orizu 0.3.3 → 0.3.5

package/src/index.ts

@@ -24,6 +24,7 @@ import { parseDatasetFile } from './file-parser.js'
 import { streamJsonlRowChunks } from './jsonl-stream.js'
 import { parseDatasetReference } from './dataset-download.js'
 import { parseGlobalFlags } from './global-flags.js'
+import { runLocalAppPreview } from './preview-runtime.js'
 import {
   assertSecureTokenTransport,
   authedFetch,
@@ -163,6 +164,12 @@ interface ScorerSummary {
   implementationKind: string
 }
 
+interface ScoreSubmitInput {
+  resultsJsonl?: string
+  aggregate?: Record<string, unknown>
+  aggregateWarning?: string
+}
+
 interface RunnerExecContext {
   prompt: {
     body: string | null
@@ -185,6 +192,8 @@ interface RunnerExecContext {
 
 type DatasetUploadSourceType = 'csv' | 'json' | 'jsonl'
 
+const MAX_README_LENGTH = 200_000
+
 interface DatasetUploadResponse {
   dataset: {
     id: string
@@ -193,6 +202,11 @@ interface DatasetUploadResponse {
     sourceType: string
     url?: string
   }
+  readmeVersion?: {
+    id: string
+    version_num: number
+    created_at: string
+  } | null
 }
 
 interface TeamMember {
@@ -256,11 +270,16 @@ function printVersion() {
 }
 
 function printUsage() {
-  printLine(`orizu global options:\n\n  --local                 Use http://localhost:3000\n  --server <url>          Use a specific server origin (for example: https://preview.example.com)\n  --version, -v           Print the orizu CLI version\n\norizu commands:\n\n  orizu login [--no-prompt-if-logged-in]\n  orizu logout\n  orizu whoami\n  orizu env [--project <team/project>] [--project-id <projectId>]\n  orizu log <event_type> --run-id <id> --sequence <n> --payload @event.json\n  orizu teams list\n  orizu teams create [--name <name>]\n  orizu teams members list [--team <teamSlug>]\n  orizu teams members add --email <email> [--team <teamSlug>]\n  orizu teams members remove --email <email> [--team <teamSlug>]\n  orizu teams members role --team <teamSlug> --email <email> --role <admin|member>\n  orizu projects list [--team <teamSlug>]\n  orizu projects create --name <name> [--team <teamSlug>]\n  orizu prompts list --project <team/project>\n  orizu prompts comments <prompt-id-or-name> --project <team/project> [--label <label> | --version <id>] [--json]\n  orizu prompts pull <prompt-id-or-name> --project <team/project> --out <dir> [--label <label> | --version <id>] [--json]\n  orizu prompts push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--json]\n  orizu prompts labels set <prompt-name> <label> --version <version-id> [--project <team/project>] [--json]\n  orizu prompts scorers set-headline <prompt-id> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]\n  orizu prompts scorers add <prompt-id> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]\n  orizu scorers list --project <team/project>\n  orizu scorers register --project <team/project> --name <name> --manifest <manifest.json> [--prompt-version <id>] [--runner-version <id>] [--label <label>] [--json]\n  orizu scorers detail <scorer-id-or-name> --project <team/project> [--json]\n  orizu scorers labels set <scorer-name> <label> --version <scorer-version-id> [--project <team/project>] [--json]\n  orizu scores submit <results.jsonl|results.json> --scorer-version <id> --subject-version <prompt-version-id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]\n  orizu judges list --project <team/project>\n  orizu judges pull <judge-id-or-name> --project <team/project> --out <dir> [--label <label> | --version <id>] [--json]\n  orizu judges push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--json]\n  orizu runners push <dir> [--project <team/project>] [--name <name>] [--label <label>] [--json]\n  orizu runners exec (--prompt <prompt-version-id> | --prompt-version <id> --runner-version <id> | --scorer-version <id>) --dataset-version <id> --split-set <id-or-name> --split <name> [--runner-dir <dir>] --out <results.jsonl|results.jsonl.gz>\n  orizu optimizers push <dir> [--project <team/project>] [--name <name>] [--label <label>] [--json]\n  orizu runs submit <results.jsonl|results.jsonl.gz> --prompt-version <id> --runner-version <id> --dataset-version <id> --split-set <id> --split <name> [--project <team/project>]\n  orizu apps list [--project <team/project>]\n  orizu apps create --project <team/project> --name <name> --dataset <datasetId> --file <path> --input-schema <json-path> --output-schema <json-path> [--component <name>]\n  orizu apps update [--app <appId>] [--project <team/project>] --file <path> --input-schema <json-path> --output-schema <json-path> [--component <name>]\n  orizu apps link-dataset --dataset <datasetId> [--app <appId>] [--project <team/project>] [--version <n>]\n  orizu apps detail --app <appId> [--project <team/project>] [--json]\n  orizu tasks list [--project <team/project>]\n  orizu tasks create --project <team/project> --dataset <datasetId> --app <appId> --title <title> --assignees <userIdOrEmail1,userIdOrEmail2> [--version <n>] [--instructions <text>] [--labels-per-item <n>] [--json]\n  orizu tasks assign --task <taskId> --assignees <userId1,userId2>\n  orizu tasks status --task <taskId> [--json]\n  orizu tasks pause --task <taskId>\n  orizu tasks unpause --task <taskId>\n  orizu datasets upload --file <path> [--project <team/project>] [--name <name>]\n  orizu datasets push <path> [--project <team/project>] [--name <name>] [--json]\n  orizu datasets versions create <datasetId|dataset-name> [--project <team/project>] [--label <label>] [--json]\n  orizu datasets splits create <datasetVersionId> [--from-file <split.json>] [--json]\n  orizu datasets download [--dataset <datasetId|datasetUrl>] [--project <team/project>] [--format <csv|json|jsonl>] [--out <path>]\n  orizu datasets append [--dataset <datasetId|datasetUrl>] [--project <team/project>] --file <path>\n  orizu datasets edit-rows [--dataset <datasetId|datasetUrl>] [--project <team/project>] --file <path>\n  orizu datasets delete-rows [--dataset <datasetId|datasetUrl>] [--project <team/project>] --row-ids <id1,id2>\n  orizu datasets delete [--dataset <datasetId|datasetUrl>] [--project <team/project>]\n  orizu datasets lock [--dataset <datasetId|datasetUrl>] [--project <team/project>] [--reason <text>]\n  orizu datasets clone [--dataset <datasetId|datasetUrl>] [--project <team/project>] [--name <name>]\n  orizu tasks export [--task <taskId>] [--format <csv|json|jsonl>] [--out <path>]`)
+  printLine(`orizu global options:\n\n  --local                 Use http://localhost:3000\n  --server <url>          Use a specific server origin (for example: https://preview.example.com)\n  --version, -v           Print the orizu CLI version\n\norizu commands:\n\n  orizu login [--no-prompt-if-logged-in]\n  orizu logout\n  orizu whoami\n  orizu env [--project <team/project>] [--project-id <projectId>]\n  orizu log <event_type> --run-id <id> --sequence <n> --payload @event.json\n  orizu teams list\n  orizu teams create [--name <name>]\n  orizu teams members list [--team <teamSlug>]\n  orizu teams members add --email <email> [--team <teamSlug>]\n  orizu teams members remove --email <email> [--team <teamSlug>]\n  orizu teams members role --team <teamSlug> --email <email> --role <admin|member>\n  orizu projects list [--team <teamSlug>]\n  orizu projects create --name <name> [--team <teamSlug>]\n  orizu prompts list --project <team/project>\n  orizu prompts comments <prompt-id-or-name> --project <team/project> [--label <label> | --version <id>] [--json]\n  orizu prompts pull <prompt-id-or-name> --project <team/project> --out <dir> [--label <label> | --version <id>] [--json]\n  orizu prompts push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--json]\n  orizu prompts labels set <prompt-name> <label> --version <version-id> [--project <team/project>] [--json]\n  orizu prompts scorers set-headline <prompt-id> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]\n  orizu prompts scorers add <prompt-id> --scorer-version <id> [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]\n  orizu scorers list --project <team/project>\n  orizu scorers register --project <team/project> --name <name> --manifest <manifest.json> [--prompt-version <id>] [--runner-version <id>] [--label <label>] [--json]\n  orizu scorers detail <scorer-id-or-name> --project <team/project> [--json]\n  orizu scorers labels set <scorer-name> <label> --version <scorer-version-id> [--project <team/project>] [--json]\n  orizu scorers exec --scorer-version <id> (--subject-version <prompt-version-id> | --optimization-run <id> --candidate <id>) --dataset-version <id> --split-set <id> --split <name> [--subject-results <jsonl>] [--dependency-score-run <alias=id>] [--dependency-results <alias=path>] [--no-submit] [--out <score.json>] [--project <team/project>] [--json]\n  orizu scores submit <results.jsonl|results.json> --scorer-version <id> (--subject-version <prompt-version-id> | --optimization-run <id> --candidate <id>) [--aggregate] [--dataset-version <id> --split-set <id> --split <name>] [--project <team/project>] [--json]\n  orizu judges list --project <team/project>\n  orizu judges pull <judge-id-or-name> --project <team/project> --out <dir> [--label <label> | --version <id>] [--json]\n  orizu judges push <dir> [--runner-version <id>] [--project <team/project>] [--parent <version-id>] [--json]\n  orizu runners push <dir> [--project <team/project>] [--name <name>] [--label <label>] [--json]\n  orizu runners exec (--prompt <prompt-version-id> | --prompt-version <id> --runner-version <id> | --scorer-version <id>) --dataset-version <id> --split-set <id-or-name> --split <name> [--runner-dir <dir>] --out <results.jsonl|results.jsonl.gz>\n  orizu optimizers push <dir> [--project <team/project>] [--name <name>] [--label <label>] [--json]\n  orizu runs submit <results.jsonl|results.jsonl.gz> --prompt-version <id> --runner-version <id> --dataset-version <id> --split-set <id> --split <name> [--project <team/project>]\n  orizu apps list [--project <team/project>]\n  orizu apps create --project <team/project> --name <name> --dataset <datasetId> --file <path> --input-schema <json-path> --output-schema <json-path> [--component <name>]\n  orizu apps update [--app <appId>] [--project <team/project>] --file <path> --input-schema <json-path> --output-schema <json-path> [--component <name>]\n  orizu apps link-dataset --dataset <datasetId> [--app <appId>] [--project <team/project>] [--version <n>]\n  orizu apps detail --app <appId> [--project <team/project>] [--json]\n  orizu apps export [--app <appId>] [--project <team/project>] [--version <n>] [--out <path>]\n  orizu tasks list [--project <team/project>]\n  orizu tasks create --project <team/project> --dataset <datasetId> --app <appId> --title <title> --assignees <userIdOrEmail1,userIdOrEmail2> [--version <n>] [--instructions <text>] [--labels-per-item <n>] [--json]\n  orizu tasks assign --task <taskId> --assignees <userId1,userId2>\n  orizu tasks status --task <taskId> [--json]\n  orizu tasks pause --task <taskId>\n  orizu tasks unpause --task <taskId>\n  orizu datasets upload --file <path> [--project <team/project>] [--name <name>] [--readme-file <README.md> | --readme-text <markdown>]\n  orizu datasets push <path> [--project <team/project>] [--name <name>] [--readme-file <README.md> | --readme-text <markdown>] [--json]\n  orizu datasets readme set <datasetId|dataset-name> [--project <team/project>] (--readme-file <README.md> | --readme-text <markdown>) [--json]\n  orizu datasets versions create <datasetId|dataset-name> [--project <team/project>] [--label <label>] [--readme-file <README.md> | --readme-text <markdown>] [--json]\n  orizu datasets splits create <datasetVersionId> [--from-file <split.json>] [--json]\n  orizu datasets download [--dataset <datasetId|datasetUrl>] [--project <team/project>] [--format <csv|json|jsonl>] [--out <path>]\n  orizu datasets append [--dataset <datasetId|datasetUrl>] [--project <team/project>] --file <path>\n  orizu datasets edit-rows [--dataset <datasetId|datasetUrl>] [--project <team/project>] --file <path>\n  orizu datasets delete-rows [--dataset <datasetId|datasetUrl>] [--project <team/project>] --row-ids <id1,id2>\n  orizu datasets delete [--dataset <datasetId|datasetUrl>] [--project <team/project>]\n  orizu datasets lock [--dataset <datasetId|datasetUrl>] [--project <team/project>] [--reason <text>]\n  orizu datasets clone [--dataset <datasetId|datasetUrl>] [--project <team/project>] [--name <name>]\n  orizu tasks export [--task <taskId>] [--format <csv|json|jsonl>] [--out <path>]`)
+  printPreviewUsage()
+}
+
+function printPreviewUsage() {
+  printLine(`\nLocal app preview:\n\n  orizu apps preview --file <path> --input-schema <json-path> --output-schema <json-path> --sample-row <json-path> [--screenshot <png-path>] [--headed] [--keep-open] [--component <name>]`)
 }
 
 function printOptimizationUsage() {
-  printLine(`\nOptimization lifecycle commands:\n\n  orizu optimizations start --project <team/project> --optimizer-version <id> --prompt-version <id[,id]> --selection-scorer <id> [--reflection-scorer <id>] [--pareto-scorer <id>] [--best-scorer <id>] --dataset-version <id> --split-set <id> [--train-split <name>] [--validation-split <name>] [--metadata <json|@file>] [--json]\n  orizu optimizations run-gepa --project <team/project> --optimizer-version-id <id> --candidate-version-id <id> --runner-version-id <id> --candidate-runner-dir <dir> --scorer-version-id <id> --scorer-runner-version-id <id> --scorer-runner-dir <dir> --dataset-version-id <id> --split-set-id <id> [--train-split train] [--val-split validation] [--log-dir logs]\n  orizu optimizations export <run-id> [--out <path>] [--json]\n  orizu optimizations pause <run-id> [--reason <text>] [--json]\n  orizu optimizations resume <run-id> [--json]\n  orizu optimizations finish <run-id> [--best-score <n>] [--best-candidate <id>] [--result-prompt-version <id>] [--metadata <json|@file>] [--json]\n  orizu optimizations fail <run-id> [--reason <text>] [--metadata <json|@file>] [--json]\n  orizu optimizations cancel <run-id> [--reason <text>] [--json]`)
+  printLine(`\nOptimization lifecycle commands:\n\n  orizu optimizations start --project <team/project> --optimizer-version <id> --prompt-version <id[,id]> --selection-scorer <id> [--reflection-scorer <id>] [--pareto-scorer <id>] [--best-scorer <id>] --dataset-version <id> --split-set <id> [--train-split <name>] [--validation-split <name>] [--metadata <json|@file>] [--json]\n  orizu optimizations run-gepa --project <team/project> --optimizer-version-id <id> --candidate-version-id <id> --runner-version-id <id> --candidate-runner-dir <dir> --scorer-version-id <id> --scorer-runner-version-id <id> --scorer-runner-dir <dir> --dataset-version-id <id> --split-set-id <id> [--train-split train] [--val-split validation] [--log-dir logs]\n  orizu optimizations export <run-id> [--out <path>] [--json]\n  orizu optimizations pause <run-id> [--reason <text>] [--json]\n  orizu optimizations resume <run-id> [--json]\n  orizu optimizations finish <run-id> [--best-score <n>] [--best-candidate <id>] [--result-prompt-version <id>] [--report <markdown|@file> | --report-file <path>] [--metadata <json|@file>] [--json]\n  orizu optimizations fail <run-id> [--reason <text>] [--report <markdown|@file> | --report-file <path>] [--metadata <json|@file>] [--json]\n  orizu optimizations cancel <run-id> [--reason <text>] [--report <markdown|@file> | --report-file <path>] [--json]`)
 }
 
 let cliArgs = process.argv.slice(2)
@@ -295,6 +314,25 @@ export function expandHomePath(path: string): string {
   return path
 }
 
+function readReadmeMarkdownFromArgs(usage: string): string | null {
+  const readmeFile = getArg('--readme-file')
+  const readmeText = getArg('--readme-text')
+
+  if (readmeFile !== null && readmeText !== null) {
+    throw new Error(`Use either --readme-file or --readme-text, not both.\n${usage}`)
+  }
+
+  const markdown = readmeFile !== null
+    ? readFileSync(expandHomePath(readmeFile), 'utf8')
+    : readmeText
+
+  if (markdown !== null && markdown.length > MAX_README_LENGTH) {
+    throw new Error(`README markdown is too large. Maximum size is ${MAX_README_LENGTH} characters.`)
+  }
+
+  return markdown
+}
+
 function createCodeVerifier(): string {
   return randomBytes(32).toString('base64url')
 }
@@ -1807,16 +1845,21 @@ async function createDatasetVersion() {
   const datasetId = getPositionalArg(3) || getArg('--dataset')
   const versionLabel = getArg('--label') || getArg('--version-label') || null
   const project = getArg('--project')
+  const usage = 'Usage: orizu datasets versions create <datasetId|dataset-name> [--project <team/project>] [--label <label>] [--readme-file <README.md> | --readme-text <markdown>] [--json]'
 
   if (!datasetId) {
-    throw new Error('Usage: orizu datasets versions create <datasetId|dataset-name> [--project <team/project>] [--label <label>] [--json]')
+    throw new Error(usage)
   }
 
+  const readmeMarkdown = readReadmeMarkdownFromArgs(usage)
   const resolvedDatasetId = await resolveDatasetIdForProject(datasetId, project)
   const response = await authedFetch(`/api/cli/datasets/${encodeURIComponent(resolvedDatasetId)}/versions`, {
     method: 'POST',
     headers: { 'Content-Type': 'application/json' },
-    body: JSON.stringify({ versionLabel }),
+    body: JSON.stringify({
+      versionLabel,
+      ...(readmeMarkdown !== null ? { readmeMarkdown } : {}),
+    }),
   })
 
   if (!response.ok) {
@@ -1970,18 +2013,23 @@ async function submitScoreResults() {
   const candidateId = getArg('--candidate')
 
   if (!resultsPath || !scorerVersionId || (!subjectPromptVersionId && (!optimizationRunId || !candidateId))) {
-    throw new Error('Usage: orizu scores submit <results.jsonl|results.json> --project <team/project> --scorer-version <id> (--subject-version <prompt-version-id> | --optimization-run <id> --candidate <id>) [--dataset-version <id> --split-set <id> --split <name>] [--json]')
+    throw new Error('Usage: orizu scores submit <results.jsonl|results.json> --project <team/project> --scorer-version <id> (--subject-version <prompt-version-id> | --optimization-run <id> --candidate <id>) [--aggregate] [--dataset-version <id> --split-set <id> --split <name>] [--json]')
   }
 
   const resultBytes = readSourceBytes(resultsPath)
   const raw = resultsPath.endsWith('.gz')
     ? gunzipSync(resultBytes).toString('utf8')
     : resultBytes.toString('utf8')
-  const resultsJsonl = normalizeScoreResultsInput(resultsPath, raw)
+  const scoreInput = normalizeScoreResultsInput(resultsPath, raw, hasArg('--aggregate'))
+  if (scoreInput.aggregateWarning) {
+    console.warn(scoreInput.aggregateWarning)
+  }
+  const aggregatePayload = scoreInput.aggregate || {}
   const response = await authedFetch(`/api/cli/scores/submit?project=${encodeURIComponent(project)}`, {
     method: 'POST',
     headers: { 'Content-Type': 'application/json' },
     body: JSON.stringify({
+      ...aggregatePayload,
       scorerVersionId,
       subjectPromptVersionId: subjectPromptVersionId || undefined,
       datasetVersionId: getArg('--dataset-version') || undefined,
@@ -1989,7 +2037,7 @@ async function submitScoreResults() {
       splitName: getArg('--split') || undefined,
       optimizationRunId: optimizationRunId || undefined,
       candidateId: candidateId || undefined,
-      resultsJsonl,
+      resultsJsonl: scoreInput.resultsJsonl,
     }),
   })
 
@@ -2015,24 +2063,161 @@ async function submitScoreResults() {
   )
 }
 
-function normalizeScoreResultsInput(sourcePath: string, raw: string): string {
+function looksLikeAggregateScoreObject(value: Record<string, unknown>): boolean {
+  return (
+    value.scoreValue !== undefined ||
+    value.score_value !== undefined ||
+    value.diagnostics !== undefined ||
+    value.feedbackSummary !== undefined ||
+    value.feedback_summary !== undefined ||
+    value.rowEvidence !== undefined ||
+    value.row_evidence !== undefined ||
+    value.dependencyScoreRunIds !== undefined ||
+    value.dependency_score_run_ids !== undefined
+  )
+}
+
+function numberFromUnknown(value: unknown): number | null {
+  if (typeof value === 'number' && Number.isFinite(value)) return value
+  if (typeof value === 'string' && value.trim()) {
+    const parsed = Number(value)
+    return Number.isFinite(parsed) ? parsed : null
+  }
+  if (typeof value === 'boolean') return value ? 1 : 0
+  return null
+}
+
+function normalizeScoreResultsInput(sourcePath: string, raw: string, aggregateMode = false): ScoreSubmitInput {
   const logicalPath = sourcePath.endsWith('.gz') ? sourcePath.slice(0, -3) : sourcePath
-  if (!logicalPath.endsWith('.json')) return raw
+  if (!logicalPath.endsWith('.json')) return { resultsJsonl: raw }
 
   try {
     const parsed = JSON.parse(raw) as unknown
+    if (aggregateMode) {
+      if (!isRecord(parsed)) {
+        throw new Error('Aggregate score results JSON must be an object')
+      }
+      return { aggregate: parsed }
+    }
+    if (isRecord(parsed) && looksLikeAggregateScoreObject(parsed)) {
+      return {
+        aggregate: parsed,
+        aggregateWarning: 'Warning: aggregate score JSON detected. Pass --aggregate to make this explicit.',
+      }
+    }
     const rows = Array.isArray(parsed) ? parsed : [parsed]
     if (rows.some(row => !isRecord(row))) {
       throw new Error('JSON score results must be an object or an array of objects')
     }
-    return rows.map(row => JSON.stringify(row)).join('\n')
+    return { resultsJsonl: rows.map(row => JSON.stringify(row)).join('\n') }
   } catch (error) {
     const message = error instanceof Error ? error.message : String(error)
     throw new Error(`Invalid score results JSON '${sourcePath}': ${message}`)
   }
 }
 
+function readOptionalJsonlFile(pathArg: string | null): string | undefined {
+  if (!pathArg) return undefined
+  const bytes = readSourceBytes(pathArg)
+  return pathArg.endsWith('.gz')
+    ? gunzipSync(bytes).toString('utf8')
+    : bytes.toString('utf8')
+}
+
+function parseAliasAssignments(value: string | null): Record<string, string> {
+  const result: Record<string, string> = {}
+  if (!value) return result
+  for (const item of parseCommaSeparated(value)) {
+    const equalsIndex = item.indexOf('=')
+    if (equalsIndex <= 0 || equalsIndex === item.length - 1) {
+      throw new Error(`Expected alias assignment in form alias=value, got '${item}'`)
+    }
+    result[item.slice(0, equalsIndex)] = item.slice(equalsIndex + 1)
+  }
+  return result
+}
+
+function readDependencyResultAssignments(value: string | null): Record<string, string> | undefined {
+  const assignments = parseAliasAssignments(value)
+  const result: Record<string, string> = {}
+  for (const [alias, path] of Object.entries(assignments)) {
+    result[alias] = readOptionalJsonlFile(path) || ''
+  }
+  return Object.keys(result).length > 0 ? result : undefined
+}
+
+async function execScorer() {
+  const project = getArg('--project') || await resolveProjectSlug(null)
+  const scorerVersionId = getArg('--scorer-version')
+  const subjectPromptVersionId = getArg('--subject-version') || getArg('--prompt-version')
+  const optimizationRunId = getArg('--optimization-run')
+  const candidateId = getArg('--candidate')
+  const datasetVersionId = getArg('--dataset-version')
+  const splitSetId = getArg('--split-set')
+  const splitName = getArg('--split')
+  const outPath = getArg('--out')
+
+  if (
+    !scorerVersionId ||
+    (!subjectPromptVersionId && (!optimizationRunId || !candidateId)) ||
+    !datasetVersionId ||
+    !splitSetId ||
+    !splitName
+  ) {
+    throw new Error('Usage: orizu scorers exec --scorer-version <id> (--subject-version <prompt-version-id> | --optimization-run <id> --candidate <id>) --dataset-version <id> --split-set <id> --split <name> [--subject-results <jsonl>] [--dependency-score-run <alias=id>] [--dependency-results <alias=path>] [--no-submit] [--out <score.json>] [--project <team/project>] [--json]')
+  }
+
+  const response = await authedFetch(`/api/cli/scorers/exec?project=${encodeURIComponent(project)}`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({
+      scorerVersionId,
+      subjectPromptVersionId: subjectPromptVersionId || undefined,
+      optimizationRunId: optimizationRunId || undefined,
+      candidateId: candidateId || undefined,
+      datasetVersionId,
+      splitSetId,
+      splitName,
+      submit: !hasArg('--no-submit'),
+      reuseExisting: !hasArg('--no-reuse-existing'),
+      subjectResultsJsonl: readOptionalJsonlFile(getArg('--subject-results') || getArg('--outputs')),
+      dependencyScoreRunIds: parseAliasAssignments(getArg('--dependency-score-run') || getArg('--dependency-score-runs')),
+      dependencyResultsJsonl: readDependencyResultAssignments(getArg('--dependency-results')),
+    }),
+  })
+
+  if (!response.ok) {
+    throw new Error(`Failed to execute scorer: ${await response.text()}`)
+  }
+
+  const data = await parseJsonResponse<Record<string, unknown>>(response, 'Scorer exec')
+  if (outPath) {
+    writeFileSync(expandHomePath(outPath), `${JSON.stringify(data.scoreResult || data, null, 2)}\n`)
+  }
+  if (hasJsonFlag()) {
+    printJson(data)
+    return
+  }
+
+  const scoreRun = isRecord(data.scoreRun) ? data.scoreRun : null
+  const scoreResult = isRecord(data.scoreResult) ? data.scoreResult : null
+  const scoreValue = scoreRun
+    ? scoreRun.scoreValue
+    : scoreResult
+      ? scoreResult.scoreValue
+      : null
+  if (scoreRun && typeof scoreRun.id === 'string') {
+    printLine(`Executed scorer ${sanitizeTerminalText(scorerVersionId)} -> score ${sanitizeTerminalText(scoreRun.id)} (${formatPercent(numberFromUnknown(scoreValue))})`)
+  } else {
+    printLine(`Executed scorer ${sanitizeTerminalText(scorerVersionId)} (${formatPercent(numberFromUnknown(scoreValue))})`)
+  }
+  if (outPath) {
+    printLine(`Wrote scorer result to ${sanitizeTerminalText(outPath)}`)
+  }
+}
+
 type OptimizationLifecycleAction = 'pause' | 'resume' | 'finish' | 'fail' | 'cancel'
+const OPTIMIZATION_REPORT_MAX_BYTES = 2 * 1024 * 1024
 
 function parseOptionalNumberFlag(name: string): number | undefined {
   const value = getArg(name)
@@ -2052,6 +2237,46 @@ function hasObjectKeys(value: Record<string, unknown>): boolean {
   return Object.keys(value).length > 0
 }
 
+function readOptimizationReportInput(): { markdown: string; sourceName: string | null } | null {
+  const report = getArg('--report')
+  const reportFile = getArg('--report-file')
+
+  if (report && reportFile) {
+    throw new Error('Use either --report or --report-file, not both')
+  }
+
+  if (!report && !reportFile) {
+    return null
+  }
+
+  let markdown: string
+  let sourceName: string | null
+
+  if (reportFile) {
+    const expandedPath = expandHomePath(reportFile)
+    markdown = readSourceFile(reportFile)
+    sourceName = basename(expandedPath)
+  } else if (report?.startsWith('@')) {
+    const path = report.slice(1)
+    const expandedPath = expandHomePath(path)
+    markdown = readSourceFile(path)
+    sourceName = basename(expandedPath)
+  } else {
+    markdown = report || ''
+    sourceName = 'inline'
+  }
+
+  if (!markdown.trim()) {
+    throw new Error('Optimization report markdown must not be blank')
+  }
+
+  if (Buffer.byteLength(markdown, 'utf8') > OPTIMIZATION_REPORT_MAX_BYTES) {
+    throw new Error(`Optimization report exceeds ${OPTIMIZATION_REPORT_MAX_BYTES} bytes`)
+  }
+
+  return { markdown, sourceName }
+}
+
 async function startOptimizationRun() {
   const project = getArg('--project') || await resolveProjectSlug(null)
   const optimizerVersionId = getArg('--optimizer-version')
@@ -2156,6 +2381,11 @@ async function updateOptimizationRunLifecycle(action: OptimizationLifecycleActio
   }
   const metadata = readJsonObjectArg(getArg('--metadata'), 'optimization metadata')
   const reason = getArg('--reason')
+  const report = readOptimizationReportInput()
+
+  if (report && (action === 'pause' || action === 'resume')) {
+    throw new Error(`orizu optimizations ${action} does not accept optimization reports`)
+  }
 
   if ((action === 'pause' || action === 'cancel') && reason) {
     metadata.reason = reason
@@ -2170,6 +2400,11 @@ async function updateOptimizationRunLifecycle(action: OptimizationLifecycleActio
     body.metadata = metadata
   }
 
+  if (report) {
+    body.reportMarkdown = report.markdown
+    body.reportSourceName = report.sourceName
+  }
+
   if (action === 'finish') {
     const bestScore = parseOptionalNumberFlag('--best-score')
     if (bestScore !== undefined) {
@@ -2707,15 +2942,61 @@ async function createAppFromFile() {
   }
 
   const data = await parseJsonResponse<{
-    app: { id: string; name: string; versionNum: number; componentName?: string }
+    app: { id: string; name: string; versionNum: number; componentName?: string; url?: string }
     warnings?: string[]
   }>(response, 'App create')
   printLine(`Created app ${sanitizeTerminalText(data.app.name)} (${sanitizeTerminalText(data.app.id)}) v${data.app.versionNum}`)
+  if (data.app.url) {
+    printLine(`View app: ${formatTerminalLink(data.app.url)}`)
+  }
   if (data.warnings?.length) {
     printLine(`Warnings: ${sanitizeTerminalText(data.warnings.join('; '))}`)
   }
 }
 
+async function previewAppFromFile() {
+  const filePath = getArg('--file')
+  const inputSchemaPath = getArg('--input-schema')
+  const outputSchemaPath = getArg('--output-schema')
+  const sampleRowPath = getArg('--sample-row')
+  const screenshotPath = getArg('--screenshot')
+  const component = getArg('--component') || undefined
+  const headed = hasArg('--headed')
+  const keepOpen = hasArg('--keep-open')
+
+  if (!filePath || !inputSchemaPath || !outputSchemaPath || !sampleRowPath) {
+    throw new Error('Usage: orizu apps preview --file <path> --input-schema <json-path> --output-schema <json-path> --sample-row <json-path> [--screenshot <png-path>] [--headed] [--keep-open] [--component <name>]')
+  }
+
+  const expandedFilePath = realpathSync(expandHomePath(filePath))
+  const inputJsonSchema = readJsonFile(inputSchemaPath)
+  const outputJsonSchema = readJsonFile(outputSchemaPath)
+  const sampleRow = readJsonFile(sampleRowPath)
+  const expandedScreenshotPath = screenshotPath ? expandHomePath(screenshotPath) : undefined
+
+  const result = await runLocalAppPreview({
+    filePath: expandedFilePath,
+    inputSchema: inputJsonSchema,
+    outputSchema: outputJsonSchema,
+    sampleRow,
+    screenshotPath: expandedScreenshotPath,
+    componentName: component,
+    headed,
+    keepOpen,
+  })
+
+  printLine(`Preview rendered: ${formatTerminalLink(result.url)}`)
+  if (result.screenshotPath) {
+    printLine(`Screenshot: ${sanitizeTerminalText(result.screenshotPath)}`)
+  }
+  if (headed || keepOpen) {
+    printLine('Headed preview is running. Close Chromium or stop the command when you are done.')
+  }
+  if (result.warnings.length > 0) {
+    printLine(`Warnings: ${sanitizeTerminalText(result.warnings.join('; '))}`)
+  }
+}
+
 async function updateAppFromFile() {
   const filePath = getArg('--file')
   const inputSchemaPath = getArg('--input-schema')
@@ -3023,6 +3304,28 @@ interface AppDetailPayload {
   projectName: string
 }
 
+interface AppExportPayload {
+  app: {
+    id: string
+    name: string
+  }
+  version: {
+    id: string
+    versionNum: number
+    createdAt?: string
+    code: string
+  }
+}
+
+function defaultAppExportFilename(appName: string, versionNum: number): string {
+  const safeName = appName
+    .trim()
+    .replace(/[^A-Za-z0-9._-]+/g, '_')
+    .replace(/^_+|_+$/g, '')
+    .slice(0, 100)
+  return `${safeName || 'app'}.v${versionNum}.tsx`
+}
+
 async function appDetail() {
   const appId = getArg('--app')
   const project = getArg('--project')
@@ -3068,6 +3371,50 @@ async function appDetail() {
   printLine(`  Updated: ${sanitizeTerminalText(detail.updatedAt)}`)
 }
 
+async function exportAppSource() {
+  const project = getArg('--project')
+  const versionArg = getArg('--version')
+  const outPathArg = getArg('--out')
+  let appId = getArg('--app')
+
+  const versionNum = versionArg && /^[1-9]\d*$/.test(versionArg)
+    ? Number.parseInt(versionArg, 10)
+    : undefined
+
+  if (versionArg && versionNum === undefined) {
+    throw new Error('--version must be a positive integer')
+  }
+
+  if (!appId) {
+    const selected = await selectAppIdInteractively(project)
+    appId = selected.appId
+  }
+
+  const query = new URLSearchParams()
+  if (project) {
+    query.set('project', project)
+  }
+  if (versionNum) {
+    query.set('version', String(versionNum))
+  }
+
+  const suffix = query.toString() ? `?${query.toString()}` : ''
+  const response = await authedFetch(`/api/cli/apps/${encodeURIComponent(appId)}/export${suffix}`)
+  if (!response.ok) {
+    throw new Error(`Failed to export app (${response.status}): ${await response.text()}`)
+  }
+
+  const data = await parseJsonResponse<AppExportPayload>(response, 'App export')
+  const filename = outPathArg
+    ? expandHomePath(outPathArg)
+    : defaultAppExportFilename(data.app.name, data.version.versionNum)
+
+  writeTextFileEnsuringDir(filename, data.version.code)
+  printLine(
+    `Saved app ${sanitizeTerminalText(data.app.name)} v${data.version.versionNum} source to ${sanitizeTerminalText(filename)}`
+  )
+}
+
 async function listTeamMembers() {
   const teamSlug = await resolveTeamSlug(getArg('--team'))
 
@@ -3154,7 +3501,8 @@ async function createDatasetFromRows(
   project: string,
   name: string,
   sourceType: DatasetUploadSourceType,
-  rows: Array<Record<string, unknown>>
+  rows: Array<Record<string, unknown>>,
+  readmeMarkdown: string | null = null
 ): Promise<DatasetUploadResponse> {
   const response = await authedFetch('/api/cli/datasets/upload', {
     method: 'POST',
@@ -3166,6 +3514,7 @@ async function createDatasetFromRows(
       name,
       rows,
       sourceType,
+      ...(readmeMarkdown !== null ? { readmeMarkdown } : {}),
     }),
   })
 
@@ -3180,7 +3529,8 @@ async function createDatasetFromRows(
 async function uploadJsonlDatasetInChunks(
   file: string,
   project: string,
-  datasetName: string
+  datasetName: string,
+  readmeMarkdown: string | null = null
 ) {
   let dataset: DatasetUploadResponse['dataset'] | null = null
   let totalUploaded = 0
@@ -3215,7 +3565,7 @@ async function uploadJsonlDatasetInChunks(
 
     try {
       if (!dataset) {
-        const data = await createDatasetFromRows(project, datasetName, 'jsonl', chunk)
+        const data = await createDatasetFromRows(project, datasetName, 'jsonl', chunk, readmeMarkdown)
         dataset = data.dataset
         totalUploaded = data.dataset.rowCount
         continue
@@ -3256,22 +3606,24 @@ async function uploadDataset() {
   const projectArg = getArg('--project')
   const fileArg = getArg('--file')
   const name = getArg('--name')
+  const usage = 'Usage: orizu datasets upload --file <path> [--project <team/project>] [--name <name>] [--readme-file <README.md> | --readme-text <markdown>]'
 
   if (!fileArg) {
-    throw new Error('Usage: orizu datasets upload --file <path> [--project <team/project>] [--name <name>]')
+    throw new Error(usage)
   }
 
+  const readmeMarkdown = readReadmeMarkdownFromArgs(usage)
   const file = expandHomePath(fileArg)
   const project = await resolveProjectSlug(projectArg)
   const datasetName = name || basename(file)
 
   if (extname(file).toLowerCase() === '.jsonl') {
-    await uploadJsonlDatasetInChunks(file, project, datasetName)
+    await uploadJsonlDatasetInChunks(file, project, datasetName, readmeMarkdown)
     return
   }
 
   const { rows, sourceType } = parseDatasetFile(file)
-  const data = await createDatasetFromRows(project, datasetName, sourceType, rows)
+  const data = await createDatasetFromRows(project, datasetName, sourceType, rows, readmeMarkdown)
 
   printLine(`Uploaded dataset ${sanitizeTerminalText(data.dataset.name)} (${sanitizeTerminalText(data.dataset.id)}) with ${data.dataset.rowCount} rows.`)
   if (data.dataset.url) {
@@ -3283,23 +3635,29 @@ async function pushDataset() {
   const projectArg = getArg('--project')
   const fileArg = getPositionalArg(2) || getArg('--file')
   const name = getArg('--name')
+  const usage = 'Usage: orizu datasets push <rows.csv|rows.json|rows.jsonl> [--project <team/project>] [--name <name>] [--readme-file <README.md> | --readme-text <markdown>] [--json]'
 
   if (!fileArg) {
-    throw new Error('Usage: orizu datasets push <rows.csv|rows.json|rows.jsonl> [--project <team/project>] [--name <name>] [--json]')
+    throw new Error(usage)
   }
 
+  const readmeMarkdown = readReadmeMarkdownFromArgs(usage)
   const file = expandHomePath(fileArg)
   const project = await resolveProjectSlug(projectArg)
   const datasetName = name || basename(file)
   const { rows, sourceType } = parseDatasetFile(file)
-  const data = await createDatasetFromRows(project, datasetName, sourceType, rows)
+  const data = await createDatasetFromRows(project, datasetName, sourceType, rows, readmeMarkdown)
 
   if (hasJsonFlag()) {
-    printJson({
+    const payload: Record<string, unknown> = {
       dataset_id: data.dataset.id,
       name: data.dataset.name,
       row_count: data.dataset.rowCount,
-    })
+    }
+    if (data.readmeVersion?.id) {
+      payload.readme_version_id = data.readmeVersion.id
+    }
+    printJson(payload)
     return
   }
 
@@ -3309,6 +3667,55 @@ async function pushDataset() {
   }
 }
 
+async function setDatasetReadme() {
+  const datasetId = getPositionalArg(3) || getArg('--dataset')
+  const project = getArg('--project')
+  const usage = 'Usage: orizu datasets readme set <datasetId|dataset-name> [--project <team/project>] (--readme-file <README.md> | --readme-text <markdown>) [--json]'
+
+  if (!datasetId) {
+    throw new Error(usage)
+  }
+
+  const readmeMarkdown = readReadmeMarkdownFromArgs(usage)
+  if (readmeMarkdown === null) {
+    throw new Error(usage)
+  }
+
+  const resolvedDatasetId = await resolveDatasetIdForProject(datasetId, project)
+  const response = await authedFetch(`/api/cli/datasets/${encodeURIComponent(resolvedDatasetId)}/readme`, {
+    method: 'POST',
+    headers: { 'Content-Type': 'application/json' },
+    body: JSON.stringify({ markdown: readmeMarkdown }),
+  })
+
+  if (!response.ok) {
+    throw new Error(`Failed to save dataset README: ${await response.text()}`)
+  }
+
+  const data = await parseJsonResponse<{
+    version: {
+      id: string
+      dataset_id: string
+      version_num: number
+      created_at: string
+    }
+  }>(response, 'Dataset README save')
+
+  if (hasJsonFlag()) {
+    printJson({
+      dataset_id: data.version.dataset_id,
+      readme_version_id: data.version.id,
+      readme_version_num: data.version.version_num,
+    })
+    return
+  }
+
+  printLine(
+    `Saved README v${data.version.version_num} for dataset ` +
+    `${sanitizeTerminalText(data.version.dataset_id)} (${sanitizeTerminalText(data.version.id)}).`
+  )
+}
+
 function getDatasetReferenceInput(): string | null {
   const fromFlag = getArg('--dataset')
   if (fromFlag) {
@@ -4180,6 +4587,11 @@ export async function main(rawArgs = process.argv.slice(2)) {
     return
   }
 
+  if (command === 'scorers' && subcommand === 'exec') {
+    await execScorer()
+    return
+  }
+
   if (command === 'runners' && subcommand === 'push') {
     await pushRunnerArtifact('runner')
     return
@@ -4244,6 +4656,11 @@ export async function main(rawArgs = process.argv.slice(2)) {
     return
   }
 
+  if (command === 'apps' && subcommand === 'preview') {
+    await previewAppFromFile()
+    return
+  }
+
   if (command === 'apps' && subcommand === 'update') {
     await updateAppFromFile()
     return
@@ -4259,6 +4676,11 @@ export async function main(rawArgs = process.argv.slice(2)) {
     return
   }
 
+  if (command === 'apps' && subcommand === 'export') {
+    await exportAppSource()
+    return
+  }
+
   if (command === 'tasks' && subcommand === 'list') {
     await listTasks()
     return
@@ -4310,6 +4732,11 @@ export async function main(rawArgs = process.argv.slice(2)) {
   }
 
   const datasetsAction = cliArgs[2]
+  if (command === 'datasets' && subcommand === 'readme' && datasetsAction === 'set') {
+    await setDatasetReadme()
+    return
+  }
+
   if (command === 'datasets' && subcommand === 'versions' && datasetsAction === 'create') {
     await createDatasetVersion()
     return