@workbench-ai/workbench 0.0.70 → 0.0.72

package/dist/index.js

@@ -4,10 +4,10 @@ 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, createWorkbenchAdapterAuthBundle, createWorkbenchReadOnlyInspectionSnapshot, diffWorkbenchVersions, evalWorkbenchSkill, improveWorkbenchSkill, initWorkbenchSkill, listWorkbenchCases, listWorkbenchAgents, listWorkbenchVersions, localWorkbenchAdapterAuthStore, parseWorkbenchAdapterAuthTarget, publishWorkbenchVersion, removeWorkbenchCase, removeWorkbenchAgent, showWorkbenchRef, switchWorkbenchVersion, syncWorkbenchRemote, workbenchJobEvidenceForSnapshot, workbenchStatusSnapshot, WorkbenchCodedError, WorkbenchUserError, } from "@workbench-ai/workbench-core";
+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 { normalizeWorkbenchSkillName } from "@workbench-ai/workbench-contract";
 import { emitError, emitResult } from "./output.js";
-import { installSnapshotToTargets, installTargetsToJson, normalizeInstallSnapshotPath, resolveInstallTargets, supportedInstallTargets, } from "./install-targets.js";
+import { installSnapshotToTargets, normalizeInstallSnapshotPath, resolveInstallTargets, supportedInstallTargets, } from "./install-targets.js";
 import { startWorkbenchOpenServer } from "./open-server.js";
 const require = createRequire(import.meta.url);
 const HELP = [
@@ -23,7 +23,7 @@ const HELP = [
     "  workbench improve [VERSION] [--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] [--list] [--dry-run] [--json]",
+    "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--dry-run] [--json]",
     "",
     "More:",
     "  workbench help --all",
@@ -36,7 +36,7 @@ const HELP_ALL = [
     "  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 publish [VERSION] [--as OWNER/SKILL] [--private|--team|--public] [--dry-run] [--json]",
-    "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--list] [--dry-run] [--json]",
+    "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--dry-run] [--json]",
     "",
     "Inspect:",
     "  workbench status [--dir DIR] [--json]",
@@ -44,10 +44,10 @@ const HELP_ALL = [
     "  workbench show REF[:PATH] [--json]",
     "  workbench diff [A..B] [--json]",
     "  workbench switch VERSION [--json]",
-    "  workbench open [--host HOST] [--port PORT] [--no-open] [--json]",
+    "  workbench open [--host HOST] [--port PORT] [--no-open]",
     "",
     "Configure:",
-    "  workbench case add [RUN_ID] | list | rm ID [--json]",
+    "  workbench case add RUN_ID [--json]",
     "  workbench agent add NAME --adapter X [--model M] [--with k=v]... | list | rm NAME [--json]",
     "",
     "Share and auth:",
@@ -65,28 +65,40 @@ const COMMAND_HELP = {
         "  workbench new [DIR] [--json]",
         "",
         "Creates a Workbench skill project.",
+        "",
+        "Example:",
+        "  workbench new earnings-prep",
     ].join("\n"),
     eval: [
         "Usage:",
         "  workbench eval [VERSION] [--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.",
+        "",
+        "Example:",
+        "  workbench eval -n 5",
     ].join("\n"),
     improve: [
         "Usage:",
         "  workbench improve [VERSION] [--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.",
+        "",
+        "Example:",
+        "  workbench improve --budget 1 -n 1",
     ].join("\n"),
     compare: [
         "Usage:",
         "  workbench compare [--skills all|LIST] [--agents all|LIST] [--versions all|A..B|LIST] [--json]",
         "",
         "Compares recorded eval evidence across selected skills, agents, and versions.",
+        "",
+        "Example:",
+        "  workbench compare --agents all",
     ].join("\n"),
     install: [
         "Usage:",
-        "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--list] [--dry-run] [--json]",
+        "  workbench install HANDLE_OR_URL [--to codex|claude|local]... [--yes] [--dry-run] [--json]",
         "",
         "Installs published Workbench Cloud source into local agent targets.",
         "",
@@ -98,12 +110,18 @@ const COMMAND_HELP = {
         "  workbench status [--dir DIR] [--json]",
         "",
         "Reports project, worktree, run, per-remote sync/publication, and auth state. --json emits the workbench.status.v1 dashboard.",
+        "",
+        "Example:",
+        "  workbench status --json",
     ].join("\n"),
     logout: [
         "Usage:",
         "  workbench logout [PROVIDER] [--json]",
         "",
         "With no provider, logs out of Workbench Cloud. With a provider such as codex or claude, removes local adapter auth.",
+        "",
+        "Example:",
+        "  workbench logout claude",
     ].join("\n"),
     show: [
         "Usage:",
@@ -111,38 +129,54 @@ const COMMAND_HELP = {
         "  workbench show REF:PATH [--json]",
         "",
         "Shows a Workbench object, lists files for file-backed objects, or prints one file.",
+        "",
+        "Example:",
+        "  workbench show run_abc12345:result.json",
     ].join("\n"),
     log: [
         "Usage:",
         "  workbench log [--runs|--versions] [--json]",
         "",
         "Shows one reverse-chronological timeline of versions and runs.",
+        "",
+        "Example:",
+        "  workbench log --runs",
     ].join("\n"),
     diff: [
         "Usage:",
         "  workbench diff [A..B] [--json]",
         "",
         "Shows changed files between two Workbench source versions.",
+        "",
+        "Example:",
+        "  workbench diff 26059f9a..eac5699c",
     ].join("\n"),
     switch: [
         "Usage:",
         "  workbench switch VERSION [--json]",
         "",
         "Switches the working skill source to a recorded Workbench version.",
+        "",
+        "Example:",
+        "  workbench switch 26059f9a",
     ].join("\n"),
     open: [
         "Usage:",
-        "  workbench open [--host HOST] [--port PORT] [--no-open] [--json]",
+        "  workbench open [--host HOST] [--port PORT] [--no-open]",
         "",
-        "Serves or emits the read-only Workbench inspection snapshot.",
+        "Serves the read-only Workbench inspection UI.",
+        "",
+        "Example:",
+        "  workbench open --no-open",
     ].join("\n"),
     case: [
         "Usage:",
-        "  workbench case list [--json]",
-        "  workbench case add [RUN_ID] [--json]",
-        "  workbench case rm ID [--json]",
+        "  workbench case add RUN_ID [--json]",
+        "",
+        "Captures a regression case from a recorded run.",
         "",
-        "Lists cases, creates a draft case, or removes a case.",
+        "Example:",
+        "  workbench case add run_abc12345",
     ].join("\n"),
     agent: [
         "Usage:",
@@ -151,18 +185,27 @@ const COMMAND_HELP = {
         "  workbench agent rm NAME [--json]",
         "",
         "Lists, adds, or removes eval agent configurations.",
+        "",
+        "Example:",
+        "  workbench agent add claude --adapter claude --model sonnet",
     ].join("\n"),
     sync: [
         "Usage:",
         "  workbench sync [REMOTE] [--dry-run] [--dir DIR] [--json]",
         "",
         "Plumbing command: synchronizes local evidence and version objects with a Workbench remote.",
+        "",
+        "Example:",
+        "  workbench sync cloud --dry-run",
     ].join("\n"),
     publish: [
         "Usage:",
         "  workbench publish [VERSION] [--as OWNER/SKILL] [--private|--team|--public] [--dry-run] [--dir DIR] [--json]",
         "",
         "Publishes installable skill source to Workbench Cloud. --as sets the linked OWNER/SKILL handle.",
+        "",
+        "Example:",
+        "  workbench publish --as acme/earnings-prep --dry-run",
     ].join("\n"),
     login: [
         "Usage:",
@@ -170,14 +213,20 @@ const COMMAND_HELP = {
         "  workbench logout [PROVIDER] [--json]",
         "",
         "Connects the CLI to Workbench Cloud or captures local adapter auth for a provider.",
+        "",
+        "Example:",
+        "  workbench login --start-only --no-open",
     ].join("\n"),
 };
 const COMMON_FLAGS = {
     json: "boolean",
 };
+const DIR_FLAG = {
+    dir: "string",
+};
 const PROJECT_FLAGS = {
     ...COMMON_FLAGS,
-    dir: "string",
+    ...DIR_FLAG,
 };
 const HELP_FLAG = {
     help: "boolean",
@@ -207,7 +256,7 @@ const COMMAND_FLAGS = {
         samples: "positive-integer",
         skills: "string",
     },
-    install: { ...COMMON_FLAGS, ...HELP_FLAG, "dry-run": "boolean", list: "boolean", to: "repeat-string", yes: "boolean" },
+    install: { ...COMMON_FLAGS, ...HELP_FLAG, "dry-run": "boolean", to: "repeat-string", yes: "boolean" },
     log: { ...PROJECT_FLAGS, ...HELP_FLAG, runs: "boolean", versions: "boolean" },
     login: {
         ...COMMON_FLAGS,
@@ -224,7 +273,7 @@ const COMMAND_FLAGS = {
     },
     logout: { ...COMMON_FLAGS, ...HELP_FLAG },
     new: { ...PROJECT_FLAGS, ...HELP_FLAG },
-    open: { ...PROJECT_FLAGS, ...HELP_FLAG, host: "string", "no-open": "boolean", port: "positive-integer" },
+    open: { ...DIR_FLAG, ...HELP_FLAG, host: "string", "no-open": "boolean", port: "port" },
     publish: {
         ...PROJECT_FLAGS,
         ...HELP_FLAG,
@@ -243,9 +292,7 @@ const COMMAND_FLAGS = {
 const SUBCOMMAND_FLAGS = {
     case: {
         flags: {
-            list: { ...PROJECT_FLAGS, ...HELP_FLAG },
             add: { ...PROJECT_FLAGS, ...HELP_FLAG },
-            rm: { ...PROJECT_FLAGS, ...HELP_FLAG },
         },
     },
     agent: {
@@ -315,26 +362,31 @@ export async function runCli(argv, io = {
                 return emitEvalFailure(runs, failedRuns, artifactIds, parsed, io);
             }
             const deltas = await evalDeltas(core, runs);
-            const nextCommands = evalSuccessNextCommands(runs);
+            const coverage = await evalCoverageSummaries(core, runs);
+            const next = await evalSuccessNextCommand(core, runs);
             return emitResult("workbench.cli.eval.v1", {
                 result: runs.map((run) => runSummary(run, artifactIds.get(run.id) ?? [])),
+                coverage: coverage,
                 deltas: deltas,
-                nextCommands: nextCommands,
+                next: next,
             }, parsed, io, () => [
                 runs.map(formatRun).join("\n"),
+                ...coverage.map(formatEvalCoverage),
                 ...deltas.map(formatEvalDelta),
-                ...(nextCommands[0] ? [`next: ${nextCommands[0]}`] : []),
+                ...(next ? [`next: ${next}`] : []),
             ].filter(Boolean).join("\n"));
         }
         if (command === "improve") {
             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 } : {}),
                 budget: intFlag(parsed, "budget"),
                 samples: intFlag(parsed, "samples"),
             });
@@ -350,12 +402,12 @@ export async function runCli(argv, io = {
                 skills: stringFlag(parsed, "skills"),
                 agents: stringFlag(parsed, "agents"),
             });
-            return output(comparison, parsed, io, () => formatComparison(comparison));
+            return output(manifestOnly(comparison), parsed, io, () => formatComparison(comparison));
         }
         if (command === "switch") {
             const versionRef = requiredPositional(parsed, 1, "workbench switch requires VERSION.");
             const version = await switchWorkbenchVersion(versionRef, core);
-            return output(versionSummary(version), parsed, io, () => `Switched to ${version.id}.`);
+            return output(versionSummary(version), parsed, io, () => `Switched to ${displayRef(version.id)}.`);
         }
         if (command === "diff") {
             const range = optionalPositional(parsed, 1) ?? await defaultDiffRange(core);
@@ -390,7 +442,7 @@ export async function runCli(argv, io = {
             }, parsed, io, () => `${result.dryRun ? "Would sync" : "Synced"} ${result.remote.name}: pushed ${result.pushed}, pulled ${result.pulled}${result.upToDate ? " (up to date)" : ""}.`);
         }
         if (command === "publish") {
-            const preview = parsed.flags["dry-run"] === true && !stringFlag(parsed, "as")
+            const preview = parsed.flags["dry-run"] === true
                 ? await previewPublishWithDerivedRemote(parsed)
                 : undefined;
             if (preview) {
@@ -403,10 +455,7 @@ export async function runCli(argv, io = {
                     pinnedInstallUrl: preview.pinnedInstallUrl,
                     dryRun: true,
                 }, parsed, io, () => [
-                    `Would publish ${preview.version.id} to remote ${preview.remote.name}.`,
-                    `Visibility: ${preview.visibility}`,
-                    `Install: ${preview.installUrl}`,
-                    `Pinned: ${preview.pinnedInstallUrl}`,
+                    `Would publish ${displayRef(preview.version.id)} as ${preview.installHandle} (${preview.visibility}).`,
                     `next: workbench install ${preview.installHandle}`,
                 ].join("\n"));
             }
@@ -427,25 +476,18 @@ export async function runCli(argv, io = {
                 pinnedInstallUrl: result.pinnedInstallUrl,
                 ...(result.dryRun ? { dryRun: true } : {}),
             }, parsed, io, () => [
-                `${result.dryRun ? "Would publish" : "Published"} ${result.version.id} to remote ${result.remote.name}.`,
-                `Visibility: ${result.visibility}`,
-                `Install: ${result.installUrl}`,
-                `Pinned: ${result.pinnedInstallUrl}`,
+                `${result.dryRun ? "Would publish" : "Published"} ${displayRef(result.version.id)} as ${result.installHandle} (${result.visibility}).`,
                 `next: workbench install ${result.installHandle}`,
             ].join("\n"));
         }
         if (command === "open") {
-            if (parsed.flags.json === true) {
-                const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
-                return output(snapshot, parsed, io, () => "Read-only Workbench inspection data is available with --json.");
-            }
             // The browser server serves committed object state through a read-only
             // snapshot path, so long-running commands do not block page loads.
             const server = await startWorkbenchOpenServer({
                 dir: dirFlag(parsed),
                 authToken: core.authToken,
                 host: stringFlag(parsed, "host"),
-                port: intFlag(parsed, "port"),
+                port: portFlag(parsed, "port"),
             });
             io.stdout.write(`Workbench: ${server.url}\n`);
             if (parsed.flags["no-open"] !== true) {
@@ -460,16 +502,18 @@ export async function runCli(argv, io = {
     }
 }
 async function handleStatus(parsed, io) {
-    const status = await workbenchStatusSnapshot(await coreOptions(parsed));
+    const core = await coreOptions(parsed);
+    const status = await workbenchStatusSnapshot(core);
     const auth = await workbenchCliAuthStatus();
+    const cliStatus = await statusWithCausalNext(status, auth, core);
     return emitResult("workbench.status.v1", {
-        project: status.project,
-        worktree: status.worktree,
-        runs: status.runs,
-        remotes: status.remotes,
+        project: cliStatus.project,
+        worktree: cliStatus.worktree,
+        runs: cliStatus.runs,
+        remotes: cliStatus.remotes,
         auth: auth,
-        next: status.next,
-    }, parsed, io, () => formatStatusSnapshot({ ...status, auth }));
+        next: cliStatus.next,
+    }, parsed, io, () => formatStatusSnapshot({ ...cliStatus, auth }));
 }
 async function handleLog(parsed, io) {
     if (parsed.flags.runs === true && parsed.flags.versions === true) {
@@ -491,7 +535,7 @@ async function handleLog(parsed, io) {
             remediation: "Run workbench log, workbench log --runs, or workbench log --versions.",
         });
     }
-    const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(await coreOptions(parsed));
+    const snapshot = await createWorkbenchInspectionSnapshot(await coreOptions(parsed));
     const includeRuns = parsed.flags.versions !== true;
     const includeVersions = parsed.flags.runs !== true;
     const entries = [
@@ -510,7 +554,7 @@ async function handleLog(parsed, io) {
             versionId: run.versionId,
             skillName: run.skillName,
             agentName: run.agentName,
-            ...(run.score !== undefined ? { score: run.score } : {}),
+            ...(scoredRunValue(run) !== undefined ? { score: scoredRunValue(run) } : {}),
         })) : []),
     ].sort((left, right) => right.createdAt.localeCompare(left.createdAt));
     return emitResult("workbench.cli.log.v1", {
@@ -534,21 +578,25 @@ async function handleShow(parsed, io) {
         return output(value, parsed, io, () => formatShow(value));
     }
     const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
-    const version = snapshot.versions.find((entry) => entry.id === objectRef);
+    const version = snapshotVersionByRef(snapshot, objectRef);
     if (version) {
         return output(fileListing("version", version.id, version.files), parsed, io, () => formatFileListing("version", version.id, version.files));
     }
-    const trace = snapshot.traces.find((entry) => entry.id === objectRef);
+    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 artifact = snapshot.artifacts.find((entry) => entry.id === objectRef);
+    const artifact = snapshotObjectByRef(snapshot.artifacts, objectRef, "artifact");
     if (artifact) {
         return output(fileListing("artifact", artifact.id, artifact.files), parsed, io, () => formatFileListing("artifact", artifact.id, artifact.files));
     }
     const details = evidenceDetailsForRunOrJob(snapshot, objectRef);
-    if (details.length > 0) {
-        return output(details, parsed, io, () => details.map(formatTraceDetail).join("\n"));
+    const evidenceFiles = evidenceFilesForRunOrJob(snapshot, objectRef);
+    if (details.length > 0 || evidenceFiles.length > 0) {
+        return output({
+            details: details.map(evidenceDetailSummary),
+            files: evidenceFiles.map(fileSummary),
+        }, parsed, io, () => formatRunOrJobEvidence(details, evidenceFiles));
     }
     const value = await showWorkbenchRef(ref, core);
     return output(value, parsed, io, () => formatShow(value));
@@ -581,22 +629,19 @@ async function handleAgent(parsed, io) {
     throw new WorkbenchUserError(`Unsupported agent command: ${subcommand}`);
 }
 async function handleCase(parsed, io) {
-    const subcommand = requiredPositional(parsed, 1, "workbench case requires list|add|rm.");
-    if (subcommand === "list") {
-        const cases = await listWorkbenchCases(await coreOptions(parsed));
-        return output(cases, parsed, io, () => cases.map((entry) => `${entry.id}\t${entry.path}`).join("\n") || "No cases.");
-    }
+    const subcommand = requiredPositional(parsed, 1, "workbench case requires add.");
     if (subcommand === "add") {
         const core = await coreOptions(parsed);
-        const sourceRef = optionalPositional(parsed, 2);
-        const record = await addWorkbenchCase({ ...core, fromTraceId: sourceRef ? await traceIdForCaseSource(core, sourceRef) : undefined });
-        return output(record, parsed, io, () => `Added case ${record.id}.`);
-    }
-    if (subcommand === "rm") {
-        const result = await removeWorkbenchCase(requiredPositional(parsed, 2, "workbench case rm requires CASE_ID."), await coreOptions(parsed));
-        return output(result, parsed, io, () => `Removed case ${result.removed}.`);
+        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(`Unsupported case command: ${subcommand}`);
+    throw new WorkbenchUserError(`Unknown command: workbench case ${subcommand}`);
 }
 async function handleAdapterLogin(provider, parsed, io) {
     const target = parseAuthTarget(provider, authProfileFlag(parsed));
@@ -685,11 +730,14 @@ function validateFlagValue(name, value, kind) {
     if (typeof value !== "string" || !value.trim()) {
         throw new WorkbenchUserError(`--${name} requires a value.`);
     }
-    if (kind === "positive-integer") {
+    if (kind === "positive-integer" || kind === "port") {
         const parsedValue = Number(value);
-        if (!Number.isInteger(parsedValue) || parsedValue <= 0) {
+        if (kind === "positive-integer" && (!Number.isInteger(parsedValue) || parsedValue <= 0)) {
             throw new WorkbenchUserError(`--${name} must be a positive integer.`);
         }
+        if (kind === "port" && (!Number.isInteger(parsedValue) || parsedValue < 0 || parsedValue > 65535)) {
+            throw new WorkbenchUserError(`--${name} must be an integer between 0 and 65535.`);
+        }
     }
 }
 const CONFIG_SCHEMA = "workbench.cli.config.v1";
@@ -714,7 +762,7 @@ async function handleLogin(parsed, io) {
     }
     if (parsed.flags["start-only"] === true && parsed.flags.wait === true) {
         throw new WorkbenchCodedError("usage", "workbench login accepts only one of --start-only or --wait.", {
-            remediation: "Run workbench login --start-only or workbench login --wait --timeout 120.",
+            remediation: "Run workbench login --start-only or workbench login --wait.",
             exitCode: 2,
         });
     }
@@ -723,22 +771,17 @@ async function handleLogin(parsed, io) {
     const timeoutSeconds = intFlag(parsed, "timeout");
     if (startOnly && timeoutSeconds !== undefined) {
         throw new WorkbenchCodedError("usage", "workbench login --timeout only applies with --wait.", {
-            remediation: "Run workbench login --start-only, then workbench login --wait --timeout 120.",
-            exitCode: 2,
-        });
-    }
-    if (waitOnly && timeoutSeconds === undefined) {
-        throw new WorkbenchCodedError("usage", "workbench login --wait requires --timeout N.", {
-            remediation: "Run workbench login --wait --timeout 120.",
+            remediation: "Run workbench login --start-only, then workbench login --wait.",
             exitCode: 2,
         });
     }
     const config = await loadConfig();
-    const baseUrl = selectWorkbenchBaseUrl({
-        explicitBaseUrl: stringFlag(parsed, "base-url"),
+    const explicitBaseUrl = stringFlag(parsed, "base-url");
+    const pending = waitOnly ? await readPendingDeviceAuthorization(explicitBaseUrl) : null;
+    const baseUrl = pending?.baseUrl ?? selectWorkbenchBaseUrl({
+        explicitBaseUrl,
         configBaseUrl: config.baseUrl,
     });
-    const pending = waitOnly ? await readPendingDeviceAuthorization(baseUrl) : null;
     const record = pending ?? await startDeviceAuthorization(baseUrl);
     const freshAuthorization = pending === null;
     if (startOnly) {
@@ -753,8 +796,8 @@ async function handleLogin(parsed, io) {
             verificationUriComplete: record.verification_uri_complete,
             userCode: record.user_code,
             expiresAt: record.expiresAt,
-            resume: "workbench login --wait --timeout 120",
-        }, parsed, io, () => `Open ${record.verification_uri_complete}\nCode: ${record.user_code}\nResume: workbench login --wait --timeout 120`);
+            resume: "workbench login --wait",
+        }, parsed, io, () => `Open ${record.verification_uri_complete}\nCode: ${record.user_code}\nResume: workbench login --wait`);
     }
     await writePendingDeviceAuthorization(record);
     if (freshAuthorization && !parsed.flags.json) {
@@ -801,9 +844,6 @@ async function handleLogout(parsed, io) {
     const config = await loadConfig();
     const baseUrl = optionalWorkbenchBaseUrl({ configBaseUrl: config.baseUrl });
     const tokenPresent = Boolean(config.accessToken);
-    if (tokenPresent && !baseUrl) {
-        throw new WorkbenchUserError("Missing Workbench API URL. Set WORKBENCH_API_URL or run `workbench login --base-url URL`.");
-    }
     let revoke = "skipped";
     if (config.accessToken && baseUrl) {
         try {
@@ -856,18 +896,6 @@ async function handleInstall(parsed, io) {
     const snapshot = await fetchWorkbenchInstallSourceSnapshot(workbenchSource, source);
     const sourceSummary = workbenchInstallSourceSummary(workbenchSource, snapshot);
     const config = await loadConfig();
-    if (parsed.flags.list === true) {
-        return emitResult("workbench.cli.install.v1", {
-            source: sourceSummary,
-            skills: [snapshot.name],
-            fileCount: snapshot.files.length,
-            targets: installTargetsToJson(supportedInstallTargets()),
-        }, parsed, io, () => [
-            `${snapshot.name}\t${snapshot.versionId}\tfiles=${snapshot.files.length}`,
-            "Targets:",
-            ...supportedInstallTargets().map((target) => `  ${target.agent}\t${target.destination}`),
-        ].join("\n"));
-    }
     const toTargets = stringsFlag(parsed, "to");
     const selectedTargets = toTargets.length > 0 ? normalizeInstallTargetNames(toTargets) : await defaultInstallTargetNames(config);
     const targets = resolveInstallTargets({
@@ -898,29 +926,57 @@ async function handleInstall(parsed, io) {
     ].join("\n"));
 }
 async function handleCloudEval(parsed, io) {
-    const started = await startCloudExecution("eval", parsed);
+    const started = await startCloudExecution("eval", parsed, io);
     const artifactIds = await artifactIdsByRunId(started.core, started.runs);
+    if (started.detached) {
+        const next = cloudDetachedNextCommand(started.runs);
+        emitResult("workbench.cli.eval.v1", {
+            result: started.runs.map((run) => runSummary(run, artifactIds.get(run.id) ?? [])),
+            detached: true,
+            next: next,
+            cloud: cloudExecutionSummary(started),
+        }, parsed, io, () => [
+            `Detached from hosted eval on ${started.remote.url}.`,
+            started.runs.map(formatRun).join("\n"),
+            ...(next ? [`next: ${next}`] : []),
+        ].filter(Boolean).join("\n"));
+        return 130;
+    }
     const failedRuns = started.runs.filter((run) => run.status === "failed" || run.status === "canceled");
     if (failedRuns.length > 0) {
         return emitEvalFailure(started.runs, failedRuns, artifactIds, parsed, io);
     }
     const deltas = await evalDeltas(started.core, started.runs);
-    const nextCommands = cloudEvalNextCommands(started.runs);
+    const next = await evalSuccessNextCommand(started.core, started.runs);
     return emitResult("workbench.cli.eval.v1", {
         result: started.runs.map((run) => runSummary(run, artifactIds.get(run.id) ?? [])),
         deltas: deltas,
-        nextCommands: nextCommands,
+        next: next,
         cloud: cloudExecutionSummary(started),
     }, parsed, io, () => [
         `Completed hosted eval on ${started.remote.url}.`,
         started.runs.map(formatRun).join("\n"),
         ...deltas.map(formatEvalDelta),
-        ...(nextCommands[0] ? [`next: ${nextCommands[0]}`] : []),
+        ...(next ? [`next: ${next}`] : []),
     ].filter(Boolean).join("\n"));
 }
 async function handleCloudImprove(parsed, io) {
-    const started = await startCloudExecution("improve", parsed);
+    const started = await startCloudExecution("improve", parsed, io);
     const artifactIds = await artifactIdsByRunId(started.core, started.runs);
+    if (started.detached) {
+        const next = cloudDetachedNextCommand(started.runs);
+        emitResult("workbench.cli.improve.v1", {
+            result: started.runs.map((run) => runSummary(run, artifactIds.get(run.id) ?? [])),
+            detached: true,
+            next: next,
+            cloud: cloudExecutionSummary(started),
+        }, parsed, io, () => [
+            `Detached from hosted improve on ${started.remote.url}.`,
+            started.runs.map(formatRun).join("\n"),
+            ...(next ? [`next: ${next}`] : []),
+        ].filter(Boolean).join("\n"));
+        return 130;
+    }
     const failedRuns = started.runs.filter((run) => run.status === "failed" || run.status === "canceled");
     if (failedRuns.length > 0) {
         const first = failedRuns[0];
@@ -934,17 +990,17 @@ async function handleCloudImprove(parsed, io) {
         });
     }
     const switchedVersionId = await switchHostedImproveVersionIfPromoted(started);
-    const nextCommands = cloudImproveNextCommands(started.runs);
+    const next = cloudImproveNextCommand(started.runs);
     return emitResult("workbench.cli.improve.v1", {
         result: started.runs.map((run) => runSummary(run, artifactIds.get(run.id) ?? [])),
-        nextCommands: nextCommands,
+        next: next,
         cloud: cloudExecutionSummary(started),
         ...(switchedVersionId ? { switchedVersionId } : {}),
     }, parsed, io, () => [
         `Completed hosted improve on ${started.remote.url}.`,
         started.runs.map(formatRun).join("\n"),
-        ...(switchedVersionId ? [`Switched local source to ${switchedVersionId}.`] : []),
-        ...(nextCommands[0] ? [`next: ${nextCommands[0]}`] : []),
+        ...(switchedVersionId ? [`Switched local source to ${displayRef(switchedVersionId)}.`] : []),
+        ...(next ? [`next: ${next}`] : []),
     ].filter(Boolean).join("\n"));
 }
 async function defaultInstallTargetNames(config) {
@@ -986,7 +1042,7 @@ async function pathExists(filePath) {
         return false;
     }
 }
-async function startCloudExecution(command, parsed) {
+async function startCloudExecution(command, parsed, io) {
     const root = dirFlag(parsed) ?? process.cwd();
     const remote = await ensureCloudRemoteForExecution(root, parsed);
     const source = parseWorkbenchInstallSource(remote.url);
@@ -1005,7 +1061,9 @@ async function startCloudExecution(command, parsed) {
         });
     }
     const core = { dir: root, authToken: token };
+    writeCloudProgress(io, `workbench cloud: syncing ${remote.name} before hosted ${command}.`);
     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}).`);
     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);
@@ -1018,9 +1076,15 @@ async function startCloudExecution(command, parsed) {
             exitCode: 1,
         });
     }
+    const initialRunIds = runs.map((run) => run.id);
+    writeCloudProgress(io, `workbench cloud: scheduled hosted ${command} on ${remote.url} (${formatCloudRunStatuses(runs)}).`);
     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}).`);
+    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")}.`);
     const completed = await waitForCloudRuns({
+        command,
         core,
+        io,
         remote,
         runs,
         initialSync: initialSyncAfter,
@@ -1029,7 +1093,9 @@ async function startCloudExecution(command, parsed) {
         core,
         remote,
         skillId,
+        initialRunIds,
         runs: completed.runs,
+        ...(completed.detached ? { detached: true } : {}),
         startVersionId: startSnapshot.status.currentVersionId ?? startSnapshot.refs.current,
         source,
         sync: {
@@ -1053,27 +1119,59 @@ async function waitForCloudRuns(input) {
     const timeoutMs = positiveIntEnv("WORKBENCH_CLOUD_RUN_TIMEOUT_MS") ?? CLOUD_RUN_TIMEOUT_MS;
     const pollIntervalMs = positiveIntEnv("WORKBENCH_CLOUD_RUN_POLL_INTERVAL_MS") ?? CLOUD_RUN_POLL_INTERVAL_MS;
     const deadline = Date.now() + timeoutMs;
-    while (true) {
-        const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(input.core);
-        const runs = runIds
-            .map((id) => snapshot.runs.find((entry) => entry.id === id))
-            .filter((run) => Boolean(run));
-        if (runs.length === runIds.length && runs.every(isTerminalRun)) {
-            return { runs, sync };
-        }
-        if (Date.now() >= deadline) {
-            throw new WorkbenchCodedError("cloud_run_pending", "Hosted Workbench run is still running.", {
-                retryable: true,
-                remediation: runIds[0] ? `Run workbench show ${runIds[0]}.` : "Run workbench log --runs.",
-                subject: {
-                    runIds,
-                    statuses: Object.fromEntries(runs.map((run) => [run.id, run.status])),
-                },
-                exitCode: 1,
-            });
+    let runs = [...input.runs];
+    let interrupted = false;
+    const onSigint = () => {
+        interrupted = true;
+        writeCloudProgress(input.io, `workbench cloud: detaching from hosted ${input.command} (${runIds.map(displayRef).join(", ")}).`);
+    };
+    process.once("SIGINT", onSigint);
+    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));
+            }
+            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}.`);
+                }
+            }
+            if (runs.length === runIds.length && runs.every(isTerminalRun)) {
+                writeCloudProgress(input.io, `workbench cloud: hosted ${input.command} finished (${formatCloudRunStatuses(runs)}).`);
+                return { runs, sync };
+            }
+            if (interrupted) {
+                return { runs, sync, detached: true };
+            }
+            if (Date.now() >= deadline) {
+                throw new WorkbenchCodedError("cloud_run_pending", "Hosted Workbench run is still running.", {
+                    retryable: true,
+                    remediation: runIds[0] ? `Run workbench show ${runIds[0]}.` : "Run workbench log --runs.",
+                    subject: {
+                        runIds,
+                        statuses: Object.fromEntries(runs.map((run) => [run.id, run.status])),
+                    },
+                    exitCode: 1,
+                });
+            }
+            await sleep(pollIntervalMs);
+            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)}).`);
         }
-        await sleep(pollIntervalMs);
-        sync = await syncWorkbenchRemote({ ...input.core, remote: input.remote.name });
+    }
+    finally {
+        process.off("SIGINT", onSigint);
     }
 }
 function isTerminalRun(run) {
@@ -1197,30 +1295,41 @@ function cloudExecutionRequestBody(command, parsed) {
         ...(command === "improve" ? { budget: intFlag(parsed, "budget") } : {}),
     };
 }
-function cloudEvalNextCommands(runs) {
-    return cloudExecutionNextCommands(runs, "workbench publish");
+function cloudImproveNextCommand(runs) {
+    return cloudExecutionNextCommand(runs, "workbench eval");
 }
-function cloudImproveNextCommands(runs) {
-    return cloudExecutionNextCommands(runs, "workbench eval");
+function cloudDetachedNextCommand(runs) {
+    const first = runs[0];
+    return first?.id ? `workbench show ${displayRef(first.id)}` : "workbench status";
 }
-function cloudExecutionNextCommands(runs, successCommand) {
+function cloudExecutionNextCommand(runs, successCommand) {
     const first = runs[0];
     if (!first) {
-        return ["workbench log --runs"];
+        return "workbench log --runs";
     }
     if (first.status === "running" || first.status === "failed" || first.status === "canceled") {
-        return [`workbench show ${first.id}`];
+        return `workbench show ${displayRef(first.id)}`;
     }
-    return [successCommand];
+    return successCommand;
 }
 function cloudExecutionSummary(started) {
     return {
         remote: started.remote.name,
         url: started.remote.url,
         skillId: started.skillId,
+        initialRunIds: started.initialRunIds,
+        ...(started.detached ? { detached: true } : {}),
         sync: started.sync,
     };
 }
+function writeCloudProgress(io, message) {
+    io.stderr.write(`${message}\n`);
+}
+function formatCloudRunStatuses(runs) {
+    return runs.length > 0
+        ? runs.map((run) => `${displayRef(run.id)}:${run.status}`).join(", ")
+        : "no runs";
+}
 function workbenchInstallSourceSummary(source, snapshot) {
     const installUrl = `${source.baseUrl}/skills/${encodeURIComponent(source.owner)}/${encodeURIComponent(source.skill)}`;
     return {
@@ -1294,12 +1403,13 @@ async function fetchWorkbenchInstallSourceSnapshot(source, displaySource) {
         throw new WorkbenchCodedError("auth_required", token
             ? `Workbench Cloud rejected the provided token while installing ${displaySource}.`
             : `Authentication is required to install ${displaySource}.`, {
-            remediation: `Run workbench login --base-url ${source.baseUrl}.`,
+            remediation: "Run workbench login.",
             exitCode: 1,
         });
     }
     if (!response.ok) {
-        throw new WorkbenchCodedError("install_failed", `Unable to download Workbench source ${displaySource}: ${response.status} ${readResponseError(text) ?? response.statusText}`, {
+        const excerpt = readResponseError(text);
+        throw new WorkbenchCodedError("install_failed", `Unable to download Workbench source ${displaySource}: ${response.status}${excerpt ? ` ${excerpt}` : response.statusText ? ` ${response.statusText}` : ""}`, {
             subject: { source: displaySource, status: response.status },
             exitCode: 1,
         });
@@ -1420,18 +1530,15 @@ function deviceAuthPath() {
     return process.env.WORKBENCH_DEVICE_AUTH?.trim() || path.join(path.dirname(configPath()), "device-auth.json");
 }
 function selectWorkbenchBaseUrl(input = {}) {
-    const baseUrl = optionalWorkbenchBaseUrl(input);
-    if (!baseUrl) {
-        throw new WorkbenchUserError("Missing Workbench API URL. Pass --base-url URL, set WORKBENCH_API_URL, or run `workbench login --base-url URL`.");
-    }
-    return baseUrl;
+    return optionalWorkbenchBaseUrl(input);
 }
 function optionalWorkbenchBaseUrl(input = {}) {
     const value = input.explicitBaseUrl ??
         input.originBaseUrl ??
         process.env.WORKBENCH_API_URL ??
-        input.configBaseUrl;
-    return value ? normalizeBaseUrl(value) : undefined;
+        input.configBaseUrl ??
+        DEFAULT_WORKBENCH_CLOUD_BASE_URL;
+    return normalizeBaseUrl(value);
 }
 function normalizeBaseUrl(value) {
     return value.trim().replace(/\/+$/u, "");
@@ -1449,7 +1556,8 @@ async function requestDeviceAuthorization(baseUrl) {
         });
     }
     if (!response.ok) {
-        throw new WorkbenchCodedError("login_denied", `Device login failed: ${readResponseError(text) ?? response.statusText}`, {
+        const excerpt = readResponseError(text);
+        throw new WorkbenchCodedError("login_denied", `Device login failed: ${response.status}${excerpt ? ` ${excerpt}` : response.statusText ? ` ${response.statusText}` : ""}`, {
             exitCode: 1,
         });
     }
@@ -1500,7 +1608,7 @@ async function pollDeviceToken(baseUrl, authorization, timeoutSeconds) {
     }
     throw new WorkbenchCodedError("login_pending", "Device login is still waiting for browser authorization.", {
         retryable: true,
-        remediation: "Authorize the device in the browser, then run workbench login --wait --timeout 120.",
+        remediation: "Authorize the device in the browser, then run workbench login --wait.",
         subject: {
             retryAfterSeconds: Math.max(1, Math.ceil(intervalMs / 1000)),
             verificationUri: authorization.verification_uri,
@@ -1524,7 +1632,8 @@ async function fetchWorkbenchUsername(baseUrl, accessToken) {
 }
 async function readPendingDeviceAuthorization(baseUrl) {
     const record = await readDeviceAuthorizationJson(deviceAuthPath());
-    if (!record || record.baseUrl !== baseUrl || Date.parse(record.expiresAt) <= Date.now()) {
+    const expectedBaseUrl = baseUrl ? normalizeBaseUrl(baseUrl) : undefined;
+    if (!record || (expectedBaseUrl && record.baseUrl !== expectedBaseUrl) || Date.parse(record.expiresAt) <= Date.now()) {
         return null;
     }
     return record;
@@ -1614,7 +1723,8 @@ async function apiRequest(apiPath, options = {}, baseUrlOverride) {
                 }
                 throw requestError;
             }
-            const requestError = new WorkbenchApiRequestError(response.status, readResponseError(text) ?? `Request failed with status ${response.status}${response.statusText ? ` ${response.statusText}` : ""}.`, text);
+            const excerpt = readResponseError(text);
+            const requestError = new WorkbenchApiRequestError(response.status, `Request failed with status ${response.status}${response.statusText ? ` ${response.statusText}` : ""}${excerpt ? `: ${excerpt}` : ""}.`, text);
             lastError = requestError;
             if (canRetry && attempt < API_REQUEST_MAX_ATTEMPTS && isTransientApiRequestError(requestError)) {
                 await sleep(250 * attempt);
@@ -1707,12 +1817,22 @@ function readResponseError(text) {
         const parsed = JSON.parse(text);
         const record = asRecord(parsed);
         const error = record?.error ?? record?.message;
-        return typeof error === "string" && error.trim() ? error : null;
+        return typeof error === "string" && error.trim() ? oneLineExcerpt(error) : null;
     }
     catch {
-        return text.trim() || null;
+        if (/<(?:!doctype|html|head|body)\b/iu.test(text)) {
+            return null;
+        }
+        return oneLineExcerpt(text);
     }
 }
+function oneLineExcerpt(text) {
+    const line = text.replace(/\s+/gu, " ").trim();
+    if (!line) {
+        return null;
+    }
+    return line.length > 180 ? `${line.slice(0, 177)}...` : line;
+}
 function parseWorkbenchCloudErrorBody(text) {
     try {
         const record = asRecord(JSON.parse(text));
@@ -2168,6 +2288,17 @@ function intFlag(parsed, name) {
     }
     return parsedValue;
 }
+function portFlag(parsed, name) {
+    const value = stringFlag(parsed, name);
+    if (!value) {
+        return undefined;
+    }
+    const parsedValue = Number(value);
+    if (!Number.isInteger(parsedValue) || parsedValue < 0 || parsedValue > 65535) {
+        throw new WorkbenchUserError(`--${name} must be an integer between 0 and 65535.`);
+    }
+    return parsedValue;
+}
 function optionalPositional(parsed, index) {
     return parsed.positionals[index];
 }
@@ -2227,19 +2358,15 @@ function parsePublishVisibilityFlags(parsed) {
 }
 async function previewPublishWithDerivedRemote(parsed) {
     const root = path.resolve(dirFlag(parsed) ?? process.cwd());
-    const core = await coreOptions(parsed);
-    await listWorkbenchVersions(core);
     const reconciledSnapshot = await createWorkbenchReadOnlyInspectionSnapshot({ dir: root });
     const link = cloudRemoteLinkTargetFromRemotes(reconciledSnapshot.remotes);
-    if (link.existing) {
-        return undefined;
-    }
-    const remote = await derivePublishCloudRemote(parsed, "workbench publish", link.name);
+    const remote = stringFlag(parsed, "as") || !link.existing
+        ? await derivePublishCloudRemote(parsed, "workbench publish", link.name)
+        : link.existing;
     const requestedVersion = optionalPositional(parsed, 1);
-    const versionId = requestedVersion && requestedVersion !== "current"
-        ? requestedVersion
-        : reconciledSnapshot.status.currentVersionId ?? reconciledSnapshot.refs.current;
-    const version = reconciledSnapshot.versions.find((entry) => entry.id === versionId);
+    const version = requestedVersion && requestedVersion !== "current"
+        ? snapshotVersionByRef(reconciledSnapshot, requestedVersion)
+        : snapshotVersionByRef(reconciledSnapshot, reconciledSnapshot.status.currentVersionId ?? reconciledSnapshot.refs.current ?? "");
     if (!version) {
         throw new WorkbenchCodedError("version_not_found", `Version not found: ${requestedVersion ?? "current"}`, {
             remediation: "Run workbench log --versions.",
@@ -2390,7 +2517,7 @@ async function artifactIdsByRunId(core, runs) {
     return byRun;
 }
 function emitEvalFailure(runs, failedRuns, artifactIds, parsed, io) {
-    const nextCommands = evalFailureNextCommands(failedRuns);
+    const next = evalFailureNextCommand(failedRuns);
     if (parsed.flags.json === true) {
         io.stdout.write(`${JSON.stringify({
             schema: "workbench.cli.eval.v1",
@@ -2401,14 +2528,14 @@ function emitEvalFailure(runs, failedRuns, artifactIds, parsed, io) {
             evidenceSaved: true,
             runs: runs.map((run) => runFailureSummary(run, artifactIds.get(run.id) ?? [])),
             failedRuns: failedRuns.map((run) => runFailureSummary(run, artifactIds.get(run.id) ?? [])),
-            nextCommands,
+            next,
         }, null, 2)}\n`);
         return 1;
     }
     io.stdout.write([
         "Eval failed; evidence was saved.",
         ...failedRuns.map(formatRun),
-        ...(nextCommands[0] ? [`next: ${nextCommands[0]}`] : []),
+        ...(next ? [`next: ${next}`] : []),
     ].join("\n") + "\n");
     return 1;
 }
@@ -2420,7 +2547,7 @@ function runSummary(run, artifactIds) {
         versionId: run.versionId,
         skillName: run.skillName,
         agentName: run.agentName,
-        ...(run.score !== undefined ? { score: run.score } : {}),
+        ...(scoredRunValue(run) !== undefined ? { score: scoredRunValue(run) } : {}),
         ...(run.latencyMs !== undefined ? { latencyMs: run.latencyMs } : {}),
         ...(run.error ? { error: run.error } : {}),
         ...(run.jobIds ? { jobIds: run.jobIds } : {}),
@@ -2435,23 +2562,18 @@ function runFailureSummary(run, artifactIds) {
         skill: run.skillName,
         status: run.status,
         versionId: run.versionId,
-        ...(run.score !== undefined ? { score: run.score } : {}),
+        ...(scoredRunValue(run) !== undefined ? { score: scoredRunValue(run) } : {}),
         ...(run.error ? { error: run.error } : {}),
         traceIds: run.traceIds,
         artifactIds: [...artifactIds],
     };
 }
-function evalFailureNextCommands(failedRuns) {
+function evalFailureNextCommand(failedRuns) {
     const first = failedRuns[0];
     if (!first) {
-        return ["workbench log --runs"];
+        return "workbench log --runs";
     }
-    return [
-        `workbench show ${first.id}`,
-        `workbench show ${first.id}:stderr.log`,
-        `workbench case add ${first.id}`,
-        `workbench improve --agents ${first.agentName} --budget 1 -n 1`,
-    ];
+    return `workbench show ${displayRef(first.id)}`;
 }
 function output(value, parsed, io, text) {
     return emitResult(commandSchema(parsed), { result: value }, parsed, io, text);
@@ -2484,12 +2606,285 @@ async function workbenchCliAuthStatus() {
         })),
     };
 }
+function scoredRunValue(run) {
+    return run.status === "succeeded" && typeof run.score === "number" ? run.score : undefined;
+}
+function scoredJobValue(job) {
+    return job.status === "succeeded" && typeof job.score === "number" ? job.score : undefined;
+}
+function snapshotHasWorkflowCase(snapshot) {
+    const currentVersion = snapshotVersionByRef(snapshot, snapshot.status.currentVersionId ?? snapshot.refs.current ?? "");
+    const caseFiles = currentVersion?.files.filter((file) => file.kind === "text" &&
+        /^\.workbench\/cases\/[^/]+\/case\.ya?ml$/u.test(file.path)) ?? [];
+    return caseFiles.some((file) => file.kind === "text" && !/\n\s*smoke:\s*true(?:\s|$)/u.test(`\n${file.content}`));
+}
+function installHandleFromStatusRemote(remote) {
+    const publicationUrl = remote.publication.status === "published" ? remote.publication.installUrl : undefined;
+    const source = parseWorkbenchInstallSource(publicationUrl ?? remote.url);
+    return source ? `${source.owner}/${source.skill}` : publicationUrl ?? remote.url;
+}
+async function statusWithCausalNext(status, auth, core) {
+    if (!status.project.initialized) {
+        return status;
+    }
+    const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core).catch(() => null);
+    const lastRun = snapshot?.runs
+        .slice()
+        .sort((left, right) => right.createdAt.localeCompare(left.createdAt))[0];
+    if ((lastRun?.status === "running" || lastRun?.status === "failed" || lastRun?.status === "canceled") && lastRun.id) {
+        return { ...status, next: `workbench show ${displayRef(lastRun.id)}` };
+    }
+    const failedRemote = status.remotes.find((remote) => remote.sync.status === "error");
+    const hasWorkflowCase = snapshot ? snapshotHasWorkflowCase(snapshot) : false;
+    const hasScoredRun = snapshot?.runs.some((run) => scoredRunValue(run) !== undefined) ?? false;
+    const canPublish = hasWorkflowCase && hasScoredRun;
+    const cloudAuthMissing = auth.workbenchCloud.status !== "authenticated";
+    const cloudRemoteNeedsAuth = status.remotes.some((remote) => remote.kind === "workbench-cloud" &&
+        (remote.sync.status !== "up_to_date" || remote.publication.status === "unpublished"));
+    if (cloudAuthMissing && (canPublish || cloudRemoteNeedsAuth)) {
+        return { ...status, next: "workbench login" };
+    }
+    if (failedRemote) {
+        return { ...status, next: `workbench sync ${failedRemote.name}` };
+    }
+    if ((snapshot?.runs.length ?? status.runs.total) === 0) {
+        return { ...status, next: "workbench eval" };
+    }
+    if (!hasWorkflowCase) {
+        return { ...status, next: "edit .workbench/cases, then run workbench eval" };
+    }
+    const cloudRemote = status.remotes.find((remote) => remote.kind === "workbench-cloud");
+    if (canPublish && !cloudRemote) {
+        return { ...status, next: "workbench publish" };
+    }
+    const unpublishedCloudRemote = status.remotes.find((remote) => remote.kind === "workbench-cloud" &&
+        remote.publication.status === "unpublished" &&
+        remote.sync.status === "up_to_date");
+    if (unpublishedCloudRemote) {
+        return { ...status, next: "workbench publish" };
+    }
+    const publishedCloudRemote = status.remotes.find((remote) => remote.kind === "workbench-cloud" &&
+        remote.publication.status === "published" &&
+        Boolean(remote.publication.installUrl));
+    if (publishedCloudRemote) {
+        return { ...status, next: `workbench install ${installHandleFromStatusRemote(publishedCloudRemote)}` };
+    }
+    return {
+        ...status,
+        next: null,
+    };
+}
+function displayRef(id) {
+    const version = /^v_([0-9a-f]{8,})$/iu.exec(id);
+    if (version?.[1]) {
+        return version[1].slice(0, 8);
+    }
+    const separator = id.indexOf("_");
+    if (separator > 0 && separator < id.length - 1) {
+        const prefix = id.slice(0, separator);
+        const suffix = id.slice(separator + 1);
+        return `${prefix}_${suffix.slice(0, 8)}`;
+    }
+    return id.length > 8 ? id.slice(0, 8) : id;
+}
+function shortenCommandRefs(command) {
+    return command.replace(/\b(?:v_[0-9a-f]{8,}|(?:run|job|trace|artifact)_[a-z0-9_-]+)/giu, (match) => displayRef(match));
+}
+function snapshotVersionByRef(snapshot, ref) {
+    const requested = ref.trim();
+    const normalized = requested === "current" ? snapshot.refs.current ?? "" : requested;
+    if (!normalized) {
+        return undefined;
+    }
+    const candidates = snapshot.versions.filter((version) => snapshotVersionRefMatches(version, normalized));
+    if (candidates.length > 1) {
+        throw new WorkbenchCodedError("ref_ambiguous", `Version ref is ambiguous: ${ref}. Candidates: ${candidates.map((version) => displayRef(version.id)).join(", ")}.`, {
+            subject: { ref, candidates: candidates.map((version) => version.id) },
+            exitCode: 2,
+        });
+    }
+    return candidates[0];
+}
+function snapshotVersionRefMatches(version, ref) {
+    const withoutVersionPrefix = ref.startsWith("v_") ? ref.slice(2) : ref;
+    return version.id === ref ||
+        version.hash === ref ||
+        version.id.startsWith(ref) ||
+        version.hash.startsWith(ref) ||
+        version.hash.startsWith(withoutVersionPrefix) ||
+        version.id.startsWith(`v_${withoutVersionPrefix}`);
+}
+function snapshotObjectByRef(entries, ref, kind) {
+    const normalized = ref.trim();
+    if (!normalized) {
+        return undefined;
+    }
+    const candidates = entries.filter((entry) => objectRefMatches(entry.id, normalized));
+    if (candidates.length > 1) {
+        throw new WorkbenchCodedError("ref_ambiguous", `${capitalize(kind)} ref is ambiguous: ${ref}. Candidates: ${candidates.map((entry) => displayRef(entry.id)).slice(0, 8).join(", ")}.`, {
+            subject: { ref, candidates: candidates.map((entry) => entry.id).slice(0, 20) },
+            exitCode: 2,
+        });
+    }
+    return candidates[0];
+}
+function objectRefMatches(id, ref) {
+    if (id === ref || id.startsWith(ref)) {
+        return true;
+    }
+    const separator = id.indexOf("_");
+    return separator > 0 && id.slice(separator + 1).startsWith(ref);
+}
+function capitalize(value) {
+    return value.length > 0 ? `${value[0].toUpperCase()}${value.slice(1)}` : value;
+}
+function runOrJobEvidenceSelection(snapshot, ref) {
+    const run = snapshotObjectByRef(snapshot.runs, ref, "run");
+    const job = snapshotObjectByRef(snapshot.jobs, ref, "job");
+    if (run && job) {
+        throw new WorkbenchCodedError("ref_ambiguous", `Run/job ref is ambiguous: ${ref}. Candidates: ${displayRef(run.id)}, ${displayRef(job.id)}.`, {
+            subject: { ref, candidates: [run.id, job.id] },
+            exitCode: 2,
+        });
+    }
+    if (run) {
+        return {
+            run,
+            jobs: snapshot.jobs.filter((entry) => entry.runId === run.id),
+        };
+    }
+    return job ? { jobs: [job] } : { jobs: [] };
+}
+function evidenceFilesForRunOrJob(snapshot, ref) {
+    const selection = runOrJobEvidenceSelection(snapshot, ref);
+    if (!selection.run && selection.jobs.length === 0) {
+        return [];
+    }
+    const traceById = new Map(snapshot.traces.map((trace) => [trace.id, trace]));
+    const artifactById = new Map(snapshot.artifacts.map((artifact) => [artifact.id, artifact]));
+    const files = selection.jobs.flatMap((job) => [
+        ...job.artifactIds.flatMap((artifactId) => {
+            const artifact = artifactById.get(artifactId);
+            return artifact
+                ? artifact.files.filter(isUserFacingEvidenceFile).map((file) => evidenceFileWithPath(file, `cases/${evidencePathSegment(job.caseId)}/jobs/${evidencePathSegment(job.id)}/${file.path}`))
+                : [];
+        }),
+        ...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}`))
+                : [];
+        }),
+    ]);
+    const seen = new Set();
+    return files.filter((file) => {
+        if (seen.has(file.path)) {
+            return false;
+        }
+        seen.add(file.path);
+        return true;
+    });
+}
+function evidenceFileWithPath(file, filePath) {
+    return {
+        ...file,
+        path: filePath.replace(/\\/gu, "/").replace(/^\/+/u, ""),
+    };
+}
+function isUserFacingEvidenceFile(file) {
+    const normalized = file.path.replace(/\\/gu, "/").replace(/^\/+/u, "");
+    return normalized.split("/").every((segment) => segment !== ".workbench");
+}
+function evidencePathSegment(value) {
+    return value.replace(/[^A-Za-z0-9._-]+/gu, "-") || "_";
+}
+function formatRunOrJobEvidence(details, files) {
+    const detailLines = details.map(formatTraceDetail).filter(Boolean);
+    const fileLines = files.length > 0 ? ["Files:", ...files.map((file) => file.path)] : [];
+    return [...detailLines, ...fileLines].join("\n") || "No evidence.";
+}
+function evidenceDetailSummary(detail) {
+    return {
+        runId: detail.runId,
+        executions: detail.executions.map((execution) => ({
+            id: execution.id,
+            status: execution.status,
+            jobIds: execution.jobIds,
+            sessions: execution.sessions.map((session) => ({
+                label: session.label,
+            })),
+            trace: {
+                events: execution.trace.events.length,
+                spans: execution.trace.spans.length,
+                summaries: execution.trace.summaries.length,
+            },
+        })),
+    };
+}
+function manifestOnly(value) {
+    if (value === null || typeof value === "string" || typeof value === "number" || typeof value === "boolean") {
+        return value;
+    }
+    if (Array.isArray(value)) {
+        return value.map(manifestOnly);
+    }
+    if (!value || typeof value !== "object") {
+        return null;
+    }
+    const record = value;
+    if (typeof record.path === "string" && typeof record.content === "string") {
+        return fileSummary(record);
+    }
+    const out = {};
+    for (const [key, child] of Object.entries(record)) {
+        if (child === undefined) {
+            continue;
+        }
+        out[key] = manifestOnly(child);
+    }
+    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 `${entry.createdAt}\tversion\t${entry.id}\tfiles=${entry.fileCount}\t${entry.message}`;
+        return `${entry.createdAt}\tversion\t${displayRef(entry.id)}\tfiles=${entry.fileCount}\t${entry.message}`;
     }
     const score = entry.score === undefined ? "n/a" : entry.score.toFixed(3);
-    return `${entry.createdAt}\trun\t${entry.id}\t${entry.status}\tversion=${entry.versionId}\tskill=${entry.skillName}\tagent=${entry.agentName}\tscore=${score}`;
+    return `${entry.createdAt}\trun\t${displayRef(entry.id)}\t${entry.status}\tversion=${displayRef(entry.versionId)}\tskill=${entry.skillName}\tagent=${entry.agentName}\tscore=${score}`;
 }
 function splitShowRef(ref) {
     const index = ref.indexOf(":");
@@ -2500,18 +2895,14 @@ function splitShowRef(ref) {
 }
 async function fileForRunOrJobRef(core, objectRef, requestedPath) {
     const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
-    const run = snapshot.runs.find((entry) => entry.id === objectRef);
-    const job = snapshot.jobs.find((entry) => entry.id === objectRef);
-    if (!run && !job) {
+    const selection = runOrJobEvidenceSelection(snapshot, objectRef);
+    if (!selection.run && selection.jobs.length === 0) {
         return null;
     }
-    const traceIds = run?.traceIds ?? job?.traceIds ?? [];
-    const traces = snapshot.traces.filter((trace) => traceIds.includes(trace.id));
-    for (const trace of traces) {
-        const file = findShowFile(trace.files, requestedPath);
-        if (file) {
-            return file;
-        }
+    const files = evidenceFilesForRunOrJob(snapshot, objectRef);
+    const file = findShowFile(files, requestedPath, objectRef);
+    if (file) {
+        return file;
     }
     throw new WorkbenchCodedError("ref_not_found", `File not found in ${objectRef}: ${requestedPath}`, {
         remediation: `Run workbench show ${objectRef}.`,
@@ -2520,12 +2911,8 @@ async function fileForRunOrJobRef(core, objectRef, requestedPath) {
     });
 }
 function evidenceDetailsForRunOrJob(snapshot, ref) {
-    const run = snapshot.runs.find((entry) => entry.id === ref);
-    const job = snapshot.jobs.find((entry) => entry.id === ref);
-    const jobs = run
-        ? snapshot.jobs.filter((entry) => entry.runId === run.id)
-        : job ? [job] : [];
-    return jobs.flatMap((entry) => {
+    const selection = runOrJobEvidenceSelection(snapshot, ref);
+    return selection.jobs.flatMap((entry) => {
         const detail = workbenchJobEvidenceForSnapshot(snapshot, {
             runId: entry.runId,
             jobId: entry.id,
@@ -2536,12 +2923,72 @@ function evidenceDetailsForRunOrJob(snapshot, ref) {
         execution.trace.events.length > 0 ||
         execution.trace.summaries.length > 0));
 }
-function findShowFile(files, requestedPath) {
+function findShowFile(files, requestedPath, objectRef) {
     const normalized = requestedPath.replace(/\\/gu, "/");
-    return files.find((file) => file.path === normalized) ??
-        files.find((file) => file.path.endsWith(`/${normalized}`)) ??
-        files.find((file) => path.basename(file.path) === normalized) ??
-        null;
+    const exact = files.filter((file) => file.path === normalized);
+    if (exact.length === 1) {
+        return exact[0];
+    }
+    const exactEquivalent = singleEquivalentShowFile(exact);
+    if (exactEquivalent) {
+        return exactEquivalent;
+    }
+    if (exact.length > 1) {
+        throw ambiguousShowPath(objectRef, requestedPath, exact);
+    }
+    const suffixCandidates = files.filter((file) => file.path.endsWith(`/${normalized}`) || path.basename(file.path) === normalized);
+    if (suffixCandidates.length === 0) {
+        return null;
+    }
+    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];
+    }
+    const equivalentCandidate = singleEquivalentShowFile(candidates);
+    if (equivalentCandidate) {
+        return equivalentCandidate;
+    }
+    if (candidates.length === 0 && suffixCandidates.length === 1) {
+        return suffixCandidates[0];
+    }
+    const equivalentSuffixCandidate = singleEquivalentShowFile(suffixCandidates);
+    if (equivalentSuffixCandidate) {
+        return equivalentSuffixCandidate;
+    }
+    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;
+    }
+    const first = files[0];
+    return files.every((file) => file.kind === first.kind && file.encoding === first.encoding && file.content === first.content)
+        ? first
+        : null;
+}
+function ambiguousShowPath(objectRef, requestedPath, candidates) {
+    const candidatePaths = candidates.map((file) => file.path);
+    return new WorkbenchCodedError("ref_ambiguous", `File path is ambiguous in ${objectRef}: ${requestedPath}. Candidates: ${candidatePaths.join(", ")}.`, {
+        remediation: `Run workbench show ${objectRef}.`,
+        subject: { ref: objectRef, path: requestedPath, candidates: candidatePaths },
+        exitCode: 2,
+    });
 }
 function fileListing(kind, id, files) {
     return {
@@ -2552,17 +2999,16 @@ function fileListing(kind, id, files) {
     };
 }
 function formatFileListing(kind, id, files) {
-    return [`${kind}\t${id}\tfiles=${files.length}`, ...files.map((file) => file.path)].join("\n");
+    return [`${kind}\t${displayRef(id)}\tfiles=${files.length}`, ...files.map((file) => file.path)].join("\n");
 }
 async function traceIdForCaseSource(core, ref) {
     const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
-    const trace = snapshot.traces.find((entry) => entry.id === ref);
+    const trace = snapshotObjectByRef(snapshot.traces, ref, "trace");
     if (trace) {
         return trace.id;
     }
-    const run = snapshot.runs.find((entry) => entry.id === ref);
-    const job = snapshot.jobs.find((entry) => entry.id === ref);
-    const traceId = run?.traceIds[0] ?? job?.traceIds[0];
+    const selection = runOrJobEvidenceSelection(snapshot, ref);
+    const traceId = selection.run?.traceIds[0] ?? selection.jobs[0]?.traceIds[0];
     if (traceId) {
         return traceId;
     }
@@ -2572,43 +3018,85 @@ async function traceIdForCaseSource(core, ref) {
         exitCode: 1,
     });
 }
+async function evalCoverageSummaries(core, runs) {
+    const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
+    const jobsByRun = new Map();
+    for (const job of snapshot.jobs) {
+        const existing = jobsByRun.get(job.runId) ?? [];
+        existing.push(job);
+        jobsByRun.set(job.runId, existing);
+    }
+    return runs.map((run) => {
+        const jobs = jobsByRun.get(run.id) ?? [];
+        const cases = new Set(jobs.map((job) => job.caseId));
+        const samples = new Set(jobs.map((job) => `${job.caseId}\0${job.sample}`));
+        return {
+            runId: run.id,
+            cases: cases.size,
+            samples: samples.size,
+            jobs: jobs.length,
+            succeeded: jobs.filter((job) => job.status === "succeeded").length,
+            failed: jobs.filter((job) => job.status === "failed" || job.status === "canceled").length,
+        };
+    });
+}
+function formatEvalCoverage(coverage) {
+    return [
+        `coverage cases=${coverage.cases}`,
+        `samples=${coverage.samples}`,
+        `jobs=${coverage.jobs}`,
+        coverage.failed > 0 ? `failed=${coverage.failed}` : undefined,
+    ].filter(Boolean).join(" ");
+}
 async function evalDeltas(core, runs) {
     const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
     return runs.map((run) => {
+        const score = scoredRunValue(run);
         const previous = snapshot.runs
             .filter((candidate) => candidate.id !== run.id &&
             candidate.skillName === run.skillName &&
             candidate.agentName === run.agentName &&
-            typeof candidate.score === "number" &&
+            scoredRunValue(candidate) !== undefined &&
             candidate.createdAt < run.createdAt)
             .sort((left, right) => right.createdAt.localeCompare(left.createdAt))[0];
+        const previousScore = previous ? scoredRunValue(previous) : undefined;
         return {
             runId: run.id,
             versionId: run.versionId,
             skillName: run.skillName,
             agentName: run.agentName,
-            ...(run.score !== undefined ? { score: run.score } : {}),
-            ...(previous?.score !== undefined ? { previousScore: previous.score } : {}),
-            ...(run.score !== undefined && previous?.score !== undefined ? { delta: run.score - previous.score } : {}),
+            ...(score !== undefined ? { score } : {}),
+            ...(previousScore !== undefined ? { previousScore } : {}),
+            ...(score !== undefined && previousScore !== undefined ? { delta: score - previousScore } : {}),
         };
     });
 }
 function formatEvalDelta(delta) {
-    const score = delta.score === undefined ? "n/a" : delta.score.toFixed(3);
+    if (delta.score === undefined) {
+        return "";
+    }
+    const score = delta.score.toFixed(3);
     if (delta.previousScore === undefined || delta.delta === undefined) {
-        return `${delta.skillName} ${delta.versionId} ${score} (was n/a)`;
+        return `${delta.skillName} ${displayRef(delta.versionId)} ${score}`;
     }
     const sign = delta.delta >= 0 ? "+" : "";
-    return `${delta.skillName} ${delta.versionId} ${score} (was ${delta.previousScore.toFixed(3)}, ${sign}${delta.delta.toFixed(3)})`;
+    return `${delta.skillName} ${displayRef(delta.versionId)} ${score} (was ${delta.previousScore.toFixed(3)}, ${sign}${delta.delta.toFixed(3)})`;
 }
-function evalSuccessNextCommands(runs) {
-    return runs.length > 0 ? ["workbench publish"] : ["workbench eval"];
+async function evalSuccessNextCommand(core, runs) {
+    if (runs.length === 0) {
+        return "workbench eval";
+    }
+    if (!runs.some((run) => scoredRunValue(run) !== undefined)) {
+        return "edit .workbench/cases, then run workbench eval";
+    }
+    const snapshot = await createWorkbenchReadOnlyInspectionSnapshot(core);
+    return snapshotHasWorkflowCase(snapshot) ? "workbench publish" : "edit .workbench/cases, then run workbench eval";
 }
 function formatStatusSnapshot(status) {
     const lines = [
         `Root: ${status.project.root}`,
         `Initialized: ${status.project.initialized ? "yes" : "no"}`,
-        ...(status.project.currentVersionId ? [`Current version: ${status.project.currentVersionId}`] : []),
+        ...(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}`] : []),
         `Runs: ${status.runs.total}${status.runs.lastStatus ? ` (last ${status.runs.lastStatus})` : ""}`,
@@ -2618,7 +3106,7 @@ function formatStatusSnapshot(status) {
                     ? [
                         "publication=published",
                         remote.publication.visibility ? `visibility=${remote.publication.visibility}` : undefined,
-                        remote.publication.versionId ? `version=${remote.publication.versionId}` : undefined,
+                        remote.publication.versionId ? `version=${displayRef(remote.publication.versionId)}` : undefined,
                         remote.publication.installUrl ? `install=${remote.publication.installUrl}` : undefined,
                         remote.publication.pinnedInstallUrl ? `pinned=${remote.publication.pinnedInstallUrl}` : undefined,
                     ].filter(Boolean).join("\t")
@@ -2629,17 +3117,16 @@ function formatStatusSnapshot(status) {
                         ? [
                             `    error[${remote.sync.lastError.code}]: ${remote.sync.lastError.message}`,
                             ...(remote.sync.lastAttemptAt ? [`    last attempt: ${remote.sync.lastAttemptAt}`] : []),
-                            ...(remote.sync.nextCommand ? [`    next: ${remote.sync.nextCommand}`] : []),
                         ]
                         : []),
                 ];
             })] : ["Remotes: none"]),
-        ...(status.next[0] ? [`next: ${status.next[0]}`] : []),
+        ...(status.next ? [`next: ${shortenCommandRefs(status.next)}`] : []),
     ];
     return lines.join("\n");
 }
 function formatVersion(version) {
-    return `${version.id}\t${version.hash.slice(0, 12)}\t${version.message}`;
+    return `${displayRef(version.id)}\t${version.hash.slice(0, 12)}\t${version.message}`;
 }
 function versionSummary(version) {
     return {
@@ -2655,50 +3142,52 @@ function formatAgent(agent) {
     return `${agent.name}\t${agent.adapter}${agent.model ? `\t${agent.model}` : ""}`;
 }
 function formatRun(run) {
-    const score = run.score === undefined ? "n/a" : run.score.toFixed(3);
+    const scoreValue = scoredRunValue(run);
+    const score = scoreValue === undefined ? "n/a" : scoreValue.toFixed(3);
     const latency = run.latencyMs === undefined ? "n/a" : `${run.latencyMs}ms`;
-    return `${run.id}\t${run.kind}\t${run.status}\tversion=${run.versionId}\tskill=${run.skillName}\tagent=${run.agentName}\tscore=${score}\tlatency=${latency}`;
+    return `${displayRef(run.id)}\t${run.kind}\t${run.status}\tversion=${displayRef(run.versionId)}\tskill=${run.skillName}\tagent=${run.agentName}\tscore=${score}\tlatency=${latency}`;
 }
 function formatImproveResult(result) {
     return [
-        `Improved ${result.version.parentIds[0] ?? "current"} -> ${result.version.id}. ${formatRun(result.run)}`,
+        `Improved ${result.version.parentIds[0] ? displayRef(result.version.parentIds[0]) : "current"} -> ${displayRef(result.version.id)}. ${formatRun(result.run)}`,
         result.switched
             ? "Switched to improved version."
             : `Did not switch: ${result.promotionReason}`,
     ].join("\n");
 }
 function formatJob(job) {
-    const score = job.score === undefined ? "n/a" : job.score.toFixed(3);
+    const scoreValue = scoredJobValue(job);
+    const score = scoreValue === undefined ? "n/a" : scoreValue.toFixed(3);
     const duration = job.durationMs === undefined ? "n/a" : `${job.durationMs}ms`;
-    return `${job.id}\trun=${job.runId}\tcase=${job.caseId}\tsample=${job.sample}\t${job.status}\tscore=${score}\tduration=${duration}`;
+    return `${displayRef(job.id)}\trun=${displayRef(job.runId)}\tcase=${job.caseId}\tsample=${job.sample}\t${job.status}\tscore=${score}\tduration=${duration}`;
 }
 function formatComparison(comparison) {
     const lines = ["version\tskill\tagent\tstatus\tscore\tcost\tlatency\trun"];
     for (const cell of comparison.cells) {
         lines.push([
-            cell.versionId,
+            displayRef(cell.versionId),
             cell.skillName,
             `${cell.agentName}@${shortObjectId(cell.agentHash)}`,
             cell.status ?? "not-run",
             cell.score === undefined ? "n/a" : cell.score.toFixed(3),
             cell.costUsd === undefined ? "n/a" : `$${cell.costUsd.toFixed(4)}`,
             cell.latencyMs === undefined ? "n/a" : `${cell.latencyMs}ms`,
-            cell.runId ?? "n/a",
+            cell.runId ? displayRef(cell.runId) : "n/a",
         ].join("\t"));
     }
     return lines.join("\n");
 }
 function shortObjectId(id) {
-    return id.length > 12 ? id.slice(0, 12) : id;
+    return id.length > 8 ? id.slice(0, 8) : id;
 }
 function formatTrace(trace) {
     const result = asRecord(trace.result);
     const status = typeof result?.status === "string" ? result.status : undefined;
-    const score = typeof result?.score === "number" ? result.score.toFixed(3) : undefined;
+    const score = status === "succeeded" && typeof result?.score === "number" ? result.score.toFixed(3) : undefined;
     const error = typeof result?.error === "string" ? result.error.split(/\r?\n/u)[0] : undefined;
     const files = trace.files.slice(0, 5).map((file) => file.path).join(",");
     return [
-        `${trace.id}\trun=${trace.runId}\tjob=${trace.jobId ?? "n/a"}\tversion=${trace.versionId}\tskill=${trace.skillName}\tagent=${trace.agentName}`,
+        `${displayRef(trace.id)}\trun=${displayRef(trace.runId)}\tjob=${trace.jobId ? displayRef(trace.jobId) : "n/a"}\tversion=${displayRef(trace.versionId)}\tskill=${trace.skillName}\tagent=${trace.agentName}`,
         status ? `status=${status}` : undefined,
         score ? `score=${score}` : undefined,
         error ? `error=${error}` : undefined,
@@ -2707,6 +3196,7 @@ function formatTrace(trace) {
 }
 function traceSummary(trace) {
     const result = asRecord(trace.result);
+    const status = typeof result?.status === "string" ? result.status : undefined;
     return {
         id: trace.id,
         runId: trace.runId,
@@ -2715,8 +3205,8 @@ function traceSummary(trace) {
         skillName: trace.skillName,
         agentName: trace.agentName,
         createdAt: trace.createdAt,
-        ...(typeof result?.status === "string" ? { status: result.status } : {}),
-        ...(typeof result?.score === "number" ? { score: result.score } : {}),
+        ...(status ? { status } : {}),
+        ...(status === "succeeded" && typeof result?.score === "number" ? { score: result.score } : {}),
         ...(typeof result?.error === "string" ? { error: singleLine(result.error) } : {}),
         fileCount: trace.files.length,
         files: trace.files.map(fileSummary),
@@ -2726,7 +3216,7 @@ function formatTraceDetail(detail) {
     return detail.executions.map((execution) => {
         const sessionLabels = execution.sessions.map((session) => session.label).join(",");
         return [
-            `${execution.id}\trun=${detail.runId}\tjobs=${execution.jobIds.join(",")}\tstatus=${execution.status}`,
+            `${execution.id}\trun=${displayRef(detail.runId)}\tjobs=${execution.jobIds.map(displayRef).join(",")}\tstatus=${execution.status}`,
             `events=${execution.trace.events.length}`,
             `spans=${execution.trace.spans.length}`,
             `summaries=${execution.trace.summaries.length}`,
@@ -2735,7 +3225,7 @@ function formatTraceDetail(detail) {
     }).join("\n");
 }
 function formatArtifact(artifact) {
-    return `${artifact.id}\trun=${artifact.runId}\tjob=${artifact.jobId}\t${artifact.kind}\tfiles=${artifact.files.length}`;
+    return `${displayRef(artifact.id)}\trun=${displayRef(artifact.runId)}\tjob=${displayRef(artifact.jobId)}\t${artifact.kind}\tfiles=${artifact.files.length}`;
 }
 function artifactSummary(artifact) {
     return {