@workbench-ai/workbench 0.0.73 → 0.0.74

package/dist/index.js

@@ -4,10 +4,11 @@ import { createRequire } from "node:module";
 import os from "node:os";
 import path from "node:path";
 import { gzipSync } from "node:zlib";
-import { addWorkbenchCase, addWorkbenchRemote, addWorkbenchAgent, compareWorkbench, createWorkbenchInspectionSnapshot, createWorkbenchAdapterAuthBundle, createWorkbenchReadOnlyInspectionSnapshot, diffWorkbenchVersions, evalWorkbenchSkill, improveWorkbenchSkill, initWorkbenchSkill, listWorkbenchAgents, listWorkbenchVersions, localWorkbenchAdapterAuthStore, parseWorkbenchAdapterAuthTarget, 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, workbenchSkillImproveCanUseQueuedAdapter, workbenchStatusSnapshot, WorkbenchCodedError, WorkbenchUserError, } from "@workbench-ai/workbench-core";
 import { normalizeWorkbenchSkillName } from "@workbench-ai/workbench-contract";
 import { emitError, emitResult } from "./output.js";
-import { installSnapshotToTargets, normalizeInstallSnapshotPath, resolveInstallTargets, supportedInstallTargets, } from "./install-targets.js";
+import { fanOutSkill, manualFanOutCommand } from "./fanout.js";
+import { installedInventoryToJson, installSnapshotToStore, normalizeInstallSnapshotPath, readInstalledSkillsInventory, } from "./install-targets.js";
 import { startWorkbenchOpenServer } from "./open-server.js";
 const require = createRequire(import.meta.url);
 const HELP = [
@@ -19,11 +20,11 @@ const HELP = [
     "",
     "Taught commands:",
     "  workbench new [DIR] [--json]",
-    "  workbench eval [VERSION] [--skills all|LIST] [--agents all|LIST] [-n N|--samples N] [--rerun] [--cloud] [--json]",
-    "  workbench improve [VERSION] [--skills LIST] [--agents LIST] [--budget N] [-n N|--samples N] [--cloud] [--json]",
+    "  workbench eval [--skills all|LIST] [--agents all|LIST] [-n N|--samples N] [--rerun] [--cloud] [--json]",
+    "  workbench improve [--skills LIST] [--agents LIST] [--budget N] [-n N|--samples N] [--cloud] [--json]",
     "  workbench compare [--skills all|LIST] [--agents all|LIST] [--versions all|A..B|LIST] [--json]",
     "  workbench publish [VERSION] [--as OWNER/SKILL] [--private|--team|--public] [--dry-run] [--json]",
-    "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--dry-run] [--json]",
+    "  workbench install [HANDLE_OR_URL] [--yes] [--dry-run] [--json]",
     "",
     "More:",
     "  workbench help --all",
@@ -32,11 +33,11 @@ const HELP_ALL = [
     "Usage:",
     "  workbench                          # = workbench status",
     "  workbench new [DIR] [--json]",
-    "  workbench eval [VERSION] [--skills all|LIST] [--agents all|LIST] [-n N|--samples N] [--rerun] [--cloud] [--json]",
+    "  workbench eval [--skills all|LIST] [--agents all|LIST] [-n N|--samples N] [--rerun] [--cloud] [--json]",
     "  workbench compare [--skills all|LIST] [--agents all|LIST] [--versions all|A..B|LIST] [--json]",
-    "  workbench improve [VERSION] [--skills LIST] [--agents LIST] [--budget N] [-n N|--samples N] [--cloud] [--json]",
+    "  workbench improve [--skills LIST] [--agents LIST] [--budget N] [-n N|--samples N] [--cloud] [--json]",
     "  workbench publish [VERSION] [--as OWNER/SKILL] [--private|--team|--public] [--dry-run] [--json]",
-    "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--dry-run] [--json]",
+    "  workbench install [HANDLE_OR_URL] [--yes] [--dry-run] [--json]",
     "",
     "Inspect:",
     "  workbench status [--dir DIR] [--json]",
@@ -47,7 +48,6 @@ const HELP_ALL = [
     "  workbench open [--host HOST] [--port PORT] [--no-open]",
     "",
     "Configure:",
-    "  workbench case add RUN_ID [--json]",
     "  workbench agent add NAME --adapter X [--model M] [--with k=v]... | list | rm NAME [--json]",
     "",
     "Share and auth:",
@@ -71,7 +71,7 @@ const COMMAND_HELP = {
     ].join("\n"),
     eval: [
         "Usage:",
-        "  workbench eval [VERSION] [--skills all|LIST] [--agents all|LIST] [-n N|--samples N] [--rerun] [--cloud] [--json]",
+        "  workbench eval [--skills all|LIST] [--agents all|LIST] [-n N|--samples N] [--rerun] [--cloud] [--json]",
         "",
         "Runs eval jobs for the selected version, measured skills, and agents. Omitted selectors use manifest defaults.",
         "",
@@ -80,7 +80,7 @@ const COMMAND_HELP = {
     ].join("\n"),
     improve: [
         "Usage:",
-        "  workbench improve [VERSION] [--skills LIST] [--agents LIST] [--budget N] [-n N|--samples N] [--cloud] [--json]",
+        "  workbench improve [--skills LIST] [--agents LIST] [--budget N] [-n N|--samples N] [--cloud] [--json]",
         "",
         "Creates one improved child version from evidence. The selected skills and agents must resolve to exactly one entry each.",
         "",
@@ -98,12 +98,12 @@ const COMMAND_HELP = {
     ].join("\n"),
     install: [
         "Usage:",
-        "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--dry-run] [--json]",
+        "  workbench install [HANDLE_OR_URL] [--yes] [--dry-run] [--json]",
         "",
-        "Installs published Workbench Cloud source into local agent targets.",
+        "Installs published source into the canonical machine skill store, or lists installed skills when no source is given.",
         "",
         "Example:",
-        "  workbench install acme/earnings-prep --to codex --yes",
+        "  workbench install acme/earnings-prep",
     ].join("\n"),
     status: [
         "Usage:",
@@ -169,15 +169,6 @@ const COMMAND_HELP = {
         "Example:",
         "  workbench open --no-open",
     ].join("\n"),
-    case: [
-        "Usage:",
-        "  workbench case add RUN_ID [--json]",
-        "",
-        "Captures a regression case from a recorded run.",
-        "",
-        "Example:",
-        "  workbench case add run_abc12345",
-    ].join("\n"),
     agent: [
         "Usage:",
         "  workbench agent list [--json]",
@@ -256,7 +247,7 @@ const COMMAND_FLAGS = {
         samples: "positive-integer",
         skills: "string",
     },
-    install: { ...COMMON_FLAGS, ...HELP_FLAG, "dry-run": "boolean", to: "repeat-string", yes: "boolean" },
+    install: { ...COMMON_FLAGS, ...HELP_FLAG, "dry-run": "boolean", yes: "boolean" },
     log: { ...PROJECT_FLAGS, ...HELP_FLAG, runs: "boolean", versions: "boolean" },
     login: {
         ...COMMON_FLAGS,
@@ -290,11 +281,6 @@ const COMMAND_FLAGS = {
     version: { ...COMMON_FLAGS, ...VERSION_FLAG },
 };
 const SUBCOMMAND_FLAGS = {
-    case: {
-        flags: {
-            add: { ...PROJECT_FLAGS, ...HELP_FLAG },
-        },
-    },
     agent: {
         flags: {
             list: { ...PROJECT_FLAGS, ...HELP_FLAG },
@@ -345,12 +331,16 @@ export async function runCli(argv, io = {
             return await handleStatus(parsed, io);
         }
         if (command === "eval") {
+            rejectExtraInput(parsed, {
+                maxPositionals: 1,
+                message: "workbench eval does not accept a VERSION argument.",
+                remediation: "Run workbench eval for current source, or use workbench show VERSION to inspect old source.",
+            });
             if (parsed.flags.cloud === true) {
                 return await handleCloudEval(parsed, io);
             }
             const runs = await evalWorkbenchSkill({
                 ...core,
-                version: optionalPositional(parsed, 1),
                 skill: stringFlag(parsed, "skills"),
                 agent: stringFlag(parsed, "agents"),
                 samples: intFlag(parsed, "samples"),
@@ -377,13 +367,17 @@ export async function runCli(argv, io = {
             ].filter(Boolean).join("\n"));
         }
         if (command === "improve") {
+            rejectExtraInput(parsed, {
+                maxPositionals: 1,
+                message: "workbench improve does not accept a VERSION argument.",
+                remediation: "Run workbench improve for current source after recording failed or reviewed eval evidence.",
+            });
             if (parsed.flags.cloud === true) {
                 return await handleCloudImprove(parsed, io);
             }
             const improverAgent = await resolveLocalImproverAgent(parsed, core);
             const result = await improveWorkbenchSkill({
                 ...core,
-                version: optionalPositional(parsed, 1),
                 skill: stringFlag(parsed, "skills"),
                 agent: stringFlag(parsed, "agents"),
                 ...(improverAgent ? { improverAgent } : {}),
@@ -423,9 +417,6 @@ export async function runCli(argv, io = {
         if (command === "agent") {
             return await handleAgent(parsed, io);
         }
-        if (command === "case") {
-            return await handleCase(parsed, io);
-        }
         if (command === "sync") {
             const result = await syncWorkbenchRemote({
                 ...core,
@@ -505,15 +496,17 @@ async function handleStatus(parsed, io) {
     const core = await coreOptions(parsed);
     const status = await workbenchStatusSnapshot(core);
     const auth = await workbenchCliAuthStatus();
-    const cliStatus = await statusWithCausalNext(status, auth, core);
+    const machine = await workbenchMachineStatus(auth);
+    const cliStatus = await statusWithCausalNext(status, auth, core, machine);
     return emitResult("workbench.status.v1", {
         project: cliStatus.project,
         worktree: cliStatus.worktree,
         runs: cliStatus.runs,
         remotes: cliStatus.remotes,
         auth: auth,
+        machine: machine,
         next: cliStatus.next,
-    }, parsed, io, () => formatStatusSnapshot({ ...cliStatus, auth }));
+    }, parsed, io, () => formatStatusSnapshot({ ...cliStatus, auth, machine }));
 }
 async function handleLog(parsed, io) {
     if (parsed.flags.runs === true && parsed.flags.versions === true) {
@@ -584,7 +577,8 @@ async function handleShow(parsed, io) {
     }
     const trace = snapshotObjectByRef(snapshot.traces, objectRef, "trace");
     if (trace) {
-        return output(fileListing("trace", trace.id, trace.files), parsed, io, () => formatFileListing("trace", trace.id, trace.files));
+        const files = trace.files.filter(isUserFacingTraceEvidenceFile);
+        return output(fileListing("trace", trace.id, files), parsed, io, () => formatFileListing("trace", trace.id, files));
     }
     const artifact = snapshotObjectByRef(snapshot.artifacts, objectRef, "artifact");
     if (artifact) {
@@ -628,21 +622,6 @@ async function handleAgent(parsed, io) {
     }
     throw new WorkbenchUserError(`Unsupported agent command: ${subcommand}`);
 }
-async function handleCase(parsed, io) {
-    const subcommand = requiredPositional(parsed, 1, "workbench case requires add.");
-    if (subcommand === "add") {
-        const core = await coreOptions(parsed);
-        const sourceRef = requiredPositional(parsed, 2, "workbench case add requires RUN_ID.");
-        rejectExtraInput(parsed, {
-            maxPositionals: 3,
-            message: "workbench case add accepts one RUN_ID argument.",
-            remediation: "Run workbench case add RUN_ID.",
-        });
-        const record = await addWorkbenchCase({ ...core, fromTraceId: await traceIdForCaseSource(core, sourceRef) });
-        return output(record, parsed, io, () => `Added draft case ${record.id}. Edit .workbench/cases/${record.path}/case.yaml before using it as score evidence.`);
-    }
-    throw new WorkbenchUserError(`Unknown command: workbench case ${subcommand}`);
-}
 async function handleAdapterLogin(provider, parsed, io) {
     const target = parseAuthTarget(provider, authProfileFlag(parsed));
     const method = authMethod(parsed, target.adapterId);
@@ -862,67 +841,87 @@ async function handleLogout(parsed, io) {
     if (tokenPresent) {
         await writeConfig({ schema: CONFIG_SCHEMA, ...(baseUrl ? { baseUrl } : {}) });
     }
-    const adapterStatuses = await localWorkbenchAdapterAuthStore(adapterAuthStoreRoot()).listStatus().catch(() => []);
-    const adapterAuthRetained = adapterStatuses.length > 0;
     return emitResult("workbench.cli.logout.v1", {
         ...(baseUrl ? { baseUrl } : {}),
         tokenPresent,
         revoke,
         configRemoved,
-        adapterAuthRetained,
+        adapterAuth: "unchanged",
     }, parsed, io, () => [
         `Logged out of Workbench${baseUrl ? ` (${baseUrl})` : ""}.`,
         `Token: ${tokenPresent ? "present" : "absent"}; revoke ${revoke}; config ${configRemoved ? "removed" : "unchanged"}.`,
-        adapterAuthRetained
-            ? "Local adapter auth records were retained; run workbench logout PROVIDER to remove them."
-            : "No local adapter auth records remain.",
+        "Local adapter auth unchanged; run workbench logout PROVIDER to remove provider credentials.",
     ].join("\n"));
 }
 async function handleInstall(parsed, io) {
-    const sourceInput = requiredPositional(parsed, 1, "workbench install requires HANDLE_OR_URL.");
+    const sourceInput = optionalPositional(parsed, 1);
+    if (!sourceInput) {
+        rejectExtraInput(parsed, {
+            maxPositionals: 1,
+            message: "workbench install with no HANDLE_OR_URL lists installed skills.",
+            remediation: "Run workbench install OWNER/SKILL to install a published skill.",
+        });
+        if (parsed.flags.yes === true || parsed.flags["dry-run"] === true) {
+            throw new WorkbenchCodedError("usage", "workbench install inventory does not accept --yes or --dry-run.", {
+                remediation: "Run workbench install --json, or run workbench install OWNER/SKILL --dry-run.",
+                exitCode: 2,
+            });
+        }
+        const inventory = await readInstalledSkillsInventory({
+            includeUpdates: true,
+            lookupLatestVersion: latestInstallVersion,
+        });
+        return emitResult("workbench.cli.installed.v1", installedInventoryToJson(inventory), parsed, io, () => formatInstalledInventory(inventory));
+    }
     rejectExtraInput(parsed, {
         maxPositionals: 2,
         message: "workbench install accepts one HANDLE_OR_URL argument.",
-        remediation: "Run workbench install OWNER/SKILL --to codex.",
+        remediation: "Run workbench install OWNER/SKILL.",
     });
     const source = await resolveWorkbenchInstallSourceInput(sourceInput);
     const workbenchSource = parseWorkbenchInstallSource(source);
     if (!workbenchSource) {
         throw new WorkbenchCodedError("usage", "workbench install requires a Workbench Cloud source URL.", {
-            remediation: "Run workbench install OWNER/SKILL --to codex.",
+            remediation: "Run workbench install OWNER/SKILL.",
             exitCode: 2,
         });
     }
     const snapshot = await fetchWorkbenchInstallSourceSnapshot(workbenchSource, source);
     const sourceSummary = workbenchInstallSourceSummary(workbenchSource, snapshot);
-    const config = await loadConfig();
-    const toTargets = stringsFlag(parsed, "to");
-    const selectedTargets = toTargets.length > 0 ? normalizeInstallTargetNames(toTargets) : await defaultInstallTargetNames(config);
-    const targets = resolveInstallTargets({
-        agents: selectedTargets.filter((target) => target !== "local"),
-        local: selectedTargets.some((target) => target === "local"),
-        skillName: snapshot.name,
-    });
-    const result = await installSnapshotToTargets({
+    const result = await installSnapshotToStore({
         snapshot,
-        targets,
         overwrite: parsed.flags.yes === true,
         dryRun: parsed.flags["dry-run"] === true,
+        provenance: {
+            handle: `${workbenchSource.owner}/${workbenchSource.skill}`,
+            versionId: snapshot.versionId,
+            baseUrl: workbenchSource.baseUrl,
+        },
     });
-    if (toTargets.length > 0 && parsed.flags["dry-run"] !== true) {
-        await writeConfig({ ...config, installTargets: selectedTargets });
-    }
+    const fanout = parsed.flags["dry-run"] === true
+        ? skippedFanOut(result.directoryName, result.destination)
+        : await fanOutSkill(result.directoryName, { skillDir: result.destination });
+    const next = installNextCommand(fanout);
     return emitResult("workbench.cli.install.v1", {
         source: sourceSummary,
         result: result.result,
-        targets: result.targets,
+        store: result.store,
+        skill: result.directoryName,
+        destination: result.destination,
+        previous: result.previous,
         filesCopied: result.filesCopied,
+        contentHash: result.contentHash,
+        provenancePath: result.provenancePath,
+        fanout: fanOutToJson(fanout),
+        next: next,
         ...(parsed.flags["dry-run"] === true ? { dryRun: true } : {}),
     }, parsed, io, () => [
         parsed.flags["dry-run"] === true
-            ? `Would install ${snapshot.name}: filesCopied=${result.filesCopied}`
-            : `Installed ${snapshot.name}: ${result.result}`,
-        ...result.targets.map((target) => `  ${target.agent}\t${target.previous}\t${target.destination}`),
+            ? `Would install ${result.directoryName} to ${result.destination}: filesCopied=${result.filesCopied}`
+            : `Installed ${result.directoryName}: ${result.result}`,
+        `  machine\t${result.previous}\t${result.destination}`,
+        formatFanOut(fanout),
+        ...(next ? [`next: ${next}`] : []),
     ].join("\n"));
 }
 async function handleCloudEval(parsed, io) {
@@ -1003,44 +1002,67 @@ async function handleCloudImprove(parsed, io) {
         ...(next ? [`next: ${next}`] : []),
     ].filter(Boolean).join("\n"));
 }
-async function defaultInstallTargetNames(config) {
-    if (config.installTargets && config.installTargets.length > 0) {
-        return config.installTargets;
+function skippedFanOut(name, destination) {
+    return {
+        status: "skipped",
+        command: manualFanOutCommand(destination, name),
+        linkedAgents: [],
+        reason: "dry-run",
+    };
+}
+function installNextCommand(fanout) {
+    return fanout.status === "failed" || (fanout.status === "skipped" && fanout.reason !== "dry-run")
+        ? fanout.command
+        : null;
+}
+function fanOutToJson(fanout) {
+    return {
+        status: fanout.status,
+        command: fanout.command,
+        linkedAgents: fanout.linkedAgents,
+        ...(fanout.additionalAgents ? { additionalAgents: fanout.additionalAgents } : {}),
+        ...(fanout.reason ? { reason: fanout.reason } : {}),
+        ...(fanout.exitCode !== undefined ? { exitCode: fanout.exitCode } : {}),
+    };
+}
+function formatFanOut(fanout) {
+    if (fanout.status === "skipped") {
+        return fanout.reason === "dry-run"
+            ? "fanout: planned"
+            : `fanout skipped: ${fanout.reason ?? "not available"}`;
     }
-    const detected = [];
-    for (const target of supportedInstallTargets()) {
-        if (target.agent === "local") {
-            continue;
-        }
-        const home = path.dirname(path.dirname(target.destination));
-        if (await pathExists(home)) {
-            detected.push(target.agent);
-        }
+    if (fanout.status === "failed") {
+        return `fanout failed: ${fanout.reason ?? "unknown failure"}`;
     }
-    return detected.length > 0 ? detected : ["local"];
-}
-function normalizeInstallTargetNames(values) {
-    const normalized = [];
-    for (const value of values) {
-        const target = value.trim().toLowerCase();
-        if (target !== "codex" && target !== "claude" && target !== "local") {
-            throw new WorkbenchCodedError("usage", `Unsupported install target: ${value}`, {
-                remediation: "Use --to codex, --to claude, or --to local.",
-                exitCode: 2,
-            });
-        }
-        normalized.push(target);
+    if (fanout.linkedAgents.length === 0) {
+        return "fanout: completed";
     }
-    return [...new Set(normalized)];
+    const suffix = fanout.additionalAgents ? ` and ${fanout.additionalAgents} more` : "";
+    return `fanned out to: ${fanout.linkedAgents.join(", ")}${suffix}`;
 }
-async function pathExists(filePath) {
-    try {
-        await fs.access(filePath);
-        return true;
-    }
-    catch {
-        return false;
+async function latestInstallVersion(record) {
+    const handle = normalizedOwnerSkillHandle(record.handle);
+    if (!handle) {
+        return undefined;
     }
+    const snapshot = await withTimeout(fetchWorkbenchInstallSourceSnapshot({
+        baseUrl: record.baseUrl,
+        owner: handle.owner,
+        skill: handle.skill,
+    }, record.handle), 3000);
+    return snapshot.versionId;
+}
+function withTimeout(promise, timeoutMs) {
+    return new Promise((resolve, reject) => {
+        const timer = setTimeout(() => reject(new Error(`Timed out after ${timeoutMs}ms.`)), timeoutMs);
+        promise.then((value) => {
+            clearTimeout(timer);
+            resolve(value);
+        }, (error) => {
+            clearTimeout(timer);
+            reject(error);
+        });
+    });
 }
 async function startCloudExecution(command, parsed, io) {
     const root = dirFlag(parsed) ?? process.cwd();
@@ -1062,12 +1084,23 @@ async function startCloudExecution(command, parsed, io) {
         });
     }
     const core = { dir: root, authToken: token };
-    writeCloudProgress(io, `workbench cloud: syncing ${remote.name} before hosted ${command}.`, showProgress);
+    const request = command === "eval"
+        ? await prepareWorkbenchCloudEvalRequest({
+            ...core,
+            skill: stringFlag(parsed, "skills"),
+            agent: stringFlag(parsed, "agents"),
+            samples: intFlag(parsed, "samples"),
+        })
+        : await prepareWorkbenchCloudImproveRequest({
+            ...core,
+            skill: stringFlag(parsed, "skills"),
+            agent: stringFlag(parsed, "agents"),
+            samples: intFlag(parsed, "samples"),
+            budget: intFlag(parsed, "budget"),
+        });
     const syncBefore = await syncWorkbenchRemote({ ...core, remote: remote.name });
-    writeCloudProgress(io, `workbench cloud: synced ${remote.name} before hosted ${command} (pushed=${syncBefore.pushed}, pulled=${syncBefore.pulled}, up-to-date=${syncBefore.upToDate}).`, showProgress);
-    const startSnapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
     const skillId = await resolveCloudSkillId(source);
-    const response = await apiRequest(`/api/workbench/skills/${encodeURIComponent(skillId)}${command === "improve" ? "/improve" : "/runs"}`, { method: "POST", body: cloudExecutionRequestBody(command, parsed) }, source.baseUrl);
+    const response = await apiRequest(`/api/workbench/skills/${encodeURIComponent(skillId)}${command === "improve" ? "/improve" : "/runs"}`, { method: "POST", body: cloudExecutionRequestBody(command, request) }, source.baseUrl);
     const runs = response.runs ?? [];
     if (runs.length === 0) {
         throw new WorkbenchCodedError("cloud_run_missing", `Workbench Cloud did not return a run for ${command}.`, {
@@ -1079,9 +1112,14 @@ async function startCloudExecution(command, parsed, io) {
     }
     const initialRunIds = runs.map((run) => run.id);
     writeCloudProgress(io, `workbench cloud: scheduled hosted ${command} on ${remote.url} (${formatCloudRunStatuses(runs)}).`, showProgress);
-    const initialSyncAfter = await syncWorkbenchRemote({ ...core, remote: remote.name });
-    writeCloudProgress(io, `workbench cloud: synced after scheduling hosted ${command} (pushed=${initialSyncAfter.pushed}, pulled=${initialSyncAfter.pulled}, up-to-date=${initialSyncAfter.upToDate}).`, showProgress);
-    writeCloudProgress(io, `workbench cloud: waiting for terminal status; press Ctrl-C to detach and resume with workbench status or workbench show ${displayRef(initialRunIds[0] ?? "run")}.`, showProgress);
+    let initialSyncAfter = syncBefore;
+    try {
+        initialSyncAfter = await syncWorkbenchRemote({ ...core, remote: remote.name });
+    }
+    catch (error) {
+        writeCloudProgress(io, `workbench cloud: sync while waiting failed; retrying (${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,
         core,
@@ -1098,7 +1136,7 @@ async function startCloudExecution(command, parsed, io) {
         initialRunIds,
         runs: completed.runs,
         ...(completed.detached ? { detached: true } : {}),
-        startVersionId: startSnapshot.status.currentVersionId ?? startSnapshot.refs.current,
+        startVersionId: request.versionId,
         source,
         sync: {
             before: { pushed: syncBefore.pushed, pulled: syncBefore.pulled, upToDate: syncBefore.upToDate },
@@ -1123,6 +1161,9 @@ async function waitForCloudRuns(input) {
     const deadline = Date.now() + timeoutMs;
     let runs = [...input.runs];
     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);
@@ -1139,17 +1180,26 @@ async function waitForCloudRuns(input) {
                 runs = runIds.map((id) => snapshotRuns.find((entry) => entry.id === id) ?? runs.find((entry) => entry.id === id))
                     .filter((run) => Boolean(run));
             }
+            let wroteProgress = false;
+            const nowMs = Date.now();
             for (const run of runs) {
                 const previous = seenStatuses.get(run.id);
                 if (previous !== run.status) {
                     seenStatuses.set(run.id, run.status);
-                    writeCloudProgress(input.io, `workbench cloud: ${displayRef(run.id)} is ${run.status}.`, input.progress);
+                    writeCloudProgress(input.io, `workbench cloud: ${formatCloudRunState(run, startedAtMs, nowMs)}.`, input.progress);
+                    wroteProgress = input.progress || wroteProgress;
                 }
             }
             if (runs.length === runIds.length && runs.every(isTerminalRun)) {
-                writeCloudProgress(input.io, `workbench cloud: hosted ${input.command} finished (${formatCloudRunStatuses(runs)}).`, input.progress);
                 return { runs, sync };
             }
+            if (wroteProgress) {
+                lastProgressAtMs = nowMs;
+            }
+            else if (input.progress && nowMs - lastProgressAtMs >= 60_000) {
+                writeCloudProgress(input.io, `workbench cloud: still waiting (${formatCloudRunStates(runs, startedAtMs, nowMs)}).`);
+                lastProgressAtMs = nowMs;
+            }
             if (interrupted) {
                 return { runs, sync, detached: true };
             }
@@ -1168,8 +1218,20 @@ async function waitForCloudRuns(input) {
             if (interrupted) {
                 return { runs, sync, detached: true };
             }
-            sync = await syncWorkbenchRemote({ ...input.core, remote: input.remote.name });
-            writeCloudProgress(input.io, `workbench cloud: synced ${input.remote.name} while waiting (${formatCloudRunStatuses(runs)}).`, input.progress);
+            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 (${message}).`, input.progress);
+                    lastSyncErrorMessage = message;
+                }
+            }
         }
     }
     finally {
@@ -1288,13 +1350,19 @@ async function resolveCloudSkillId(source) {
     }
     return skill.id;
 }
-function cloudExecutionRequestBody(command, parsed) {
+function cloudExecutionRequestBody(command, request) {
     return {
-        version: optionalPositional(parsed, 1),
-        skill: stringFlag(parsed, "skills"),
-        agent: stringFlag(parsed, "agents"),
-        samples: intFlag(parsed, "samples"),
-        ...(command === "improve" ? { budget: intFlag(parsed, "budget") } : {}),
+        versionId: request.versionId,
+        evalHash: request.evalHash,
+        skill: request.skill,
+        skillBundleHash: request.skillBundleHash,
+        agent: request.agent,
+        agentHash: request.agentHash,
+        samples: request.samples,
+        ...(command === "improve" ? {
+            budget: request.budget,
+            evidenceTraceIds: request.evidenceTraceIds,
+        } : {}),
     };
 }
 function cloudImproveNextCommand(runs) {
@@ -1335,6 +1403,17 @@ function formatCloudRunStatuses(runs) {
         ? runs.map((run) => `${displayRef(run.id)}:${run.status}`).join(", ")
         : "no runs";
 }
+function formatCloudRunStates(runs, startedAtMs, nowMs) {
+    return runs.length > 0
+        ? runs.map((run) => formatCloudRunState(run, startedAtMs, nowMs)).join(", ")
+        : `no runs (${elapsedSeconds(startedAtMs, nowMs)}s)`;
+}
+function formatCloudRunState(run, startedAtMs, nowMs) {
+    return `${displayRef(run.id)} ${run.status} (${elapsedSeconds(startedAtMs, nowMs)}s)`;
+}
+function elapsedSeconds(startedAtMs, nowMs) {
+    return Math.max(0, Math.floor((nowMs - startedAtMs) / 1000));
+}
 function workbenchInstallSourceSummary(source, snapshot) {
     const installUrl = `${source.baseUrl}/skills/${encodeURIComponent(source.owner)}/${encodeURIComponent(source.skill)}`;
     return {
@@ -1495,7 +1574,6 @@ async function loadConfig() {
         ...(typeof parsed.baseUrl === "string" ? { baseUrl: normalizeBaseUrl(parsed.baseUrl) } : {}),
         ...(typeof parsed.accessToken === "string" ? { accessToken: parsed.accessToken } : {}),
         ...(typeof parsed.username === "string" ? { username: parsed.username } : {}),
-        ...(Array.isArray(parsed.installTargets) ? { installTargets: normalizeInstallTargetNames(parsed.installTargets.flatMap((entry) => typeof entry === "string" ? [entry] : [])) } : {}),
     };
 }
 // Single resolver for the Workbench Cloud token used by every authenticated
@@ -2250,7 +2328,7 @@ function flagSpecForParsedPrefix(positionals, flags) {
     return allowedFlagsForCommand({ positionals: [...positionals], flags: {} }, command);
 }
 function addFlag(flags, name, value) {
-    if (name === "with" || name === "to") {
+    if (name === "with") {
         const existing = flags[name];
         flags[name] = Array.isArray(existing)
             ? [...existing, String(value)]
@@ -2274,14 +2352,6 @@ function stringFlag(parsed, name) {
     const value = parsed.flags[name];
     return typeof value === "string" ? value : undefined;
 }
-function stringsFlag(parsed, name) {
-    const value = parsed.flags[name];
-    return Array.isArray(value)
-        ? value
-        : typeof value === "string"
-            ? [value]
-            : [];
-}
 function intFlag(parsed, name) {
     const value = stringFlag(parsed, name);
     if (!value) {
@@ -2382,12 +2452,17 @@ async function previewPublishWithDerivedRemote(parsed) {
     return {
         remote,
         version,
-        visibility: parsePublishVisibilityFlags(parsed) ?? "private",
+        visibility: parsePublishVisibilityFlags(parsed) ??
+            normalizePublishVisibility(reconciledSnapshot.refs["publication/visibility"]) ??
+            "private",
         installHandle: installHandleFromCloudRemote(remote),
         installUrl: remote.url,
         pinnedInstallUrl: `${remote.url}/releases/${encodeURIComponent(version.id)}`,
     };
 }
+function normalizePublishVisibility(value) {
+    return value === "private" || value === "internal" || value === "public" ? value : undefined;
+}
 async function ensurePublishRemote(parsed) {
     const core = await coreOptions(parsed);
     const root = path.resolve(dirFlag(parsed) ?? process.cwd());
@@ -2460,7 +2535,7 @@ async function resolveWorkbenchInstallSourceInput(input) {
     const handle = normalizedOwnerSkillHandle(input);
     if (!handle) {
         throw new WorkbenchCodedError("usage", "workbench install expects OWNER/SKILL or a Workbench Cloud skill URL.", {
-            remediation: "Run workbench install OWNER/SKILL --to codex.",
+            remediation: "Run workbench install OWNER/SKILL.",
             exitCode: 2,
         });
     }
@@ -2611,6 +2686,20 @@ async function workbenchCliAuthStatus() {
         })),
     };
 }
+async function workbenchMachineStatus(auth) {
+    const inventory = await readInstalledSkillsInventory();
+    return {
+        installedSkillCount: inventory.skills.length,
+        stores: inventory.stores,
+        connectedProviders: auth.adapters
+            .filter((entry) => entry.status === "connected")
+            .map((entry) => ({
+            adapter: entry.adapter,
+            ...(entry.slot ? { slot: entry.slot } : {}),
+            profile: entry.profile,
+        })),
+    };
+}
 function scoredRunValue(run) {
     return run.status === "succeeded" && typeof run.score === "number" ? run.score : undefined;
 }
@@ -2628,9 +2717,12 @@ function installHandleFromStatusRemote(remote) {
     const source = parseWorkbenchInstallSource(publicationUrl ?? remote.url);
     return source ? `${source.owner}/${source.skill}` : publicationUrl ?? remote.url;
 }
-async function statusWithCausalNext(status, auth, core) {
+async function statusWithCausalNext(status, auth, core, machine) {
     if (!status.project.initialized) {
-        return status;
+        return {
+            ...status,
+            next: machine.installedSkillCount > 0 ? "workbench install" : status.next,
+        };
     }
     const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core).catch(() => null);
     const lastRun = snapshot?.runs
@@ -2767,28 +2859,61 @@ function evidenceFilesForRunOrJob(snapshot, ref) {
     }
     const traceById = new Map(snapshot.traces.map((trace) => [trace.id, trace]));
     const artifactById = new Map(snapshot.artifacts.map((artifact) => [artifact.id, artifact]));
-    const files = selection.jobs.flatMap((job) => [
+    const candidates = selection.jobs.flatMap((job) => [
         ...job.artifactIds.flatMap((artifactId) => {
             const artifact = artifactById.get(artifactId);
             return artifact
-                ? artifact.files.filter(isUserFacingEvidenceFile).map((file) => evidenceFileWithPath(file, `cases/${evidencePathSegment(job.caseId)}/jobs/${evidencePathSegment(job.id)}/${file.path}`))
+                ? artifact.files.filter(isUserFacingEvidenceFile).map((file) => ({
+                    file: evidenceFileWithPath(file, `cases/${evidencePathSegment(job.caseId)}/jobs/${evidencePathSegment(job.id)}/${file.path}`),
+                    jobId: job.id,
+                    source: "artifact",
+                }))
                 : [];
         }),
         ...job.traceIds.flatMap((traceId) => {
             const trace = traceById.get(traceId);
             return trace
-                ? trace.files.filter(isUserFacingEvidenceFile).map((file) => evidenceFileWithPath(file, `cases/${evidencePathSegment(job.caseId)}/jobs/${evidencePathSegment(job.id)}/traces/${evidencePathSegment(trace.id)}/${file.path}`))
+                ? trace.files.filter(isUserFacingTraceEvidenceFile).map((file) => ({
+                    file: evidenceFileWithPath(file, `cases/${evidencePathSegment(job.caseId)}/jobs/${evidencePathSegment(job.id)}/traces/${evidencePathSegment(trace.id)}/${file.path}`),
+                    jobId: job.id,
+                    source: "trace",
+                }))
                 : [];
         }),
     ]);
+    return canonicalEvidenceFiles(candidates);
+}
+function canonicalEvidenceFiles(candidates) {
     const seen = new Set();
-    return files.filter((file) => {
+    const sameJobArtifactFiles = new Set();
+    const files = [];
+    for (const candidate of candidates) {
+        const file = candidate.file;
         if (seen.has(file.path)) {
-            return false;
+            continue;
         }
         seen.add(file.path);
-        return true;
-    });
+        const equivalentKey = sameJobEquivalentEvidenceKey(candidate);
+        if (candidate.source === "trace" && sameJobArtifactFiles.has(equivalentKey)) {
+            continue;
+        }
+        if (candidate.source === "artifact") {
+            sameJobArtifactFiles.add(equivalentKey);
+        }
+        files.push(file);
+    }
+    return files;
+}
+function sameJobEquivalentEvidenceKey(candidate) {
+    const file = candidate.file;
+    return [
+        candidate.jobId,
+        path.basename(file.path),
+        file.kind ?? "text",
+        file.encoding ?? "utf8",
+        file.executable === true ? "1" : "0",
+        file.content,
+    ].join("\0");
 }
 function evidenceFileWithPath(file, filePath) {
     return {
@@ -2800,6 +2925,13 @@ function isUserFacingEvidenceFile(file) {
     const normalized = file.path.replace(/\\/gu, "/").replace(/^\/+/u, "");
     return normalized.split("/").every((segment) => segment !== ".workbench");
 }
+function isUserFacingTraceEvidenceFile(file) {
+    if (!isUserFacingEvidenceFile(file)) {
+        return false;
+    }
+    const basename = path.basename(file.path.replace(/\\/gu, "/"));
+    return basename !== "request.json" && basename !== "result.json" && basename !== "trace.json";
+}
 function evidencePathSegment(value) {
     return value.replace(/[^A-Za-z0-9._-]+/gu, "-") || "_";
 }
@@ -2948,17 +3080,6 @@ function findShowFile(files, requestedPath, objectRef) {
     const candidates = normalized === "stderr.log"
         ? suffixCandidates.filter((file) => file.content.length > 0)
         : suffixCandidates;
-    const canonicalCandidates = candidates.filter(isCanonicalEvidenceFileCandidate);
-    if (canonicalCandidates.length === 1) {
-        return canonicalCandidates[0];
-    }
-    const equivalentCanonicalCandidate = singleEquivalentShowFile(canonicalCandidates);
-    if (equivalentCanonicalCandidate) {
-        return equivalentCanonicalCandidate;
-    }
-    if (canonicalCandidates.length > 1) {
-        throw ambiguousShowPath(objectRef, requestedPath, canonicalCandidates);
-    }
     if (candidates.length === 1) {
         return candidates[0];
     }
@@ -2975,9 +3096,6 @@ function findShowFile(files, requestedPath, objectRef) {
     }
     throw ambiguousShowPath(objectRef, requestedPath, candidates.length > 0 ? candidates : suffixCandidates);
 }
-function isCanonicalEvidenceFileCandidate(file) {
-    return !file.path.includes("/traces/") && !file.path.includes("/artifacts/");
-}
 function singleEquivalentShowFile(files) {
     if (files.length <= 1) {
         return null;
@@ -3006,23 +3124,6 @@ function fileListing(kind, id, files) {
 function formatFileListing(kind, id, files) {
     return [`${kind}\t${displayRef(id)}\tfiles=${files.length}`, ...files.map((file) => file.path)].join("\n");
 }
-async function traceIdForCaseSource(core, ref) {
-    const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
-    const trace = snapshotObjectByRef(snapshot.traces, ref, "trace");
-    if (trace) {
-        return trace.id;
-    }
-    const selection = runOrJobEvidenceSelection(snapshot, ref);
-    const traceId = selection.run?.traceIds[0] ?? selection.jobs[0]?.traceIds[0];
-    if (traceId) {
-        return traceId;
-    }
-    throw new WorkbenchCodedError("ref_not_found", `Run, job, or trace not found: ${ref}`, {
-        remediation: "Run workbench log, then workbench case add RUN_ID.",
-        subject: { ref },
-        exitCode: 1,
-    });
-}
 async function evalCoverageSummaries(core, runs) {
     const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
     const jobsByRun = new Map();
@@ -3101,6 +3202,12 @@ function formatStatusSnapshot(status) {
     const lines = [
         `Root: ${status.project.root}`,
         `Initialized: ${status.project.initialized ? "yes" : "no"}`,
+        ...(status.project.initialized ? [] : [
+            `Installed skills: ${status.machine?.installedSkillCount ?? 0}`,
+            `Connected providers: ${status.machine?.connectedProviders.length
+                ? status.machine.connectedProviders.map((entry) => `${entry.adapter}/${entry.profile}`).join(", ")
+                : "none"}`,
+        ]),
         ...(status.project.currentVersionId ? [`Current version: ${displayRef(status.project.currentVersionId)}`] : []),
         ...(status.project.defaultSkill ? [`Default skill: ${status.project.defaultSkill}`] : []),
         ...(status.project.defaultAgent ? [`Default agent: ${status.project.defaultAgent}`] : []),
@@ -3130,6 +3237,36 @@ function formatStatusSnapshot(status) {
     ];
     return lines.join("\n");
 }
+function formatInstalledInventory(inventory) {
+    if (inventory.skills.length === 0) {
+        return [
+            "No skills installed.",
+            ...(inventory.next ? [`next: ${inventory.next}`] : []),
+        ].join("\n");
+    }
+    const lines = [
+        "store\tname\tversion\tstatus\tsource",
+        ...inventory.skills.map(formatInstalledSkill),
+        ...(inventory.next ? [`next: ${inventory.next}`] : []),
+    ];
+    return lines.join("\n");
+}
+function formatInstalledSkill(skill) {
+    return [
+        skill.store,
+        skill.name,
+        skill.versionId ? shortInstalledVersion(skill.versionId) : "-",
+        skill.status,
+        skill.handle ?? "(no provenance)",
+    ].join("\t");
+}
+function shortInstalledVersion(versionId) {
+    return versionId.startsWith("v_") && versionId.length > 10
+        ? displayRef(versionId)
+        : versionId.length > 12
+            ? versionId.slice(0, 12)
+            : versionId;
+}
 function formatVersion(version) {
     return `${displayRef(version.id)}\t${version.hash.slice(0, 12)}\t${version.message}`;
 }