@workbench-ai/workbench 0.0.74 → 0.0.75

package/dist/fanout.js

@@ -34,7 +34,7 @@ export async function fanOutSkill(name, options) {
             "--global",
             "--yes",
         ];
-        const child = await runNode(args, scrubAgentRuntimeEnv(process.env), FANOUT_TIMEOUT_MS);
+        const child = await runNode(args, minimalChildEnv(process.env), FANOUT_TIMEOUT_MS);
         const { linkedAgents, additionalAgents } = parseFanOutAgents(`${child.stdout}\n${child.stderr}`);
         if (child.timedOut) {
             return {
@@ -143,29 +143,40 @@ function runNode(args, env, timeoutMs) {
         });
     });
 }
-function scrubAgentRuntimeEnv(env) {
-    const next = { ...env };
-    for (const key of [
-        "AI_AGENT",
-        "CODEX_SANDBOX",
-        "CODEX_CI",
-        "CODEX_THREAD_ID",
-        "CURSOR_TRACE_ID",
-        "CURSOR_AGENT",
-        "CURSOR_EXTENSION_HOST_ROLE",
-        "GEMINI_CLI",
-        "ANTIGRAVITY_AGENT",
-        "AUGMENT_AGENT",
-        "OPENCODE_CLIENT",
-        "CLAUDECODE",
-        "CLAUDE_CODE",
-        "CLAUDE_CODE_IS_COWORK",
-        "REPL_ID",
-        "COPILOT_MODEL",
-        "COPILOT_ALLOW_ALL",
-        "COPILOT_GITHUB_TOKEN",
-    ]) {
-        delete next[key];
+// The skills CLI changes behavior when it detects it is running inside an
+// agent (via @vercel/detect-agent reading agent-specific environment
+// variables). Fan-out must always run in machine mode, so the child gets a
+// minimal allowlisted environment instead of a hand-maintained blacklist of
+// every agent's marker variables.
+function minimalChildEnv(env) {
+    const allowed = [
+        "PATH",
+        "HOME",
+        "USER",
+        "SHELL",
+        "TMPDIR",
+        "TMP",
+        "TEMP",
+        "LANG",
+        "NODE_EXTRA_CA_CERTS",
+        "HTTP_PROXY",
+        "HTTPS_PROXY",
+        "NO_PROXY",
+        "http_proxy",
+        "https_proxy",
+        "no_proxy",
+        "SYSTEMROOT",
+        "SystemRoot",
+        "COMSPEC",
+        "APPDATA",
+        "LOCALAPPDATA",
+        "USERPROFILE",
+    ];
+    const next = {};
+    for (const key of Object.keys(env)) {
+        if (allowed.includes(key) || key.startsWith("XDG_") || key.startsWith("LC_")) {
+            next[key] = env[key];
+        }
     }
     return next;
 }

package/dist/index.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAiEA,MAAM,WAAW,KAAK;IACpB,MAAM,EAAE,MAAM,CAAC,cAAc,CAAC;IAC9B,MAAM,EAAE,MAAM,CAAC,cAAc,CAAC;CAC/B;AAuTD,wBAAsB,MAAM,CAAC,IAAI,EAAE,SAAS,MAAM,EAAE,EAAE,EAAE,GAAE,KAGzD,GAAG,OAAO,CAAC,MAAM,CAAC,CAsMlB"}
+{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAgEA,MAAM,WAAW,KAAK;IACpB,MAAM,EAAE,MAAM,CAAC,cAAc,CAAC;IAC9B,MAAM,EAAE,MAAM,CAAC,cAAc,CAAC;CAC/B;AAuTD,wBAAsB,MAAM,CAAC,IAAI,EAAE,SAAS,MAAM,EAAE,EAAE,EAAE,GAAE,KAGzD,GAAG,OAAO,CAAC,MAAM,CAAC,CAoMlB"}

package/dist/index.js

@@ -4,7 +4,7 @@ import { createRequire } from "node:module";
 import os from "node:os";
 import path from "node:path";
 import { gzipSync } from "node:zlib";
-import { addWorkbenchRemote, addWorkbenchAgent, compareWorkbench, createWorkbenchInspectionSnapshot, createWorkbenchAdapterAuthBundle, createWorkbenchReadOnlyInspectionSnapshot, diffWorkbenchVersions, evalWorkbenchSkill, improveWorkbenchSkill, initWorkbenchSkill, listWorkbenchAgents, listWorkbenchVersions, localWorkbenchAdapterAuthStore, parseWorkbenchAdapterAuthTarget, prepareWorkbenchCloudEvalRequest, prepareWorkbenchCloudImproveRequest, publishWorkbenchVersion, removeWorkbenchAgent, showWorkbenchRef, switchWorkbenchVersion, syncWorkbenchRemote, workbenchJobEvidenceForSnapshot, workbenchSkillImproveCanUseQueuedAdapter, workbenchStatusSnapshot, WorkbenchCodedError, WorkbenchUserError, } from "@workbench-ai/workbench-core";
+import { addWorkbenchRemote, addWorkbenchAgent, compareWorkbench, createWorkbenchInspectionSnapshot, createWorkbenchAdapterAuthBundle, createWorkbenchReadOnlyInspectionSnapshot, diffWorkbenchVersions, evalWorkbenchSkill, improveWorkbenchSkill, initWorkbenchSkill, listWorkbenchAgents, listWorkbenchVersions, localWorkbenchAdapterAuthStore, parseWorkbenchAdapterAuthTarget, prepareWorkbenchCloudEvalRequest, prepareWorkbenchCloudImproveRequest, publishWorkbenchVersion, removeWorkbenchAgent, showWorkbenchRef, switchWorkbenchVersion, syncWorkbenchRemote, workbenchJobEvidenceForSnapshot, workbenchStatusSnapshot, WorkbenchCodedError, WorkbenchUserError, } from "@workbench-ai/workbench-core";
 import { normalizeWorkbenchSkillName } from "@workbench-ai/workbench-contract";
 import { emitError, emitResult } from "./output.js";
 import { fanOutSkill, manualFanOutCommand } from "./fanout.js";
@@ -375,12 +375,10 @@ export async function runCli(argv, io = {
             if (parsed.flags.cloud === true) {
                 return await handleCloudImprove(parsed, io);
             }
-            const improverAgent = await resolveLocalImproverAgent(parsed, core);
             const result = await improveWorkbenchSkill({
                 ...core,
                 skill: stringFlag(parsed, "skills"),
                 agent: stringFlag(parsed, "agents"),
-                ...(improverAgent ? { improverAgent } : {}),
                 budget: intFlag(parsed, "budget"),
                 samples: intFlag(parsed, "samples"),
             });
@@ -1110,15 +1108,9 @@ async function startCloudExecution(command, parsed, io) {
             exitCode: 1,
         });
     }
+    const syncAfterSchedule = await syncWorkbenchRemote({ ...core, remote: remote.name });
     const initialRunIds = runs.map((run) => run.id);
     writeCloudProgress(io, `workbench cloud: scheduled hosted ${command} on ${remote.url} (${formatCloudRunStatuses(runs)}).`, showProgress);
-    let initialSyncAfter = syncBefore;
-    try {
-        initialSyncAfter = await syncWorkbenchRemote({ ...core, remote: remote.name });
-    }
-    catch (error) {
-        writeCloudProgress(io, `workbench cloud: sync while waiting failed; retrying (${oneLineExcerpt(errorMessage(error)) ?? "unknown error"}).`, showProgress);
-    }
     writeCloudProgress(io, `workbench cloud: waiting for terminal status; press Ctrl-C to detach and resume with workbench show ${displayRef(initialRunIds[0] ?? "run")}.`, showProgress);
     const completed = await waitForCloudRuns({
         command,
@@ -1127,7 +1119,9 @@ async function startCloudExecution(command, parsed, io) {
         progress: showProgress,
         remote,
         runs,
-        initialSync: initialSyncAfter,
+        source,
+        skillId,
+        initialSync: syncAfterSchedule,
     });
     return {
         core,
@@ -1163,7 +1157,6 @@ async function waitForCloudRuns(input) {
     let interrupted = false;
     const startedAtMs = Date.now();
     let lastProgressAtMs = startedAtMs;
-    let lastSyncErrorMessage;
     const onSigint = () => {
         interrupted = true;
         writeCloudProgress(input.io, `workbench cloud: detaching from hosted ${input.command} (${runIds.map(displayRef).join(", ")}).`, input.progress);
@@ -1172,14 +1165,7 @@ async function waitForCloudRuns(input) {
     const seenStatuses = new Map();
     try {
         while (true) {
-            const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(input.core);
-            const snapshotRuns = runIds
-                .map((id) => snapshot.runs.find((entry) => entry.id === id))
-                .filter((run) => Boolean(run));
-            if (snapshotRuns.length > 0) {
-                runs = runIds.map((id) => snapshotRuns.find((entry) => entry.id === id) ?? runs.find((entry) => entry.id === id))
-                    .filter((run) => Boolean(run));
-            }
+            runs = await fetchCloudRuns(input.source.baseUrl, input.skillId, runIds, runs);
             let wroteProgress = false;
             const nowMs = Date.now();
             for (const run of runs) {
@@ -1191,6 +1177,7 @@ async function waitForCloudRuns(input) {
                 }
             }
             if (runs.length === runIds.length && runs.every(isTerminalRun)) {
+                sync = await syncWorkbenchRemote({ ...input.core, remote: input.remote.name });
                 return { runs, sync };
             }
             if (wroteProgress) {
@@ -1218,26 +1205,18 @@ async function waitForCloudRuns(input) {
             if (interrupted) {
                 return { runs, sync, detached: true };
             }
-            try {
-                sync = await syncWorkbenchRemote({ ...input.core, remote: input.remote.name });
-                if (lastSyncErrorMessage) {
-                    writeCloudProgress(input.io, "workbench cloud: sync while waiting recovered.", input.progress);
-                    lastSyncErrorMessage = undefined;
-                }
-            }
-            catch (error) {
-                const message = oneLineExcerpt(errorMessage(error)) ?? "unknown error";
-                if (message !== lastSyncErrorMessage) {
-                    writeCloudProgress(input.io, `workbench cloud: sync while waiting failed; retrying (${message}).`, input.progress);
-                    lastSyncErrorMessage = message;
-                }
-            }
         }
     }
     finally {
         process.off("SIGINT", onSigint);
     }
 }
+async function fetchCloudRuns(baseUrl, skillId, runIds, fallback) {
+    const responses = await Promise.all(runIds.map((runId) => apiRequest(`/api/workbench/skills/${encodeURIComponent(skillId)}/runs/${encodeURIComponent(runId)}`, {}, baseUrl)));
+    return runIds
+        .map((runId, index) => responses[index]?.run ?? fallback.find((run) => run.id === runId))
+        .filter((run) => Boolean(run));
+}
 function isTerminalRun(run) {
     return run.status === "succeeded" || run.status === "failed" || run.status === "canceled";
 }
@@ -1353,15 +1332,11 @@ async function resolveCloudSkillId(source) {
 function cloudExecutionRequestBody(command, request) {
     return {
         versionId: request.versionId,
-        evalHash: request.evalHash,
         skill: request.skill,
-        skillBundleHash: request.skillBundleHash,
         agent: request.agent,
-        agentHash: request.agentHash,
         samples: request.samples,
         ...(command === "improve" ? {
             budget: request.budget,
-            evidenceTraceIds: request.evidenceTraceIds,
         } : {}),
     };
 }
@@ -2701,10 +2676,10 @@ async function workbenchMachineStatus(auth) {
     };
 }
 function scoredRunValue(run) {
-    return run.status === "succeeded" && typeof run.score === "number" ? run.score : undefined;
+    return typeof run.score === "number" ? run.score : undefined;
 }
 function scoredJobValue(job) {
-    return job.status === "succeeded" && typeof job.score === "number" ? job.score : undefined;
+    return typeof job.score === "number" ? job.score : undefined;
 }
 function snapshotHasWorkflowCase(snapshot) {
     const currentVersion = snapshotVersionByRef(snapshot, snapshot.status.currentVersionId ?? snapshot.refs.current ?? "");
@@ -2981,41 +2956,6 @@ function manifestOnly(value) {
     }
     return out;
 }
-async function resolveLocalImproverAgent(parsed, core) {
-    if (stringFlag(parsed, "agents")) {
-        return undefined;
-    }
-    const agents = await listWorkbenchAgents(core).catch(() => []);
-    const status = await workbenchStatusSnapshot(core).catch(() => undefined);
-    const defaultAgentName = status?.project.defaultAgent ?? agents[0]?.name;
-    const defaultAgent = agents.find((agent) => agent.name === defaultAgentName);
-    if (defaultAgent && workbenchSkillImproveCanUseQueuedAdapter(defaultAgent)) {
-        return undefined;
-    }
-    const connected = await localWorkbenchAdapterAuthStore(adapterAuthStoreRoot()).listStatus().catch(() => []);
-    const candidates = connected
-        .filter((entry) => entry.status === "connected" &&
-        (entry.adapterId === "claude" || entry.adapterId === "codex"))
-        .sort((left, right) => {
-        const adapterRank = (adapter) => adapter === "claude" ? 0 : adapter === "codex" ? 1 : 2;
-        return adapterRank(left.adapterId) - adapterRank(right.adapterId) ||
-            (Date.parse(right.updatedAt ?? "") || 0) - (Date.parse(left.updatedAt ?? "") || 0);
-    });
-    const selected = candidates[0];
-    if (!selected) {
-        throw new WorkbenchCodedError("auth_required", "workbench improve needs a connected improver.", {
-            remediation: "Run workbench login claude (or codex) to connect an improver.",
-            exitCode: 1,
-        });
-    }
-    return {
-        name: selected.adapterId,
-        adapter: selected.adapterId,
-        config: {
-            auth: selected.slot ? { [selected.slot]: selected.profile } : selected.profile,
-        },
-    };
-}
 function formatLogEntry(entry) {
     if (entry.kind === "version") {
         return `${entry.createdAt}\tversion\t${displayRef(entry.id)}\tfiles=${entry.fileCount}\t${entry.message}`;

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@workbench-ai/workbench",
-  "version": "0.0.74",
+  "version": "0.0.75",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/workbench-ai/workbench.git",
@@ -22,10 +22,10 @@
   "dependencies": {
     "skills": "1.5.11",
     "yaml": "^2.8.2",
-    "@workbench-ai/workbench-built-in-adapters": "0.0.74",
-    "@workbench-ai/workbench-contract": "0.0.74",
-    "@workbench-ai/workbench-protocol": "0.0.74",
-    "@workbench-ai/workbench-core": "0.0.74"
+    "@workbench-ai/workbench-built-in-adapters": "0.0.75",
+    "@workbench-ai/workbench-contract": "0.0.75",
+    "@workbench-ai/workbench-core": "0.0.75",
+    "@workbench-ai/workbench-protocol": "0.0.75"
   },
   "devDependencies": {
     "@tailwindcss/postcss": "^4.2.2",
@@ -36,7 +36,7 @@
     "react-dom": "^19.2.0",
     "typescript": "^5.9.2",
     "vitest": "^3.2.4",
-    "@workbench-ai/workbench-ui": "0.0.74"
+    "@workbench-ai/workbench-ui": "0.0.75"
   },
   "scripts": {
     "build": "rm -rf dist && tsc -p tsconfig.json && chmod 755 dist/workbench.js && node ./scripts/build-dev-open-assets.mjs",