@artemiskit/cli 0.2.0 → 0.2.3

package/dist/src/cli.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../src/cli.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAWpC,wBAAgB,SAAS,IAAI,OAAO,CAuCnC"}
+{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../src/cli.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAYpC,wBAAgB,SAAS,IAAI,OAAO,CAwCnC"}

package/dist/src/commands/baseline.d.ts

@@ -0,0 +1,9 @@
+/**
+ * Baseline command - Manage baseline runs for regression detection
+ */
+import { Command } from 'commander';
+/**
+ * Create the main baseline command with subcommands
+ */
+export declare function baselineCommand(): Command;
+//# sourceMappingURL=baseline.d.ts.map

package/dist/src/commands/baseline.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"baseline.d.ts","sourceRoot":"","sources":["../../../src/commands/baseline.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAocpC;;GAEG;AACH,wBAAgB,eAAe,IAAI,OAAO,CAWzC"}

package/dist/src/commands/history.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"history.d.ts","sourceRoot":"","sources":["../../../src/commands/history.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AA8FpC,wBAAgB,cAAc,IAAI,OAAO,CAiFxC"}
+{"version":3,"file":"history.d.ts","sourceRoot":"","sources":["../../../src/commands/history.ts"],"names":[],"mappings":"AAAA;;GAEG;AAIH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AA4IpC,wBAAgB,cAAc,IAAI,OAAO,CAmFxC"}

package/dist/src/commands/redteam.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"redteam.d.ts","sourceRoot":"","sources":["../../../src/commands/redteam.ts"],"names":[],"mappings":"AAAA;;GAEG;AAkCH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAkCpC,wBAAgB,cAAc,IAAI,OAAO,CA6bxC"}
+{"version":3,"file":"redteam.d.ts","sourceRoot":"","sources":["../../../src/commands/redteam.ts"],"names":[],"mappings":"AAAA;;GAEG;AAsCH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAoCpC,wBAAgB,cAAc,IAAI,OAAO,CAycxC"}

package/dist/src/commands/run.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/commands/run.ts"],"names":[],"mappings":"AAAA;;GAEG;AAYH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAgVpC,wBAAgB,UAAU,IAAI,OAAO,CAmTpC"}
+{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/commands/run.ts"],"names":[],"mappings":"AAAA;;GAEG;AAiBH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAyiBpC,wBAAgB,UAAU,IAAI,OAAO,CAggBpC"}

package/dist/src/commands/stress.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"stress.d.ts","sourceRoot":"","sources":["../../../src/commands/stress.ts"],"names":[],"mappings":"AAAA;;GAEG;AAmBH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAmCpC,wBAAgB,aAAa,IAAI,OAAO,CAuRvC"}
+{"version":3,"file":"stress.d.ts","sourceRoot":"","sources":["../../../src/commands/stress.ts"],"names":[],"mappings":"AAAA;;GAEG;AAoBH,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAsCpC,wBAAgB,aAAa,IAAI,OAAO,CA+SvC"}

package/dist/src/config/schema.d.ts

@@ -13,6 +13,7 @@ declare const ProviderConfigSchema: z.ZodObject<{
     deploymentName: z.ZodOptional<z.ZodString>;
     apiVersion: z.ZodOptional<z.ZodString>;
     embeddingDeploymentName: z.ZodOptional<z.ZodString>;
+    modelFamily: z.ZodOptional<z.ZodString>;
     underlyingProvider: z.ZodOptional<z.ZodEnum<["openai", "azure", "anthropic", "google", "mistral"]>>;
 }, "strip", z.ZodTypeAny, {
     apiKey?: string | undefined;
@@ -25,6 +26,7 @@ declare const ProviderConfigSchema: z.ZodObject<{
     deploymentName?: string | undefined;
     apiVersion?: string | undefined;
     embeddingDeploymentName?: string | undefined;
+    modelFamily?: string | undefined;
     underlyingProvider?: "openai" | "anthropic" | "google" | "mistral" | "azure" | undefined;
 }, {
     apiKey?: string | undefined;
@@ -37,6 +39,7 @@ declare const ProviderConfigSchema: z.ZodObject<{
     deploymentName?: string | undefined;
     apiVersion?: string | undefined;
     embeddingDeploymentName?: string | undefined;
+    modelFamily?: string | undefined;
     underlyingProvider?: "openai" | "anthropic" | "google" | "mistral" | "azure" | undefined;
 }>;
 declare const StorageConfigSchema: z.ZodObject<{
@@ -73,6 +76,7 @@ export declare const ArtemisConfigSchema: z.ZodObject<{
         deploymentName: z.ZodOptional<z.ZodString>;
         apiVersion: z.ZodOptional<z.ZodString>;
         embeddingDeploymentName: z.ZodOptional<z.ZodString>;
+        modelFamily: z.ZodOptional<z.ZodString>;
         underlyingProvider: z.ZodOptional<z.ZodEnum<["openai", "azure", "anthropic", "google", "mistral"]>>;
     }, "strip", z.ZodTypeAny, {
         apiKey?: string | undefined;
@@ -85,6 +89,7 @@ export declare const ArtemisConfigSchema: z.ZodObject<{
         deploymentName?: string | undefined;
         apiVersion?: string | undefined;
         embeddingDeploymentName?: string | undefined;
+        modelFamily?: string | undefined;
         underlyingProvider?: "openai" | "anthropic" | "google" | "mistral" | "azure" | undefined;
     }, {
         apiKey?: string | undefined;
@@ -97,6 +102,7 @@ export declare const ArtemisConfigSchema: z.ZodObject<{
         deploymentName?: string | undefined;
         apiVersion?: string | undefined;
         embeddingDeploymentName?: string | undefined;
+        modelFamily?: string | undefined;
         underlyingProvider?: "openai" | "anthropic" | "google" | "mistral" | "azure" | undefined;
     }>>>;
     storage: z.ZodOptional<z.ZodObject<{
@@ -161,6 +167,7 @@ export declare const ArtemisConfigSchema: z.ZodObject<{
         deploymentName?: string | undefined;
         apiVersion?: string | undefined;
         embeddingDeploymentName?: string | undefined;
+        modelFamily?: string | undefined;
         underlyingProvider?: "openai" | "anthropic" | "google" | "mistral" | "azure" | undefined;
     }> | undefined;
     storage?: {
@@ -195,6 +202,7 @@ export declare const ArtemisConfigSchema: z.ZodObject<{
         deploymentName?: string | undefined;
         apiVersion?: string | undefined;
         embeddingDeploymentName?: string | undefined;
+        modelFamily?: string | undefined;
         underlyingProvider?: "openai" | "anthropic" | "google" | "mistral" | "azure" | undefined;
     }> | undefined;
     storage?: {

package/dist/src/config/schema.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../../src/config/schema.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,QAAA,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAexB,CAAC;AAEH,QAAA,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;EAMvB,CAAC;AAcH,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAS9B,CAAC;AAEH,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAChE,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAClE,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC"}
+{"version":3,"file":"schema.d.ts","sourceRoot":"","sources":["../../../src/config/schema.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,CAAC,EAAE,MAAM,KAAK,CAAC;AAExB,QAAA,MAAM,oBAAoB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAkBxB,CAAC;AAEH,QAAA,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;EAMvB,CAAC;AAcH,eAAO,MAAM,mBAAmB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;EAS9B,CAAC;AAEH,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC;AAChE,MAAM,MAAM,cAAc,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,oBAAoB,CAAC,CAAC;AAClE,MAAM,MAAM,aAAa,GAAG,CAAC,CAAC,KAAK,CAAC,OAAO,mBAAmB,CAAC,CAAC"}

package/dist/src/utils/adapter.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"adapter.d.ts","sourceRoot":"","sources":["../../../src/utils/adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,KAAK,EAAE,YAAY,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AACrE,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AACvD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,MAAM,WAAW,oBAAoB;IACnC,yCAAyC;IACzC,QAAQ,EAAE,MAAM,CAAC;IACjB,2DAA2D;IAC3D,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,2BAA2B;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,0BAA0B;IAC1B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,yCAAyC;IACzC,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,sCAAsC;IACtC,UAAU,CAAC,EAAE,aAAa,GAAG,IAAI,CAAC;IAClC,wCAAwC;IACxC,cAAc,CAAC,EAAE,YAAY,CAAC;IAC9B,qCAAqC;IACrC,WAAW,CAAC,EAAE,YAAY,CAAC;CAC5B;AAOD;;GAEG;AACH,MAAM,WAAW,mBAAmB;IAClC,oDAAoD;IACpD,aAAa,EAAE,aAAa,CAAC;IAC7B,+DAA+D;IAC/D,cAAc,EAAE,cAAc,CAAC;CAChC;AAED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,oBAAoB,GAAG,mBAAmB,CA8ErF;AA+WD;;GAEG;AACH,wBAAgB,yBAAyB,CACvC,WAAW,CAAC,EAAE,MAAM,EACpB,gBAAgB,CAAC,EAAE,MAAM,EACzB,cAAc,CAAC,EAAE,MAAM,GACtB;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,YAAY,CAAA;CAAE,CAK5C;AAED;;GAEG;AACH,wBAAgB,sBAAsB,CACpC,QAAQ,CAAC,EAAE,MAAM,EACjB,aAAa,CAAC,EAAE,MAAM,EACtB,WAAW,CAAC,EAAE,MAAM,GACnB;IAAE,KAAK,EAAE,MAAM,GAAG,SAAS,CAAC;IAAC,MAAM,EAAE,YAAY,GAAG,SAAS,CAAA;CAAE,CAKjE;AAED;;GAEG;AACH,wBAAgB,eAAe,CAC7B,WAAW,CAAC,EAAE,MAAM,EACpB,gBAAgB,CAAC,EAAE,MAAM,EACzB,cAAc,CAAC,EAAE,MAAM,GACtB,MAAM,CAER;AAED;;GAEG;AACH,wBAAgB,YAAY,CAC1B,QAAQ,CAAC,EAAE,MAAM,EACjB,aAAa,CAAC,EAAE,MAAM,EACtB,WAAW,CAAC,EAAE,MAAM,GACnB,MAAM,GAAG,SAAS,CAEpB"}
+{"version":3,"file":"adapter.d.ts","sourceRoot":"","sources":["../../../src/utils/adapter.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AACtD,OAAO,KAAK,EAAE,YAAY,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AACrE,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,kBAAkB,CAAC;AACvD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,kBAAkB,CAAC;AAEtD,MAAM,WAAW,oBAAoB;IACnC,yCAAyC;IACzC,QAAQ,EAAE,MAAM,CAAC;IACjB,2DAA2D;IAC3D,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,2BAA2B;IAC3B,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,0BAA0B;IAC1B,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,yCAAyC;IACzC,cAAc,CAAC,EAAE,cAAc,CAAC;IAChC,sCAAsC;IACtC,UAAU,CAAC,EAAE,aAAa,GAAG,IAAI,CAAC;IAClC,wCAAwC;IACxC,cAAc,CAAC,EAAE,YAAY,CAAC;IAC9B,qCAAqC;IACrC,WAAW,CAAC,EAAE,YAAY,CAAC;CAC5B;AAOD;;GAEG;AACH,MAAM,WAAW,mBAAmB;IAClC,oDAAoD;IACpD,aAAa,EAAE,aAAa,CAAC;IAC7B,+DAA+D;IAC/D,cAAc,EAAE,cAAc,CAAC;CAChC;AAED;;;;;GAKG;AACH,wBAAgB,kBAAkB,CAAC,OAAO,EAAE,oBAAoB,GAAG,mBAAmB,CA8ErF;AAsXD;;GAEG;AACH,wBAAgB,yBAAyB,CACvC,WAAW,CAAC,EAAE,MAAM,EACpB,gBAAgB,CAAC,EAAE,MAAM,EACzB,cAAc,CAAC,EAAE,MAAM,GACtB;IAAE,QAAQ,EAAE,MAAM,CAAC;IAAC,MAAM,EAAE,YAAY,CAAA;CAAE,CAK5C;AAED;;GAEG;AACH,wBAAgB,sBAAsB,CACpC,QAAQ,CAAC,EAAE,MAAM,EACjB,aAAa,CAAC,EAAE,MAAM,EACtB,WAAW,CAAC,EAAE,MAAM,GACnB;IAAE,KAAK,EAAE,MAAM,GAAG,SAAS,CAAC;IAAC,MAAM,EAAE,YAAY,GAAG,SAAS,CAAA;CAAE,CAKjE;AAED;;GAEG;AACH,wBAAgB,eAAe,CAC7B,WAAW,CAAC,EAAE,MAAM,EACpB,gBAAgB,CAAC,EAAE,MAAM,EACzB,cAAc,CAAC,EAAE,MAAM,GACtB,MAAM,CAER;AAED;;GAEG;AACH,wBAAgB,YAAY,CAC1B,QAAQ,CAAC,EAAE,MAAM,EACjB,aAAa,CAAC,EAAE,MAAM,EACtB,WAAW,CAAC,EAAE,MAAM,GACnB,MAAM,GAAG,SAAS,CAEpB"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@artemiskit/cli",
-  "version": "0.2.0",
+  "version": "0.2.3",
   "description": "Command-line interface for ArtemisKit LLM evaluation toolkit",
   "type": "module",
   "license": "Apache-2.0",
@@ -45,11 +45,11 @@
     "test": "bun test"
   },
   "dependencies": {
-    "@artemiskit/adapter-openai": "0.1.7",
-    "@artemiskit/adapter-vercel-ai": "0.1.7",
-    "@artemiskit/core": "0.2.0",
-    "@artemiskit/redteam": "0.2.0",
-    "@artemiskit/reports": "0.2.0",
+    "@artemiskit/adapter-openai": "0.1.10",
+    "@artemiskit/adapter-vercel-ai": "0.1.10",
+    "@artemiskit/core": "0.2.3",
+    "@artemiskit/redteam": "0.2.3",
+    "@artemiskit/reports": "0.2.3",
     "chalk": "^5.3.0",
     "cli-table3": "^0.6.3",
     "commander": "^12.0.0",

package/src/cli.ts

@@ -4,6 +4,7 @@
 
 import { Command } from 'commander';
 import { version } from '../package.json';
+import { baselineCommand } from './commands/baseline';
 import { compareCommand } from './commands/compare';
 import { historyCommand } from './commands/history';
 import { initCommand } from './commands/init';
@@ -45,6 +46,7 @@ export function createCLI(): Command {
 
   program.addCommand(initCommand());
   program.addCommand(runCommand());
+  program.addCommand(baselineCommand());
   program.addCommand(compareCommand());
   program.addCommand(historyCommand());
   program.addCommand(reportCommand());

package/src/commands/baseline.ts

@@ -0,0 +1,473 @@
+/**
+ * Baseline command - Manage baseline runs for regression detection
+ */
+
+import type { BaselineMetadata, BaselineStorageAdapter } from '@artemiskit/core';
+import chalk from 'chalk';
+import { Command } from 'commander';
+import { loadConfig } from '../config/loader.js';
+import { createSpinner, icons, isTTY, padText, renderError } from '../ui/index.js';
+import { createStorage } from '../utils/storage.js';
+
+interface BaselineSetOptions {
+  scenario?: string;
+  tag?: string;
+  config?: string;
+}
+
+interface BaselineListOptions {
+  config?: string;
+  json?: boolean;
+}
+
+interface BaselineRemoveOptions {
+  config?: string;
+  force?: boolean;
+}
+
+/**
+ * Check if storage adapter supports baselines
+ */
+function isBaselineStorage(storage: unknown): storage is BaselineStorageAdapter {
+  return (
+    typeof storage === 'object' &&
+    storage !== null &&
+    'setBaseline' in storage &&
+    'getBaseline' in storage &&
+    'listBaselines' in storage &&
+    'removeBaseline' in storage
+  );
+}
+
+/**
+ * Render baselines table for TTY
+ */
+function renderBaselinesTable(baselines: BaselineMetadata[]): string {
+  const scenarioWidth = 30;
+  const runIdWidth = 16;
+  const rateWidth = 12;
+  const dateWidth = 20;
+  const tagWidth = 12;
+
+  const width =
+    2 + scenarioWidth + 1 + runIdWidth + 1 + rateWidth + 1 + dateWidth + 1 + tagWidth + 2;
+  const border = '═'.repeat(width - 2);
+
+  const formatHeaderRow = () => {
+    const scenarioPad = padText('Scenario', scenarioWidth);
+    const runIdPad = padText('Run ID', runIdWidth);
+    const ratePad = padText('Success Rate', rateWidth, 'right');
+    const datePad = padText('Created', dateWidth, 'right');
+    const tagPad = padText('Tag', tagWidth, 'right');
+    return `║ ${scenarioPad} ${runIdPad} ${ratePad} ${datePad} ${tagPad} ║`;
+  };
+
+  const lines = [
+    `╔${border}╗`,
+    `║${padText('BASELINES', width - 2, 'center')}║`,
+    `╠${border}╣`,
+    formatHeaderRow(),
+    `╟${'─'.repeat(width - 2)}╢`,
+  ];
+
+  for (const baseline of baselines) {
+    const rateColor =
+      baseline.metrics.successRate >= 0.9
+        ? chalk.green
+        : baseline.metrics.successRate >= 0.7
+          ? chalk.yellow
+          : chalk.red;
+
+    const truncScenario =
+      baseline.scenario.length > scenarioWidth - 2
+        ? `${baseline.scenario.slice(0, scenarioWidth - 3)}…`
+        : baseline.scenario;
+    const scenarioPad = padText(truncScenario, scenarioWidth);
+    const runIdPad = padText(baseline.runId, runIdWidth);
+
+    const rateValue = `${(baseline.metrics.successRate * 100).toFixed(1)}%`;
+    const ratePad = padText(rateValue, rateWidth, 'right');
+    const rateColored = rateColor(ratePad);
+
+    const dateObj = new Date(baseline.createdAt);
+    const dateStr = `${dateObj.toLocaleDateString()} ${dateObj.toLocaleTimeString([], { hour: '2-digit', minute: '2-digit' })}`;
+    const datePad = padText(dateStr, dateWidth, 'right');
+
+    const tagPad = padText(baseline.tag || '-', tagWidth, 'right');
+
+    lines.push(`║ ${scenarioPad} ${runIdPad} ${rateColored} ${datePad} ${tagPad} ║`);
+  }
+
+  lines.push(`╚${border}╝`);
+
+  return lines.join('\n');
+}
+
+/**
+ * Render baselines as plain text for CI/non-TTY
+ */
+function renderBaselinesPlain(baselines: BaselineMetadata[]): string {
+  const lines = ['=== BASELINES ===', ''];
+
+  for (const baseline of baselines) {
+    const rate = `${(baseline.metrics.successRate * 100).toFixed(1)}%`;
+    const date = new Date(baseline.createdAt).toLocaleString();
+    const tag = baseline.tag ? ` [${baseline.tag}]` : '';
+    lines.push(`${baseline.scenario}  ${baseline.runId}  ${rate}  ${date}${tag}`);
+  }
+
+  return lines.join('\n');
+}
+
+/**
+ * Create the baseline set subcommand
+ */
+function baselineSetCommand(): Command {
+  const cmd = new Command('set');
+
+  cmd
+    .description('Set a run as the baseline for regression comparison')
+    .argument('<run-id>', 'Run ID to set as baseline')
+    .option('-s, --scenario <name>', 'Override scenario name (defaults to scenario from run)')
+    .option('-t, --tag <tag>', 'Optional tag/description for the baseline')
+    .option('--config <path>', 'Path to config file')
+    .action(async (runId: string, options: BaselineSetOptions) => {
+      const spinner = createSpinner('Setting baseline...');
+      spinner.start();
+
+      try {
+        const config = await loadConfig(options.config);
+        const storage = createStorage({ fileConfig: config });
+
+        if (!isBaselineStorage(storage)) {
+          spinner.fail('Error');
+          console.log();
+          console.log(
+            renderError({
+              title: 'Baselines Not Supported',
+              reason: 'Current storage adapter does not support baseline management',
+              suggestions: [
+                'Use local storage which supports baselines',
+                'Check your storage configuration in artemis.config.yaml',
+              ],
+            })
+          );
+          process.exit(1);
+        }
+
+        const baseline = await storage.setBaseline(options.scenario || '', runId, options.tag);
+
+        spinner.succeed('Baseline set successfully');
+        console.log();
+        console.log(`${icons.passed} ${chalk.bold('Baseline created')}`);
+        console.log();
+        console.log(`  Scenario:     ${chalk.cyan(baseline.scenario)}`);
+        console.log(`  Run ID:       ${chalk.dim(baseline.runId)}`);
+        console.log(
+          `  Success Rate: ${chalk.green(`${(baseline.metrics.successRate * 100).toFixed(1)}%`)}`
+        );
+        console.log(
+          `  Test Cases:   ${baseline.metrics.passedCases}/${baseline.metrics.totalCases} passed`
+        );
+        if (baseline.tag) {
+          console.log(`  Tag:          ${chalk.dim(baseline.tag)}`);
+        }
+        console.log();
+        console.log(
+          chalk.dim('Future runs of this scenario will be compared against this baseline.')
+        );
+      } catch (error) {
+        spinner.fail('Error');
+        console.log();
+        console.log(
+          renderError({
+            title: 'Failed to Set Baseline',
+            reason: (error as Error).message,
+            suggestions: [
+              'Check that the run ID exists',
+              'Run "artemiskit history" to see available runs',
+              'Verify storage configuration',
+            ],
+          })
+        );
+        process.exit(1);
+      }
+    });
+
+  return cmd;
+}
+
+/**
+ * Create the baseline list subcommand
+ */
+function baselineListCommand(): Command {
+  const cmd = new Command('list');
+
+  cmd
+    .description('List all baselines')
+    .option('--config <path>', 'Path to config file')
+    .option('--json', 'Output as JSON')
+    .action(async (options: BaselineListOptions) => {
+      const spinner = createSpinner('Loading baselines...');
+      spinner.start();
+
+      try {
+        const config = await loadConfig(options.config);
+        const storage = createStorage({ fileConfig: config });
+
+        if (!isBaselineStorage(storage)) {
+          spinner.fail('Error');
+          console.log();
+          console.log(
+            renderError({
+              title: 'Baselines Not Supported',
+              reason: 'Current storage adapter does not support baseline management',
+              suggestions: ['Use local storage which supports baselines'],
+            })
+          );
+          process.exit(1);
+        }
+
+        const baselines = await storage.listBaselines();
+        spinner.succeed('Loaded baselines');
+        console.log();
+
+        if (baselines.length === 0) {
+          console.log(chalk.dim('No baselines set.'));
+          console.log();
+          console.log(chalk.dim('Set a baseline with:'));
+          console.log(chalk.dim('  artemiskit baseline set <run-id>'));
+          return;
+        }
+
+        if (options.json) {
+          console.log(JSON.stringify(baselines, null, 2));
+          return;
+        }
+
+        if (isTTY) {
+          console.log(renderBaselinesTable(baselines));
+        } else {
+          console.log(renderBaselinesPlain(baselines));
+        }
+
+        console.log();
+        console.log(
+          chalk.dim(`${baselines.length} baseline${baselines.length === 1 ? '' : 's'} configured`)
+        );
+      } catch (error) {
+        spinner.fail('Error');
+        console.log();
+        console.log(
+          renderError({
+            title: 'Failed to List Baselines',
+            reason: (error as Error).message,
+            suggestions: ['Verify storage configuration'],
+          })
+        );
+        process.exit(1);
+      }
+    });
+
+  return cmd;
+}
+
+/**
+ * Create the baseline remove subcommand
+ */
+function baselineRemoveCommand(): Command {
+  const cmd = new Command('remove');
+
+  cmd
+    .description('Remove a baseline')
+    .argument('<identifier>', 'Run ID of the baseline to remove (or scenario name with --scenario)')
+    .option('--config <path>', 'Path to config file')
+    .option('-f, --force', 'Skip confirmation prompt')
+    .option('-s, --scenario', 'Treat identifier as scenario name instead of run ID')
+    .action(async (identifier: string, options: BaselineRemoveOptions & { scenario?: boolean }) => {
+      const spinner = createSpinner('Removing baseline...');
+
+      try {
+        const config = await loadConfig(options.config);
+        const storage = createStorage({ fileConfig: config });
+
+        if (!isBaselineStorage(storage)) {
+          console.log(
+            renderError({
+              title: 'Baselines Not Supported',
+              reason: 'Current storage adapter does not support baseline management',
+              suggestions: ['Use local storage which supports baselines'],
+            })
+          );
+          process.exit(1);
+        }
+
+        // Check if baseline exists first - by run ID or scenario name
+        const existing = options.scenario
+          ? await storage.getBaseline(identifier)
+          : await storage.getBaselineByRunId(identifier);
+
+        if (!existing) {
+          console.log();
+          console.log(
+            chalk.yellow(
+              options.scenario
+                ? `No baseline found for scenario: ${identifier}`
+                : `No baseline found with run ID: ${identifier}`
+            )
+          );
+          console.log();
+          console.log(chalk.dim('List baselines with:'));
+          console.log(chalk.dim('  artemiskit baseline list'));
+          process.exit(1);
+        }
+
+        // Confirm if not forced
+        if (!options.force && isTTY) {
+          const { promptConfirm } = await import('../ui/index.js');
+          const confirmed = await promptConfirm(
+            `Remove baseline for "${existing.scenario}"? (Run ID: ${existing.runId})`,
+            false
+          );
+          if (!confirmed) {
+            console.log(chalk.dim('Cancelled.'));
+            return;
+          }
+        }
+
+        spinner.start();
+        const removed = options.scenario
+          ? await storage.removeBaseline(identifier)
+          : await storage.removeBaselineByRunId(identifier);
+
+        if (removed) {
+          spinner.succeed('Baseline removed');
+          console.log();
+          console.log(`${icons.passed} Removed baseline for: ${chalk.cyan(existing.scenario)}`);
+        } else {
+          spinner.fail('Baseline not found');
+        }
+      } catch (error) {
+        spinner.fail('Error');
+        console.log();
+        console.log(
+          renderError({
+            title: 'Failed to Remove Baseline',
+            reason: (error as Error).message,
+            suggestions: [
+              'Check the run ID or scenario name',
+              'Run "artemiskit baseline list" to see baselines',
+            ],
+          })
+        );
+        process.exit(1);
+      }
+    });
+
+  return cmd;
+}
+
+/**
+ * Create the baseline get subcommand
+ */
+function baselineGetCommand(): Command {
+  const cmd = new Command('get');
+
+  cmd
+    .description('Get baseline details')
+    .argument('<identifier>', 'Run ID of the baseline (or scenario name with --scenario)')
+    .option('--config <path>', 'Path to config file')
+    .option('--json', 'Output as JSON')
+    .option('-s, --scenario', 'Treat identifier as scenario name instead of run ID')
+    .action(
+      async (
+        identifier: string,
+        options: { config?: string; json?: boolean; scenario?: boolean }
+      ) => {
+        try {
+          const config = await loadConfig(options.config);
+          const storage = createStorage({ fileConfig: config });
+
+          if (!isBaselineStorage(storage)) {
+            console.log(
+              renderError({
+                title: 'Baselines Not Supported',
+                reason: 'Current storage adapter does not support baseline management',
+                suggestions: ['Use local storage which supports baselines'],
+              })
+            );
+            process.exit(1);
+          }
+
+          // Look up by run ID or scenario name
+          const baseline = options.scenario
+            ? await storage.getBaseline(identifier)
+            : await storage.getBaselineByRunId(identifier);
+
+          if (!baseline) {
+            console.log(
+              chalk.yellow(
+                options.scenario
+                  ? `No baseline found for scenario: ${identifier}`
+                  : `No baseline found with run ID: ${identifier}`
+              )
+            );
+            process.exit(1);
+          }
+
+          if (options.json) {
+            console.log(JSON.stringify(baseline, null, 2));
+            return;
+          }
+
+          console.log();
+          console.log(chalk.bold(`Baseline: ${baseline.scenario}`));
+          console.log();
+          console.log(`  Run ID:       ${baseline.runId}`);
+          console.log(`  Created:      ${new Date(baseline.createdAt).toLocaleString()}`);
+          console.log(
+            `  Success Rate: ${chalk.green(`${(baseline.metrics.successRate * 100).toFixed(1)}%`)}`
+          );
+          console.log(
+            `  Test Cases:   ${baseline.metrics.passedCases}/${baseline.metrics.totalCases}`
+          );
+          console.log(`  Latency:      ${baseline.metrics.medianLatencyMs}ms (median)`);
+          console.log(`  Tokens:       ${baseline.metrics.totalTokens.toLocaleString()}`);
+          if (baseline.tag) {
+            console.log(`  Tag:          ${baseline.tag}`);
+          }
+          console.log();
+        } catch (error) {
+          console.log(
+            renderError({
+              title: 'Failed to Get Baseline',
+              reason: (error as Error).message,
+              suggestions: [
+                'Check the run ID or scenario name',
+                'Run "artemiskit baseline list" to see baselines',
+              ],
+            })
+          );
+          process.exit(1);
+        }
+      }
+    );
+
+  return cmd;
+}
+
+/**
+ * Create the main baseline command with subcommands
+ */
+export function baselineCommand(): Command {
+  const cmd = new Command('baseline');
+
+  cmd.description('Manage baseline runs for regression detection');
+
+  cmd.addCommand(baselineSetCommand());
+  cmd.addCommand(baselineListCommand());
+  cmd.addCommand(baselineRemoveCommand());
+  cmd.addCommand(baselineGetCommand());
+
+  return cmd;
+}