAppToolCallRequest
-} from "./chunk-35OYRVVH.js";
+} from "./chunk-US7DOPAH.js";
 import {
   external_exports,
   yaml
@@ -12965,7 +12965,6 @@ export {
   ApproveDatasetImportMappingRequestSchema,
   ModelVersionSchema,
   ModelEvaluationReceiptSchema,
-  ModelEvaluationStopReceiptSchema,
   ModelRunSchema,
   summarizeModelEvaluationTaskEfficiency,
   LocalModelChatConfigurationSchema,

package/dist/chunks/{chunk-PFNX5XOT.js → chunk-6QWTEC2D.js}

@@ -4,7 +4,7 @@ import {
   getAnyCliOptionKind,
   getCliCommandDefinition,
   getCliOptionKind
-} from "./chunk-R4PRXXTC.js";
+} from "./chunk-JSRCXDIL.js";
 import {
   parseJsonOption
 } from "./chunk-MUMPWI4O.js";

package/dist/chunks/{chunk-J4LLKY6E.js → chunk-C4DAHPVI.js}

@@ -3,7 +3,7 @@ import { createRequire as __openpondCreateRequire } from "node:module"; var requ
 // ../server/src/constants.ts
 var DEFAULT_HOST = "127.0.0.1";
 var DEFAULT_PORT = 17874;
-var VERSION = "0.0.51";
+var VERSION = "0.0.52";
 var HOSTED_CHAT_SYSTEM_PROMPT = "You are OpenPond Chat. Respond in the user's language. If the user's latest message is language-neutral, ambiguous, or only a short test, respond in English. Be concise and directly answer the latest request. Do not use emojis. Use Markdown when it improves scanability. If you emit reasoning or thinking content, keep it sparse and user-readable: at most one short sentence for major progress, decisions, or blockers. Omit reasoning for routine searches, reads, tool calls, and obvious next steps. Do not restate the user request, narrate every action, or include code blocks, code excerpts, diffs, raw tool payloads, or markdown examples in reasoning. Put necessary code or exact snippets only in the final assistant answer. When the user asks to show or preview an image that is available as a workspace path or signed image URL, use Markdown image syntax like ![description](path-or-url) instead of a bare path or raw HTML. For workspace-backed answers, use a Codex-style shape: brief outcome first, then Changed Files if files changed, then Verification if checks ran. End with a concise final answer that summarizes the result and verification when relevant. Do not mention raw tool JSON, internal repo paths, or origin/remote URLs unless the user explicitly asks for them or they are necessary to explain a git/deploy failure.";
 var APP_PREFERENCES_CACHE_TYPE = "app_preferences";
 var APP_PREFERENCES_CACHE_KEY = "global";

package/dist/chunks/{chunk-IGSEFSMA.js → chunk-EIRW6Z6S.js}

@@ -12,7 +12,7 @@ import {
   saveConfig,
   saveProfileApiKey,
   setActiveProfile
-} from "./chunk-3PYU3XCS.js";
+} from "./chunk-55KT7K5P.js";
 import {
   DEFAULT_OPENPOND_API_BASE_URL,
   DEFAULT_OPENPOND_WEB_BASE_URL,

package/dist/chunks/{chunk-V3HADQL5.js → chunk-FFD2YSZR.js}

@@ -1,7 +1,7 @@
 import { createRequire as __openpondCreateRequire } from "node:module"; var require = __openpondCreateRequire(import.meta.url);
 import {
   now
-} from "./chunk-J4LLKY6E.js";
+} from "./chunk-C4DAHPVI.js";
 
 // ../server/src/workspace/workspace-command.ts
 import { spawn, spawnSync } from "node:child_process";

package/dist/chunks/{chunk-R4PRXXTC.js → chunk-JSRCXDIL.js}

@@ -397,7 +397,7 @@ var CLI_COMMAND_REGISTRY = [
     name: "help",
     usage: "openpond help",
     optionSchema: {},
-    handler: async () => (await import("./help-WJBMX37U.js")).printHelp()
+    handler: async () => (await import("./help-DHGNXUA4.js")).printHelp()
   },
   {
     name: "version",
@@ -409,19 +409,19 @@ var CLI_COMMAND_REGISTRY = [
     name: "check-update",
     usage: "openpond check-update",
     optionSchema: {},
-    handler: async () => (await import("./core-commands-5BRR7TU4.js")).runCheckUpdate()
+    handler: async () => (await import("./core-commands-SND555SC.js")).runCheckUpdate()
   },
   {
     name: "login",
     usage: "openpond login [--base-url <url>]",
     optionSchema: CLI_GLOBAL_OPTION_SCHEMA,
-    handler: async ({ options }) => (await import("./core-commands-5BRR7TU4.js")).runLogin(options)
+    handler: async ({ options }) => (await import("./core-commands-SND555SC.js")).runLogin(options)
   },
   {
     name: "init",
     usage: "openpond init [--path <dir>]",
     optionSchema: { path: "string" },
-    handler: async ({ options }) => (await import("./profile-B62ZWPTQ.js")).runOpenPondInitCommand(options)
+    handler: async ({ options }) => (await import("./profile-PLWXQHKF.js")).runOpenPondInitCommand(options)
   },
   {
     name: "profile",
@@ -449,13 +449,13 @@ var CLI_COMMAND_REGISTRY = [
       publishHostedSource: "boolean",
       teamId: "string"
     },
-    handler: async ({ options, rest }) => (await import("./profile-B62ZWPTQ.js")).runOpenPondProfileCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./profile-PLWXQHKF.js")).runOpenPondProfileCommand(options, rest)
   },
   {
     name: "agents",
     usage: "openpond agents [--json]",
     optionSchema: { json: "boolean" },
-    handler: async ({ options, rest }) => (await import("./profile-B62ZWPTQ.js")).runOpenPondAgentsCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./profile-PLWXQHKF.js")).runOpenPondAgentsCommand(options, rest)
   },
   ...profileSdkAliasDefinitions(),
   {
@@ -468,25 +468,25 @@ var CLI_COMMAND_REGISTRY = [
       "openpond extension remove <owner/repo> [--json]"
     ],
     optionSchema: { all: "boolean", json: "boolean", ref: "string" },
-    handler: async ({ options, rest }) => (await import("./extension-V3NZ7MU7.js")).runOpenPondExtensionCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./extension-3IKF6YQE.js")).runOpenPondExtensionCommand(options, rest)
   },
   {
     name: "profiles",
     usage: "openpond profiles [args]",
     optionSchema: { json: "boolean" },
-    handler: async ({ options, rest }) => (await import("./core-commands-5BRR7TU4.js")).runProfiles(options, rest)
+    handler: async ({ options, rest }) => (await import("./core-commands-SND555SC.js")).runProfiles(options, rest)
   },
   {
     name: "account",
     usage: "openpond account",
     optionSchema: { json: "boolean" },
-    handler: async () => (await import("./core-commands-5BRR7TU4.js")).runAccount()
+    handler: async () => (await import("./core-commands-SND555SC.js")).runAccount()
   },
   {
     name: "health",
     usage: "openpond health",
     optionSchema: { json: "boolean" },
-    handler: async ({ options }) => (await import("./core-commands-5BRR7TU4.js")).runHealth(options)
+    handler: async ({ options }) => (await import("./core-commands-SND555SC.js")).runHealth(options)
   },
   {
     name: "app-server",
@@ -494,7 +494,7 @@ var CLI_COMMAND_REGISTRY = [
     optionSchema: {
       storeDir: "string"
     },
-    handler: async ({ options, rest }) => (await import("./app-layer-DCUTQ4TM.js")).runOpenPondServerCommand("app-server", options, rest)
+    handler: async ({ options, rest }) => (await import("./app-layer-T26MAAGI.js")).runOpenPondServerCommand("app-server", options, rest)
   },
   {
     name: "serve",
@@ -505,7 +505,7 @@ var CLI_COMMAND_REGISTRY = [
       listen: "string",
       port: "integer"
     },
-    handler: async ({ options, rest }) => (await import("./app-layer-DCUTQ4TM.js")).runOpenPondServerCommand("serve", options, rest)
+    handler: async ({ options, rest }) => (await import("./app-layer-T26MAAGI.js")).runOpenPondServerCommand("serve", options, rest)
   },
   {
     name: "ui",
@@ -518,7 +518,7 @@ var CLI_COMMAND_REGISTRY = [
       port: "integer",
       webRoot: "string"
     },
-    handler: async ({ options, rest }) => (await import("./app-layer-DCUTQ4TM.js")).runOpenPondServerCommand("web", options, rest)
+    handler: async ({ options, rest }) => (await import("./app-layer-T26MAAGI.js")).runOpenPondServerCommand("web", options, rest)
   },
   {
     name: "tui",
@@ -532,13 +532,13 @@ var CLI_COMMAND_REGISTRY = [
       provider: "string",
       server: "string"
     },
-    handler: async ({ options, rest }) => (await import("./app-layer-DCUTQ4TM.js")).runOpenPondTerminalCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./app-layer-T26MAAGI.js")).runOpenPondTerminalCommand(options, rest)
   },
   {
     name: "chat",
     usage: "openpond chat (--message-file <path>|--message <text>|--stdin) --non-interactive [--provider <id>] [--model <id>] [--agent <id>] [--project <id>] [--resume <session-id>] [--yes] [--approval-policy <policy>] [--json] [--timeout-sec <n>] [--max-output-bytes <n>] [--sandbox <mode>]",
     optionSchema: CHAT_OPTION_SCHEMA,
-    handler: async ({ options, rest }) => (await import("./app-layer-DCUTQ4TM.js")).runOpenPondTerminalCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./app-layer-T26MAAGI.js")).runOpenPondTerminalCommand(options, rest)
   },
   {
     name: "tool",
@@ -574,7 +574,7 @@ var CLI_COMMAND_REGISTRY = [
     name: "sandbox-template",
     usage: "openpond sandbox-template <command> [args]",
     optionSchema: SANDBOX_TEMPLATE_OPTION_SCHEMA,
-    handler: async ({ options, rest }) => (await import("./sandbox-template-GEWTAWAR.js")).runSandboxTemplateCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./sandbox-template-FVW2UTI4.js")).runSandboxTemplateCommand(options, rest)
   },
   {
     name: "repo",
@@ -619,13 +619,13 @@ var CLI_COMMAND_REGISTRY = [
       toolset: "string",
       url: "string"
     },
-    handler: async ({ options, rest }) => (await import("./organizations-RHOF5H4Z.js")).runOrganizationsCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./organizations-HKVC4CSA.js")).runOrganizationsCommand(options, rest)
   },
   {
     name: "project",
     usage: "openpond project <list|create|upsert|get|update|sync|source-upload|archive> [args]",
     optionSchema: PROJECT_OPTION_SCHEMA,
-    handler: async ({ options, rest }) => (await import("./project-agent-QA25HHDV.js")).runProjectCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./project-agent-JNSP56KG.js")).runProjectCommand(options, rest)
   },
   {
     name: "agent",
@@ -637,7 +637,7 @@ var CLI_COMMAND_REGISTRY = [
       "openpond agent source publish <agentId> --team-id <id> [--expected-manifest-hash <hash>]"
     ],
     optionSchema: AGENT_OPTION_SCHEMA,
-    handler: async ({ options, rest }) => (await import("./project-agent-QA25HHDV.js")).runAgentCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./project-agent-JNSP56KG.js")).runAgentCommand(options, rest)
   },
   {
     name: "desktop-test",
@@ -658,13 +658,13 @@ var CLI_COMMAND_REGISTRY = [
       token: "string",
       tokenFile: "string"
     },
-    handler: async ({ options, rest }) => (await import("./desktop-test-EQ2BGZOM.js")).runDesktopTestCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./desktop-test-WSWXNSFS.js")).runDesktopTestCommand(options, rest)
   },
   {
     name: "sandbox",
     usage: "openpond sandbox <command> [args]",
     optionSchema: SANDBOX_OPTION_SCHEMA,
-    handler: async ({ options, rest }) => (await import("./sandbox-command-JFEVPU4X.js")).runSandboxCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./sandbox-command-Z6LNNSSL.js")).runSandboxCommand(options, rest)
   },
   {
     name: "training",
@@ -712,7 +712,7 @@ var CLI_COMMAND_REGISTRY = [
       skipTool: "boolean",
       stream: "boolean"
     },
-    handler: async ({ options, rest }) => (await import("./opchat-W7DSKNNB.js")).runOpChatCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./opchat-XYKT5KKO.js")).runOpChatCommand(options, rest)
   },
   {
     name: "teams-bot",
@@ -721,7 +721,7 @@ var CLI_COMMAND_REGISTRY = [
       json: "boolean",
       teamId: "string"
     },
-    handler: async ({ options, rest }) => (await import("./teams-bot-ODSMRNE4.js")).runTeamsBotCommand(options, rest)
+    handler: async ({ options, rest }) => (await import("./teams-bot-FNPT4Q2B.js")).runTeamsBotCommand(options, rest)
   },
   {
     name: "apps",
@@ -742,7 +742,7 @@ var CLI_COMMAND_REGISTRY = [
     name: "opentool",
     usage: "openpond opentool <init|validate|build> [args]",
     optionSchema: {},
-    handler: async ({ rest }) => (await import("./core-commands-5BRR7TU4.js")).runOpentool(rest)
+    handler: async ({ rest }) => (await import("./core-commands-SND555SC.js")).runOpentool(rest)
   }
 ];
 var CLI_COMMAND_BY_NAME = /* @__PURE__ */ new Map();
@@ -814,11 +814,11 @@ function profileSdkAliasDefinitions() {
     name,
     usage: `openpond ${name} [args]`,
     optionSchema: PROFILE_SDK_OPTION_SCHEMA,
-    handler: async ({ options, rest }) => (await import("./profile-B62ZWPTQ.js")).runOpenPondProfileSdkAlias(name, options, rest)
+    handler: async ({ options, rest }) => (await import("./profile-PLWXQHKF.js")).runOpenPondProfileSdkAlias(name, options, rest)
   }));
 }
 async function runToolCommand({ options, rest }) {
-  const { runToolList, runToolRun } = await import("./core-commands-5BRR7TU4.js");
+  const { runToolList, runToolRun } = await import("./core-commands-SND555SC.js");
   const subcommand = rest[0];
   if (subcommand === "list") {
     const target = rest[1];
@@ -842,8 +842,8 @@ async function runToolCommand({ options, rest }) {
   throw new Error("usage: tool <list|run> <handle>/<repo> [args]");
 }
 async function runBacktestCommand({ options, rest }) {
-  const { runBacktestRun } = await import("./core-commands-5BRR7TU4.js");
-  const { runBacktestEvents, runBacktestGet } = await import("./apps-JDRCDAPE.js");
+  const { runBacktestRun } = await import("./core-commands-SND555SC.js");
+  const { runBacktestEvents, runBacktestGet } = await import("./apps-7KLTYKHN.js");
   const subcommand = rest[0] || "run";
   if (subcommand === "run") {
     const target = rest[1];
@@ -877,7 +877,7 @@ async function runBacktestCommand({ options, rest }) {
   throw new Error("usage: backtest <run|events|get> <handle>/<repo> [args]");
 }
 async function runDeployCommand({ options, rest }) {
-  const { runDeployWatch } = await import("./core-commands-5BRR7TU4.js");
+  const { runDeployWatch } = await import("./core-commands-SND555SC.js");
   const subcommand = rest[0] || "watch";
   if (subcommand !== "watch") {
     throw new Error(
@@ -893,7 +893,7 @@ async function runDeployCommand({ options, rest }) {
   await runDeployWatch(options, target);
 }
 async function runTemplateCommand({ options, rest }) {
-  const { runTemplateBranches, runTemplateStatus, runTemplateUpdate } = await import("./core-commands-5BRR7TU4.js");
+  const { runTemplateBranches, runTemplateStatus, runTemplateUpdate } = await import("./core-commands-SND555SC.js");
   const subcommand = rest[0] || "status";
   const target = rest[1];
   if (!target) {
@@ -918,7 +918,7 @@ async function runTemplateCommand({ options, rest }) {
   );
 }
 async function runRepoCommand({ options, rest }) {
-  const { runRepoCreate, runRepoPush } = await import("./core-commands-5BRR7TU4.js");
+  const { runRepoCreate, runRepoPush } = await import("./core-commands-SND555SC.js");
   const subcommand = rest[0] || "create";
   if (subcommand === "create") {
     await runRepoCreate(options, rest.slice(1));
@@ -945,7 +945,7 @@ async function runAppsCommand({ options, rest }) {
     runAppsSummary,
     runAppsTools,
     runAppsToolsExecute
-  } = await import("./apps-JDRCDAPE.js");
+  } = await import("./apps-7KLTYKHN.js");
   const subcommand = rest[0];
   if (subcommand === "list") {
     await runAppsList(options);

package/dist/chunks/{chunk-UDTHREHB.js → chunk-KNKZSVGO.js}

@@ -21,7 +21,7 @@ import {
   loadOpenPondAccountContext,
   loadOpenPondHostedModels,
   streamOpenPondHostedChatTurn
-} from "./chunk-IGSEFSMA.js";
+} from "./chunk-EIRW6Z6S.js";
 import {
   assertSafeBranchName,
   isGeneratedWorkspacePath,
@@ -34,14 +34,14 @@ import {
   truncatePatch,
   uniqueSortedPaths,
   workspaceImageContentType
-} from "./chunk-V3HADQL5.js";
+} from "./chunk-FFD2YSZR.js";
 import {
   HOSTED_CHAT_SYSTEM_PROMPT,
   VERSION,
   event,
   now,
   textFromUnknown
-} from "./chunk-J4LLKY6E.js";
+} from "./chunk-C4DAHPVI.js";
 import {
   AppPreferencesSchema,
   ApplyCreateImproveRunActionRequestSchema,
@@ -158,7 +158,7 @@ import {
   runAgentSdkProjectCommand,
   runProfileSkillCommand,
   sidebarFileBookmarkId
-} from "./chunk-3PYU3XCS.js";
+} from "./chunk-55KT7K5P.js";
 import {
   AgentSnapshotSchema,
   AttemptReceiptSchema,
@@ -236,7 +236,7 @@ import {
   stableId,
   uniqueEventRefs,
   validateConnectedAppProviderOperationRequest
-} from "./chunk-35OYRVVH.js";
+} from "./chunk-US7DOPAH.js";
 import {
   external_exports
 } from "./chunk-WYDVQOBL.js";
@@ -7869,11 +7869,12 @@ var MAX_PDF_ARTIFACTS = 10;
 var MAX_PDF_BYTES = 20 * 1024 * 1024;
 var execFileAsync = promisify(execFile);
 async function loadBoundedRefinerContext(store, trigger, observations) {
-  const [session, turn, events, turns] = await Promise.all([
+  const [session, turn, events, turns, usageRecords] = await Promise.all([
     store.getSession(trigger.runRef),
     store.getTurn(trigger.turnId),
     store.runtimeEventsForSession(trigger.runRef, { limit: 1e3 }),
-    store.turnsForSession(trigger.runRef, 1e3)
+    store.turnsForSession(trigger.runRef, 1e3),
+    store.listModelUsageRecords({ turnId: trigger.turnId, limit: 1e4 })
   ]);
   const orderedTurns = [...turns].sort(
     (left, right) => left.startedAt.localeCompare(right.startedAt)
@@ -7900,6 +7901,23 @@ async function loadBoundedRefinerContext(store, trigger, observations) {
     limit: trigger.policy.maxEvidenceEvents
   });
   const assistantOutput = assistantOutputForTurn(events, trigger.turnId);
+  const observationKinds = observations.map((observation) => observation.kind);
+  const receiptProfile = tasksetGradeExecutionProfile(events, trigger.turnId);
+  const durableUsage = {
+    modelRequestCount: usageRecords.length || receiptProfile.modelRequestCount,
+    promptTokens: usageRecords.reduce(
+      (total, record9) => total + (record9.promptTokens ?? 0),
+      0
+    ) || receiptProfile.promptTokens,
+    completionTokens: usageRecords.reduce(
+      (total, record9) => total + (record9.completionTokens ?? 0),
+      0
+    ) || receiptProfile.completionTokens,
+    totalTokens: usageRecords.reduce(
+      (total, record9) => total + (record9.totalTokens ?? 0),
+      0
+    ) || receiptProfile.totalTokens
+  };
   return {
     task: {
       prompt: turn?.prompt ? redactAndBoundRefinerText(turn.prompt, 8e3) : null,
@@ -7908,6 +7926,18 @@ async function loadBoundedRefinerContext(store, trigger, observations) {
       previousAssistantOutput: previousTurn ? assistantOutputForTurn(events, previousTurn.id) : null
     },
     artifactDiagnostics: await inspectBoundedPdfArtifactDiagnostics(session?.cwd),
+    executionProfile: {
+      modelRequestCount: durableUsage.modelRequestCount,
+      failedModelRequestCount: usageRecords.filter(
+        (record9) => record9.status === "failed" || record9.status === "interrupted"
+      ).length,
+      promptTokens: durableUsage.promptTokens,
+      completionTokens: durableUsage.completionTokens,
+      totalTokens: durableUsage.totalTokens,
+      toolFailureCount: observationKinds.filter((kind) => kind === "tool_failure").length,
+      retryCount: observationKinds.filter((kind) => kind === "retry").length,
+      recoveryCount: observationKinds.filter((kind) => kind === "recovery").length
+    },
     eventExcerpts: evidenceEvents.map((runtimeEvent) => {
       const data = asRecord(runtimeEvent.data);
       const result = asRecord(data.result);
@@ -7927,6 +7957,34 @@ async function loadBoundedRefinerContext(store, trigger, observations) {
     })
   };
 }
+function tasksetGradeExecutionProfile(events, turnId) {
+  const fallback = {
+    modelRequestCount: 0,
+    promptTokens: 0,
+    completionTokens: 0,
+    totalTokens: 0
+  };
+  const diagnostic = [...events].reverse().find(
+    (runtimeEvent) => runtimeEvent.turnId === turnId && runtimeEvent.name === "diagnostic" && runtimeEvent.action === "taskset_grade" && typeof runtimeEvent.output === "string"
+  );
+  if (!diagnostic?.output) return fallback;
+  try {
+    const payload = asRecord(JSON.parse(diagnostic.output));
+    const attempt = asRecord(payload.attempt);
+    const usage = asRecord(attempt.usage);
+    return {
+      modelRequestCount: nonnegativeInteger(attempt.modelRequestCount),
+      promptTokens: nonnegativeInteger(usage.promptTokens),
+      completionTokens: nonnegativeInteger(usage.completionTokens),
+      totalTokens: nonnegativeInteger(usage.totalTokens)
+    };
+  } catch {
+    return fallback;
+  }
+}
+function nonnegativeInteger(value) {
+  return typeof value === "number" && Number.isFinite(value) && value >= 0 ? Math.trunc(value) : 0;
+}
 async function inspectBoundedPdfArtifactDiagnostics(cwd) {
   if (!cwd) return [];
   const pdfs = [];
@@ -11022,6 +11080,19 @@ async function runLocalHarnessRefinerWorker(input) {
     trigger,
     observations
   );
+  const currentObservationHashes = new Set(
+    observations.map((observation) => observation.contentHash)
+  );
+  const recentObservations = (await input.store.listHarnessImprovementArtifacts(
+    workspace.id,
+    "observation",
+    100
+  )).filter((observation) => !currentObservationHashes.has(observation.contentHash)).sort((left, right) => right.createdAt.localeCompare(left.createdAt)).slice(0, 20).map((observation) => ({
+    ...boundedObservationEvidence(observation),
+    runRef: observation.runRef,
+    turnId: observation.turnId,
+    createdAt: observation.createdAt
+  }));
   const recentOutcomes = (await input.store.listHarnessImprovementArtifacts(
     workspace.id,
     "refiner_outcome",
@@ -11039,6 +11110,8 @@ async function runLocalHarnessRefinerWorker(input) {
     task: boundedContext.task,
     eventExcerpts: boundedContext.eventExcerpts,
     artifactDiagnostics: boundedContext.artifactDiagnostics,
+    executionProfile: boundedContext.executionProfile,
+    recentObservations,
     recentOutcomes,
     sourceFiles: source.files,
     sourceCatalog: source.catalog,
@@ -41835,7 +41908,7 @@ function uniqueEvidence(run, evidence) {
 }
 async function verifyAppliedProfileCommit(input) {
   const command = input.command ?? (async (name, args, cwd, env) => {
-    const { runWorkspaceCommand } = await import("./workspaces-7YHBJG5P.js");
+    const { runWorkspaceCommand } = await import("./workspaces-YLNB6DS7.js");
     return runWorkspaceCommand(name, args, cwd, env);
   });
   const result = await command("git", ["rev-parse", "HEAD"], input.repoPath);

package/dist/chunks/{chunk-CSN3DLWJ.js → chunk-M2OSF2YN.js}

@@ -28,7 +28,7 @@ import {
   compareSemver,
   fetchLatestNpmVersion,
   getInstalledCliVersion
-} from "./chunk-R4PRXXTC.js";
+} from "./chunk-JSRCXDIL.js";
 import {
   checkOpenPondApiHealth,
   createHeadlessApps,

package/dist/chunks/{chunk-35OYRVVH.js → chunk-US7DOPAH.js}

@@ -2770,6 +2770,17 @@ var LocalHarnessRefinerEvidenceSchema = external_exports.object({
   }).strict(),
   eventExcerpts: external_exports.array(external_exports.record(external_exports.string(), external_exports.unknown())).max(20),
   artifactDiagnostics: external_exports.array(external_exports.record(external_exports.string(), external_exports.unknown())).max(20),
+  executionProfile: external_exports.object({
+    modelRequestCount: external_exports.number().int().nonnegative(),
+    failedModelRequestCount: external_exports.number().int().nonnegative(),
+    promptTokens: external_exports.number().int().nonnegative(),
+    completionTokens: external_exports.number().int().nonnegative(),
+    totalTokens: external_exports.number().int().nonnegative(),
+    toolFailureCount: external_exports.number().int().nonnegative(),
+    retryCount: external_exports.number().int().nonnegative(),
+    recoveryCount: external_exports.number().int().nonnegative()
+  }).strict(),
+  recentObservations: external_exports.array(external_exports.record(external_exports.string(), external_exports.unknown())).max(20),
   recentOutcomes: external_exports.array(external_exports.object({
     id: external_exports.string().trim().min(1).max(2e3),
     decision: external_exports.enum(["no_action", "proposed"]),
@@ -2882,7 +2893,10 @@ function refinerMessages(evidence2) {
         "The task's assistantOutputLinkCount and artifactDiagnostics are objective observations, not decision rules. Failed artifact diagnostics can contradict a claimed successful visual check; decide whether the evidence supports a reusable Harness correction, an external route, or no action. When a user requests linked evidence, named sources without clickable links do not satisfy the request; an explicit request for links authorizes including them and must not be excused as a generic URL-formatting constraint.",
         "For claims presented as current web verification, consider whether user-visible citations let the user inspect the supporting evidence even when the request did not literally say 'include links'. Source names and hidden retrieval metadata alone do not make a current factual claim verifiable.",
         "A recovered error can still justify improvement when the same avoidable first attempt is likely to recur. Ordinary successful work, one-off artifact details, and continuation of the current task usually require no_action.",
+        "executionProfile is bounded cost evidence for the completed turn. Use request, token, tool-failure, retry, and recovery counts to distinguish a cheap recovery from a material recurring tax. High cost alone is not a reason to edit the Harness, but repeated repair loops supported by recentObservations can justify removing the failed first strategy for future related work.",
+        "recentObservations is a bounded window of earlier raw improvement observations from this Harness workspace. Use it to detect recurrence across distinct real turns even when an earlier Refiner outcome was no_action. Match the reusable root behavior, not merely a shared tool name, topic, artifact type, or benchmark family.",
         "recentOutcomes is a small bounded window of earlier Refiner decisions from this Harness workspace. Use it as recurrence evidence only when you judge the underlying behavior to be related; repeated no_action decisions do not force a proposal, and differently worded incidents may still share one root behavior.",
+        "Optimize future related work, not the already completed turn. Prefer a small instruction or workflow correction that removes the repeated failed attempt, redundant search, full rewrite, or unnecessary output while preserving the requested result. Do not prescribe a library, command, file format, or subject-specific workaround unless the durable Harness already standardizes that workflow.",
         "Propose only the reusable root behavior. Do not encode the task's subject, named entities, business facts, requested artifact outline, benchmark wording, or transient paths. A durable proposal must plausibly help materially different future tasks with the same failure class; otherwise choose no_action or route the underlying runtime/product concern.",
         "Choose the smallest correct layer. Use memory for durable user facts or preferences, prompt for broad behavior, skill for a reusable workflow, and agent for a reusable role. Use route for runtime, product, taskset, or training concerns that this step must not mutate.",
         "Do not confuse 'no safe Harness edit' with no_action. If the evidence exposes a durable defect owned by runtime, product, evaluation, or training, return route even when the agent recovered and completed the task.",
@@ -3823,7 +3837,7 @@ var harnessRefinerBenchmarkRelease = TasksetReleaseSchema.parse({
       ]
     }
   ],
-  "contentHash": "05dbf9058c09b75bc950ca5cdf4ac03650dba7bb0cb74a3d9e152b16c110a2f7",
+  "contentHash": "4cef91a9c92df39d16f741b4d901dbde6b62e72bd8a48647a4a81c0d517d9634",
   "environment": {
     "defaultTimeoutMs": 9e5,
     "deterministicSeeds": false,
@@ -3879,14 +3893,15 @@ var harnessRefinerBenchmarkRelease = TasksetReleaseSchema.parse({
       "weight": 1
     }
   ],
-  "id": "harness-refiner-public-v1",
+  "id": "harness-refiner-08112026",
   "metadata": {
     "adaptationSplit": "validation",
     "benchmark": "harness-refiner",
     "frozenEvaluationSplit": "frozen_eval",
     "primaryMetric": "paired_foreground_provider_tokens",
-    "protocolVersion": "1",
+    "protocolVersion": "2",
     "qualityPolicy": "hard_non_regression",
+    "refinementMode": "sequential_product_lifecycle",
     "toolDeclarationSource": "openpond-production-model-tool-definitions",
     "trainingSideEffect": false
   },
@@ -3903,7 +3918,7 @@ var harnessRefinerBenchmarkRelease = TasksetReleaseSchema.parse({
       "expectedOutput"
     ]
   },
-  "revision": 3,
+  "revision": 1,
   "schemaVersion": "openpond.tasksetRelease.v2",
   "tasks": [
     {

package/dist/chunks/{cli-2FTM4SXH.js → cli-Z2CULBCU.js}

@@ -3,12 +3,12 @@ import {
   resolveWebLaunchMessages,
   runOpenPondAppServerCli,
   runOpenPondServerCli
-} from "./chunk-LYEXNSCO.js";
+} from "./chunk-3UPQUD6Y.js";
 import "./chunk-NFNEENNR.js";
-import "./chunk-IGSEFSMA.js";
-import "./chunk-J4LLKY6E.js";
-import "./chunk-3PYU3XCS.js";
-import "./chunk-35OYRVVH.js";
+import "./chunk-EIRW6Z6S.js";
+import "./chunk-C4DAHPVI.js";
+import "./chunk-55KT7K5P.js";
+import "./chunk-US7DOPAH.js";
 import "./chunk-WYDVQOBL.js";
 import "./chunk-KHNHUA2H.js";
 import "./chunk-UACZUFIA.js";