@workbench-ai/workbench 0.0.88 → 0.0.90

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,CA2MlB"}
+{"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,CAmNlB"}

package/dist/index.js

@@ -415,11 +415,15 @@ export async function runCli(argv, io = {
             return await handleAgent(parsed, io);
         }
         if (command === "sync") {
+            const beforeRuns = parsed.flags["dry-run"] === true
+                ? undefined
+                : await runEvidenceFingerprints(core).catch(() => undefined);
             const result = await syncWorkbenchRemote({
                 ...core,
                 remote: optionalPositional(parsed, 1),
                 dryRun: parsed.flags["dry-run"] === true,
             });
+            const next = result.dryRun ? null : await syncNextCommand(core, beforeRuns);
             return emitResult("workbench.cli.sync.v1", {
                 remote: result.remote,
                 status: result.dryRun ? "dry_run" : "synced",
@@ -427,8 +431,12 @@ export async function runCli(argv, io = {
                 pulled: result.pulled,
                 changed: syncChanged(result),
                 publication: result.publication,
+                next: next,
                 ...(result.dryRun ? { dryRun: true } : {}),
-            }, parsed, io, () => `${result.dryRun ? "Would sync" : "Synced"} ${result.remote.name}: pushed ${result.pushed}, pulled ${result.pulled}${result.upToDate ? " (up to date)" : ""}.`);
+            }, parsed, io, () => [
+                `${result.dryRun ? "Would sync" : "Synced"} ${result.remote.name}: pushed ${result.pushed}, pulled ${result.pulled}${result.upToDate ? " (up to date)" : ""}.`,
+                ...(next ? [`next: ${next}`] : []),
+            ].join("\n"));
         }
         if (command === "publish") {
             const preview = parsed.flags["dry-run"] === true
@@ -614,12 +622,14 @@ async function handleAgent(parsed, io) {
         if (!adapter) {
             throw new WorkbenchUserError("workbench agent add requires --adapter ADAPTER.");
         }
+        const config = parseWithFlags(parsed);
+        validateAgentCommandConfig(config);
         const agent = await addWorkbenchAgent({
             ...(await coreOptions(parsed)),
             name,
             adapter,
             model: stringFlag(parsed, "model"),
-            config: parseWithFlags(parsed),
+            config,
         });
         return output(agent, parsed, io, () => `Added agent ${formatAgent(agent)}.`);
     }
@@ -656,7 +666,17 @@ async function handleAdapterLogin(provider, parsed, io) {
 async function handleAdapterLogout(provider, parsed, io) {
     const target = parseAuthTarget(provider, authProfileFlag(parsed));
     await localWorkbenchAdapterAuthStore(adapterAuthStoreRoot()).disconnect(target);
-    const remote = await deleteAdapterConnectionRemote(target, parsed);
+    const remote = await deleteAdapterConnectionRemote(target, parsed).catch((error) => {
+        if (error instanceof WorkbenchCodedError && error.code === "auth_required") {
+            return {
+                status: "not_authenticated",
+                sync: "skipped",
+                reason: "not_authenticated",
+                remediation: "Run workbench login.",
+            };
+        }
+        throw error;
+    });
     return emitResult("workbench.cli.logout.v1", {
         provider: target.adapterId,
         localAdapter: {
@@ -1061,7 +1081,6 @@ async function startCloudExecution(command, parsed, io) {
     const showProgress = true;
     const interrupt = createCloudInterruptController(command, io, showProgress);
     try {
-        writeCloudProgress(io, `workbench cloud: preparing hosted ${command}.`, showProgress);
         const remote = await cloudPreScheduleStep(command, interrupt, ensureCloudRemoteForExecution(root, parsed));
         const source = parseWorkbenchInstallSource(remote.url);
         if (!source) {
@@ -1079,6 +1098,7 @@ async function startCloudExecution(command, parsed, io) {
             });
         }
         const core = { dir: root, authToken: token };
+        writeCloudProgress(io, `workbench cloud: preparing hosted ${command}.`, showProgress);
         writeCloudProgress(io, "workbench cloud: preparing current source.", showProgress);
         const request = command === "eval"
             ? await cloudPreScheduleStep(command, interrupt, prepareWorkbenchCloudEvalRequest({
@@ -1506,7 +1526,6 @@ function hostedImproveResult(started, artifactIds, switchedVersion) {
     const runs = started.runs.map((run) => runSummary(run, artifactIds.get(run.id) ?? []));
     return {
         run: runs[0] ?? null,
-        runs,
         switched: Boolean(switchedVersion),
         promoted: Boolean(switchedVersion),
         ...(switchedVersion ? { version: versionSummary(switchedVersion) } : {}),
@@ -1524,6 +1543,49 @@ function cloudSyncSummary(sync) {
 function syncChanged(sync) {
     return sync.pushed > 0 || sync.pulled > 0;
 }
+async function syncNextCommand(core, beforeRuns) {
+    if (beforeRuns) {
+        const changedRun = await latestChangedRunAfterSync(core, beforeRuns);
+        if (changedRun) {
+            return `workbench show ${displayRef(changedRun.id)}`;
+        }
+    }
+    const status = await workbenchStatusSnapshot(core);
+    const auth = await workbenchCliAuthStatus();
+    const cliStatus = await statusWithCausalNext(status, auth, core, {
+        installedSkillCount: 0,
+        stores: [],
+        connectedProviders: [],
+    });
+    return cliStatus.next ?? null;
+}
+async function latestChangedRunAfterSync(core, beforeRuns) {
+    const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core).catch(() => null);
+    const changedRuns = snapshot?.runs
+        .filter((run) => beforeRuns.get(run.id) !== runEvidenceFingerprint(run))
+        .sort((left, right) => runEvidenceTime(right).localeCompare(runEvidenceTime(left))) ?? [];
+    return changedRuns[0] ?? null;
+}
+async function runEvidenceFingerprints(core) {
+    const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
+    return new Map(snapshot.runs.map((run) => [run.id, runEvidenceFingerprint(run)]));
+}
+function runEvidenceFingerprint(run) {
+    return JSON.stringify({
+        status: run.status,
+        score: run.score,
+        costUsd: run.costUsd,
+        latencyMs: run.latencyMs,
+        jobIds: run.jobIds ?? [],
+        traceIds: run.traceIds,
+        finishedAt: run.finishedAt,
+        outputVersionId: run.outputVersionId,
+        error: run.error,
+    });
+}
+function runEvidenceTime(run) {
+    return run.finishedAt ?? run.createdAt;
+}
 function writeCloudProgress(io, message, enabled = true) {
     if (!enabled) {
         return;
@@ -2741,6 +2803,33 @@ function parseWithFlags(parsed) {
         return [entry.slice(0, eq), parseScalar(entry.slice(eq + 1))];
     }));
 }
+function validateAgentCommandConfig(config) {
+    for (const key of ["command", "improveCommand"]) {
+        const value = config[key];
+        if (typeof value !== "string") {
+            continue;
+        }
+        const expanded = expandedRuntimeEnvPath(value);
+        if (!expanded) {
+            continue;
+        }
+        throw new WorkbenchCodedError("usage", `--with ${key}=... contains ${expanded.path}, which usually means the shell expanded a Workbench runtime variable before Workbench received it.`, {
+            remediation: `Wrap the assignment in single quotes, for example --with '${key}=... >> "${expanded.replacement}"'.`,
+            exitCode: 2,
+        });
+    }
+}
+function expandedRuntimeEnvPath(value) {
+    for (const entry of [
+        { path: "/SKILL.md", replacement: "$SKILL_DIR/SKILL.md", pattern: /(^|[\s"'=])\/SKILL\.md(?=$|[\s"'])/u },
+        { path: "/result.json", replacement: "$OUTPUT_DIR/result.json", pattern: /(^|[\s"'=])\/result\.json(?=$|[\s"'])/u },
+    ]) {
+        if (entry.pattern.test(value)) {
+            return entry;
+        }
+    }
+    return null;
+}
 function parseScalar(value) {
     if (value === "true") {
         return true;
@@ -3692,7 +3781,7 @@ function formatTraceDetail(detail, refs = {}) {
     return detail.executions.map((execution) => {
         const sessionLabels = execution.sessions.map((session) => session.label).join(",");
         return [
-            `${execution.id}\trun=${refs.runRefs?.get(detail.runId) ?? displayRef(detail.runId)}\tjobs=${execution.jobIds.map((id) => refs.jobRefs?.get(id) ?? displayRef(id)).join(",")}\tstatus=${execution.status}`,
+            `${formatExecutionEvidenceLabel(detail, execution)}\trun=${refs.runRefs?.get(detail.runId) ?? displayRef(detail.runId)}\tjobs=${execution.jobIds.map((id) => refs.jobRefs?.get(id) ?? displayRef(id)).join(",")}\tstatus=${execution.status}`,
             `events=${execution.trace.events.length}`,
             `spans=${execution.trace.spans.length}`,
             `summaries=${execution.trace.summaries.length}`,
@@ -3700,6 +3789,11 @@ function formatTraceDetail(detail, refs = {}) {
         ].filter(Boolean).join("\t");
     }).join("\n");
 }
+function formatExecutionEvidenceLabel(detail, execution) {
+    return execution.jobIds.length === 1 && execution.id === `job:${detail.runId}:${execution.jobIds[0]}`
+        ? "evidence"
+        : execution.id;
+}
 function formatArtifact(artifact) {
     return `${displayRef(artifact.id)}\trun=${displayRef(artifact.runId)}\tjob=${displayRef(artifact.jobId)}\t${artifact.kind}\tfiles=${artifact.files.length}`;
 }

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@workbench-ai/workbench",
-  "version": "0.0.88",
+  "version": "0.0.90",
   "repository": {
     "type": "git",
     "url": "git+https://github.com/workbench-ai/workbench.git",
@@ -21,10 +21,10 @@
   ],
   "dependencies": {
     "yaml": "^2.8.2",
-    "@workbench-ai/workbench-contract": "0.0.88",
-    "@workbench-ai/workbench-built-in-adapters": "0.0.88",
-    "@workbench-ai/workbench-protocol": "0.0.88",
-    "@workbench-ai/workbench-core": "0.0.88"
+    "@workbench-ai/workbench-built-in-adapters": "0.0.90",
+    "@workbench-ai/workbench-contract": "0.0.90",
+    "@workbench-ai/workbench-core": "0.0.90",
+    "@workbench-ai/workbench-protocol": "0.0.90"
   },
   "devDependencies": {
     "@tailwindcss/postcss": "^4.2.2",
@@ -35,7 +35,7 @@
     "react-dom": "^19.2.0",
     "typescript": "^5.9.2",
     "vitest": "^3.2.4",
-    "@workbench-ai/workbench-ui": "0.0.88"
+    "@workbench-ai/workbench-ui": "0.0.90"
   },
   "scripts": {
     "build": "rm -rf dist && tsc -p tsconfig.json && chmod 755 dist/workbench.js && node ./scripts/build-dev-open-assets.mjs",