@workbench-ai/workbench 0.0.65 → 0.0.67

package/dist/local-archive.js

@@ -1,6 +1,6 @@
 import { promises as fs } from "node:fs";
 import path from "node:path";
-import { buildWorkbenchTraceSessionsFromFiles, candidateRecordWithoutDerivedFields, compactWorkbenchRuntimeJobForExchange, mergeWorkbenchRuntimeCandidateForExchange, sanitizeWorkbenchRuntimeCandidateForExchange, sanitizeWorkbenchRuntimeJobForExchange, selectExecutionOutputFilesForInspection, workbenchRuntimeExplicitActiveId, workbenchRuntimeBundleStats, workbenchRuntimeCandidateIdentityForExchange, workbenchSurfaceFilesEqualForExchange, } from "@workbench-ai/workbench-core";
+import { buildWorkbenchTraceSessionsFromFiles, candidateRecordWithoutDerivedFields, compactWorkbenchRuntimeJobForExchange, mergeWorkbenchRuntimeCandidateForExchange, sanitizeWorkbenchRuntimeCandidateForExchange, sanitizeWorkbenchRuntimeJobForExchange, selectExecutionOutputFilesForInspection, isSurfaceSnapshotFile, jsonRecord, normalizeRelativePath, readSurfaceFiles, workbenchRuntimeBundleStats, workbenchRuntimeCandidateIdentityForExchange, workbenchRuntimeProjectedActiveId, workbenchSurfaceFilesEqualForExchange, writeSurfaceFiles, } from "@workbench-ai/workbench-core";
 const RUNTIME_DIR = ".workbench/runtime";
 const CANDIDATE_RECORDS_DIR = "candidates";
 export function localRuntimeDir(workspace) {
@@ -73,12 +73,17 @@ export async function saveLocalJobs(workspace, jobs) {
 }
 export async function exportLocalRuntimeBundle(workspace, options = {}) {
     const snapshot = await loadLocalArchive(workspace);
-    const jobs = (await readLocalJobs(workspace)).map(compactWorkbenchRuntimeJobForExchange);
+    const archivedJobs = await readLocalJobs(workspace);
+    const jobs = archivedJobs.map(compactWorkbenchRuntimeJobForExchange);
+    const executionFiles = (await Promise.all(archivedJobs.map(async (job) => ({
+        jobId: job.id,
+        files: await readLocalExecutionFiles(workspace, job.id),
+    })))).filter((group) => group.files.length > 0);
     const activeId = options.currentBenchmarkFingerprint
-        ? workbenchRuntimeExplicitActiveId({
+        ? workbenchRuntimeProjectedActiveId({
             candidates: snapshot.candidates,
+            evaluations: snapshot.evaluations,
             runs: snapshot.runs,
-            preferredActiveId: snapshot.activeId,
             benchmarkFingerprint: options.currentBenchmarkFingerprint,
         })
         : snapshot.activeId;
@@ -93,7 +98,7 @@ export async function exportLocalRuntimeBundle(workspace, options = {}) {
         evaluations: snapshot.evaluations.map((evaluation) => ({ ...evaluation })),
         runs: snapshot.runs.map((run) => ({ ...run })),
         jobs,
-        executionFiles: [],
+        executionFiles,
         events: snapshot.events.map((event) => ({ ...event })),
     };
 }
@@ -163,10 +168,10 @@ export async function importLocalRuntimeBundle(workspace, bundle, currentBenchma
         changed ||= didChange;
     }, runtimeJobsEqualForExchange).sort((left, right) => (left.startedAt ?? left.createdAt).localeCompare(right.startedAt ?? right.createdAt) ||
         left.id.localeCompare(right.id));
-    const activeId = workbenchRuntimeExplicitActiveId({
+    const activeId = workbenchRuntimeProjectedActiveId({
         candidates,
+        evaluations,
         runs,
-        preferredActiveId: bundle.activeId ?? null,
         benchmarkFingerprint: currentBenchmarkFingerprint,
     });
     if (activeId !== snapshot.activeId) {
@@ -682,14 +687,6 @@ function completedJobOutputFiles(job) {
     }
     return output.files.filter(isSurfaceSnapshotFile).map((file) => ({ ...file }));
 }
-function isSurfaceSnapshotFile(value) {
-    const record = jsonRecord(value);
-    return (typeof record.path === "string" &&
-        (record.kind === "text" || record.kind === "binary") &&
-        (record.encoding === "utf8" || record.encoding === "base64") &&
-        typeof record.content === "string" &&
-        typeof record.executable === "boolean");
-}
 function readExecutionPurpose(job) {
     const input = jsonRecord(job.input);
     return stringValue(jsonRecord(input.execution).purpose);
@@ -751,15 +748,9 @@ function traceEvent(args) {
 }
 function traceUsageSummary(value) {
     const record = jsonRecord(value);
-    const usage = Object.keys(jsonRecord(record.total)).length > 0
-        ? jsonRecord(record.total)
-        : Object.keys(jsonRecord(record.improver)).length > 0
-            ? jsonRecord(record.improver)
-            : Object.keys(jsonRecord(record.runner)).length > 0
-                ? jsonRecord(record.runner)
-                : Object.keys(jsonRecord(record.engine)).length > 0
-                    ? jsonRecord(record.engine)
-                    : record;
+    const usage = ["total", "improver", "runner", "engine"]
+        .map((key) => jsonRecord(record[key]))
+        .find((entry) => Object.keys(entry).length > 0) ?? record;
     if (Object.keys(usage).length === 0) {
         return null;
     }
@@ -779,11 +770,6 @@ function traceUsageSummary(value) {
         pricing_source: stringValue(usage.pricingSource) ?? stringValue(usage.pricing_source),
     };
 }
-function jsonRecord(value) {
-    return value && typeof value === "object" && !Array.isArray(value)
-        ? value
-        : {};
-}
 function stringValue(value) {
     return typeof value === "string" && value.length > 0 ? value : null;
 }
@@ -850,70 +836,3 @@ async function writeJson(filePath, value) {
     await fs.mkdir(path.dirname(filePath), { recursive: true });
     await fs.writeFile(filePath, `${JSON.stringify(value, null, 2)}\n`);
 }
-async function writeSurfaceFiles(root, files) {
-    await fs.mkdir(root, { recursive: true });
-    for (const file of files) {
-        const target = path.join(root, normalizeRelativePath(file.path));
-        await fs.mkdir(path.dirname(target), { recursive: true });
-        const body = file.encoding === "base64" ? Buffer.from(file.content, "base64") : Buffer.from(file.content, "utf8");
-        await fs.writeFile(target, body);
-        if (file.executable) {
-            await fs.chmod(target, 0o755).catch(() => undefined);
-        }
-    }
-}
-async function readSurfaceFiles(root) {
-    const decoder = new TextDecoder("utf-8", { fatal: true });
-    const files = [];
-    async function walk(directory) {
-        const entries = await fs.readdir(directory, { withFileTypes: true }).catch(() => []);
-        for (const entry of entries) {
-            const absolutePath = path.join(directory, entry.name);
-            if (entry.isDirectory()) {
-                await walk(absolutePath);
-                continue;
-            }
-            if (!entry.isFile()) {
-                continue;
-            }
-            const body = await fs.readFile(absolutePath);
-            const relativePath = normalizeRelativePath(path.relative(root, absolutePath).replace(/\\/gu, "/"));
-            const stats = await fs.stat(absolutePath);
-            const content = encodeContent(body, decoder);
-            files.push({
-                path: relativePath,
-                kind: content.encoding === "base64" ? "binary" : "text",
-                encoding: content.encoding,
-                content: content.content,
-                executable: (stats.mode & 0o111) !== 0,
-            });
-        }
-    }
-    await walk(root);
-    return files.sort((left, right) => left.path.localeCompare(right.path));
-}
-function encodeContent(body, decoder) {
-    try {
-        return {
-            encoding: "utf8",
-            content: decoder.decode(body),
-        };
-    }
-    catch {
-        return {
-            encoding: "base64",
-            content: body.toString("base64"),
-        };
-    }
-}
-function normalizeRelativePath(filePath) {
-    const normalized = filePath.replace(/\\/gu, "/").replace(/^\/+/u, "");
-    if (!normalized || normalized.includes("\0")) {
-        throw new Error("File paths must be non-empty relative paths.");
-    }
-    const parts = normalized.split("/");
-    if (parts.some((part) => part === ".." || part === "." || part === "")) {
-        throw new Error(`Unsafe relative file path: ${filePath}`);
-    }
-    return normalized;
-}

package/dist/local-inspection.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"local-inspection.d.ts","sourceRoot":"","sources":["../src/local-inspection.ts"],"names":[],"mappings":"AAEA,OAAO,EAaL,KAAK,mBAAmB,EAGzB,MAAM,8BAA8B,CAAC;AAetC,OAAO,EAGL,KAAK,kBAAkB,EAExB,MAAM,qBAAqB,CAAC;AAgB7B,MAAM,WAAW,+BAA+B;IAC9C,SAAS,EAAE,MAAM,CAAC;IAClB,iBAAiB,CAAC,EAAE,MAAM,OAAO,CAAC,kBAAkB,CAAC,CAAC;CACvD;AAED,wBAAgB,8BAA8B,CAC5C,OAAO,EAAE,+BAA+B,GACvC,mBAAmB,CAqCrB;AAED,wBAAgB,8BAA8B,CAC5C,SAAS,EAAE,MAAM,GAChB,MAAM,OAAO,CAAC,kBAAkB,CAAC,CAiBnC"}
+{"version":3,"file":"local-inspection.d.ts","sourceRoot":"","sources":["../src/local-inspection.ts"],"names":[],"mappings":"AAEA,OAAO,EAgBL,KAAK,mBAAmB,EAIzB,MAAM,8BAA8B,CAAC;AAetC,OAAO,EAGL,KAAK,kBAAkB,EAExB,MAAM,qBAAqB,CAAC;AAgB7B,MAAM,WAAW,+BAA+B;IAC9C,SAAS,EAAE,MAAM,CAAC;IAClB,iBAAiB,CAAC,EAAE,MAAM,OAAO,CAAC,kBAAkB,CAAC,CAAC;CACvD;AAED,wBAAgB,8BAA8B,CAC5C,OAAO,EAAE,+BAA+B,GACvC,mBAAmB,CA2DrB;AAkBD,wBAAgB,8BAA8B,CAC5C,SAAS,EAAE,MAAM,GAChB,MAAM,OAAO,CAAC,kBAAkB,CAAC,CAiBnC"}

package/dist/local-inspection.js

@@ -1,5 +1,5 @@
 import path from "node:path";
-import { WorkbenchInspectionError, candidateRecordWithoutDerivedFields, candidateSummaryFromRecord, createWorkbenchInspection, loadAuthoredWorkbenchSourceDocument, traceSessionLabel, } from "@workbench-ai/workbench-core";
+import { WorkbenchInspectionError, candidateRecordWithoutDerivedFields, candidateSummaryFromRecord, createCandidateFilePreview, createWorkbenchInspection, loadAuthoredWorkbenchSourceDocument, selectedFilePath, summarizeCandidateFiles, traceSessionLabel, } from "@workbench-ai/workbench-core";
 import { localBenchmarkFingerprint } from "./benchmark-fingerprint.js";
 import { loadLocalArchiveIndex, readLocalCandidateFilesForId, readLocalCandidateRecord, readLocalEvaluationRecord, readLocalExecutionFiles, readLocalJobInRun, readLocalRunJobs, readLocalRunRecord, } from "./local-archive.js";
 import { readLocalAuthoredProjectSource, readLocalProjectSource, WORKBENCH_BENCHMARK_FILE, } from "./project-source.js";
@@ -14,14 +14,22 @@ export function createLocalWorkbenchInspection(options) {
         projectId: "local",
         snapshot: () => localBenchmarkSnapshot(context),
         spec: (input) => localSpecDocument(context, input.fingerprint),
-        sourceFiles: (input) => localBenchmarkMountedFiles(context, input.fingerprint),
+        sourceFiles: async (input) => {
+            const files = await localBenchmarkMountedFiles(context, input.fingerprint);
+            return summarizeCandidateFiles(files, files.map((file) => file.path));
+        },
+        sourceFileSurface: async (input) => {
+            const files = await localBenchmarkMountedFiles(context, input.fingerprint);
+            return localFileSurface(files, files.map((file) => file.path), input.path, input.view);
+        },
         candidate: (input) => readCandidateForInspection(context.workspace, input.id),
         candidateFiles: async (input) => {
             const candidate = await readCandidateForInspection(context.workspace, input.id);
-            return {
-                files: await readCandidateFilesForInspection(context.workspace, input.id),
-                changedPaths: candidate.fileChanges,
-            };
+            return summarizeCandidateFiles(await readCandidateFilesForInspection(context.workspace, input.id), candidate.fileChanges);
+        },
+        candidateFileSurface: async (input) => {
+            const candidate = await readCandidateForInspection(context.workspace, input.id);
+            return localFileSurface(await readCandidateFilesForInspection(context.workspace, input.id), candidate.fileChanges, input.path, input.view);
         },
         evaluation: (input) => readEvaluationForInspection(context.workspace, input.id),
         run: async (input) => {
@@ -34,12 +42,29 @@ export function createLocalWorkbenchInspection(options) {
             };
         },
         jobInRun: (input) => readExecutionJobForRun(context.workspace, input.runId, input.jobId),
-        executionFiles: (input) => readExecutionFilesForRun(context.workspace, input.runId, input.jobId),
+        executionFiles: async (input) => {
+            const files = await readExecutionFilesForRun(context.workspace, input.runId, input.jobId);
+            return summarizeCandidateFiles(files, files.map((file) => file.path));
+        },
+        executionFileSurface: async (input) => {
+            const files = await readExecutionFilesForRun(context.workspace, input.runId, input.jobId);
+            return localFileSurface(files, files.map((file) => file.path), input.path, input.view);
+        },
         traceForJob: readLocalAggregateTrace,
         traceSessionsForJob: readLocalTraceSessions,
     };
     return createWorkbenchInspection(backend);
 }
+function localFileSurface(files, changedPaths, path, view = "rendered") {
+    const summaries = summarizeCandidateFiles(files, changedPaths);
+    const previewPath = selectedFilePath(path, summaries);
+    return {
+        files: summaries,
+        preview: previewPath
+            ? createCandidateFilePreview({ files, path: previewPath, view })
+            : null,
+    };
+}
 export function createLocalProjectSourceReader(workspace) {
     const resolvedWorkspace = path.resolve(workspace);
     let cached = null;
@@ -90,9 +115,20 @@ async function localSpecDocument(context, benchmarkFingerprint) {
         currentFingerprint &&
         requestedFingerprint !== currentFingerprint) {
         const snapshot = await loadLocalArchiveIndex(workspace);
-        const document = localHistoricalBenchmarkDocument(snapshot, requestedFingerprint);
-        if (document) {
-            return document;
+        const source = localHistoricalBenchmarkSource(snapshot, requestedFingerprint);
+        if (source) {
+            return loadAuthoredWorkbenchSourceDocument({
+                sourceYaml: source.sourceYaml,
+                path: WORKBENCH_BENCHMARK_FILE,
+                sourceFiles: [{
+                        path: WORKBENCH_BENCHMARK_FILE,
+                        kind: "text",
+                        encoding: "utf8",
+                        content: source.sourceYaml,
+                        executable: false,
+                    }],
+                cases: source.engineResolveFiles,
+            });
         }
         throw new WorkbenchInspectionError(`Benchmark version not found: ${requestedFingerprint}`, { status: 404 });
     }
@@ -116,12 +152,16 @@ async function localBenchmarkMountedFiles(context, benchmarkFingerprint) {
         currentFingerprint &&
         requestedFingerprint !== currentFingerprint) {
         const snapshot = await loadLocalArchiveIndex(workspace);
-        return localHistoricalBenchmarkFiles(snapshot, requestedFingerprint);
+        const source = localHistoricalBenchmarkSource(snapshot, requestedFingerprint);
+        if (source) {
+            return source.engineResolveFiles.map((file) => ({ ...file }));
+        }
+        throw new WorkbenchInspectionError(`Benchmark version not found: ${requestedFingerprint}`, { status: 404 });
     }
     return inspectableEngineCaseFiles(projectSource.engineCases);
 }
 function publicLocalRunSummary(run) {
-    const { executionFingerprint: _executionFingerprint, ...summary } = run;
+    const { executionFingerprint: _executionFingerprint, input: _input, ...summary } = run;
     return summary;
 }
 async function readCurrentBenchmarkFingerprint(context) {
@@ -147,21 +187,17 @@ function caseSummaryFilesFromEngineCases(engineCases, files) {
         })),
     ];
 }
-function localHistoricalBenchmarkDocument(snapshot, benchmarkFingerprint) {
-    const candidate = snapshot.candidates.find((entry) => entry.benchmarkFingerprint === benchmarkFingerprint && readBenchmarkSourceMetadata(entry));
-    const source = candidate ? readBenchmarkSourceMetadata(candidate) : null;
-    if (!source?.sourceYaml) {
-        return null;
+function localHistoricalBenchmarkSource(snapshot, benchmarkFingerprint) {
+    for (const run of snapshot.runs) {
+        if (run.benchmarkFingerprint !== benchmarkFingerprint) {
+            continue;
+        }
+        const source = readRunSourceInput(run);
+        if (source) {
+            return source;
+        }
     }
-    return loadAuthoredWorkbenchSourceDocument({
-        sourceYaml: source.sourceYaml,
-        path: WORKBENCH_BENCHMARK_FILE,
-        sourceFiles: source.files,
-        cases: localHistoricalBenchmarkFiles(snapshot, benchmarkFingerprint),
-    });
-}
-function localHistoricalBenchmarkFiles(_snapshot, _benchmarkFingerprint) {
-    return [];
+    return null;
 }
 function inspectableEngineCaseFiles(engineCases) {
     return engineCases.flatMap((bundle) => engineCaseFiles(bundle).map((file) => ({
@@ -195,18 +231,19 @@ async function readRunForInspection(workspace, runId) {
 async function readCandidateFilesForInspection(workspace, candidateId) {
     return await readArchiveRecord("Candidate", candidateId, () => readLocalCandidateFilesForId(workspace, candidateId));
 }
-function readBenchmarkSourceMetadata(candidate) {
-    const benchmark = asRecord(asRecord(candidate.meta)?.benchmark);
-    const files = Array.isArray(benchmark?.files)
-        ? benchmark.files
+function readRunSourceInput(run) {
+    const input = asRecord(run.input);
+    const sourceYaml = typeof input?.sourceYaml === "string" ? input.sourceYaml : null;
+    if (!input || !sourceYaml) {
+        return null;
+    }
+    const engineResolveFiles = Array.isArray(input.engineResolveFiles)
+        ? input.engineResolveFiles
             .map(readSurfaceSnapshotFile)
             .filter((file) => file !== null)
+            .sort((left, right) => left.path.localeCompare(right.path))
         : [];
-    const sourceYaml = files.find((file) => file.path === WORKBENCH_BENCHMARK_FILE)?.content ?? null;
-    if (!sourceYaml) {
-        return null;
-    }
-    return { sourceYaml, files };
+    return { sourceYaml, engineResolveFiles };
 }
 function readSurfaceSnapshotFile(value) {
     const record = asRecord(value);

package/package.json

@@ -1,6 +1,11 @@
 {
   "name": "@workbench-ai/workbench",
-  "version": "0.0.65",
+  "version": "0.0.67",
+  "repository": {
+    "type": "git",
+    "url": "git+https://github.com/workbench-ai/workbench.git",
+    "directory": "packages/cli"
+  },
   "type": "module",
   "publishConfig": {
     "registry": "https://registry.npmjs.org/",
@@ -16,9 +21,9 @@
   ],
   "dependencies": {
     "yaml": "^2.8.2",
-    "@workbench-ai/workbench-built-in-adapters": "0.0.65",
-    "@workbench-ai/workbench-protocol": "0.0.65",
-    "@workbench-ai/workbench-core": "0.0.65"
+    "@workbench-ai/workbench-built-in-adapters": "0.0.67",
+    "@workbench-ai/workbench-core": "0.0.67",
+    "@workbench-ai/workbench-protocol": "0.0.67"
   },
   "devDependencies": {
     "@tailwindcss/postcss": "^4.2.2",