@kradle/cli 0.0.4 → 0.0.6

package/dist/lib/evaluation/tui.js

@@ -0,0 +1,130 @@
+import { jsx as _jsx, Fragment as _Fragment, jsxs as _jsxs } from "react/jsx-runtime";
+import { Box, render, Text, useInput } from "ink";
+import { useEffect, useMemo, useState } from "react";
+import { STATUS_ICONS } from "./types.js";
+const formatElapsed = (startTime) => {
+    const elapsed = Date.now() - startTime;
+    const seconds = Math.floor(elapsed / 1000);
+    const minutes = Math.floor(seconds / 60);
+    const hours = Math.floor(minutes / 60);
+    if (hours > 0)
+        return `${hours}h ${minutes % 60}m`;
+    if (minutes > 0)
+        return `${minutes}m ${seconds % 60}s`;
+    return `${seconds}s`;
+};
+const getVisibleRows = () => {
+    const terminalHeight = process.stdout.rows || 24;
+    return Math.max(1, terminalHeight - 6 - 1); // Header (3 lines) + footer (3 lines) + 1 line of offset
+};
+const getVisibleColumns = () => {
+    const terminalWidth = process.stdout.columns || 80;
+    return Math.max(1, terminalWidth - 1);
+};
+const RenderRunLine = ({ state, total, isSelected, padding, }) => {
+    const { icon, color } = STATUS_ICONS[state.status] ?? STATUS_ICONS.queued;
+    const indexLabel = `${String(state.index + 1).padStart(padding, " ")}/${total}`;
+    const statusLabel = state.status.padEnd(12);
+    const startTime = state.startTime ?? null;
+    const showElapsed = startTime !== null &&
+        state.status !== "completed" &&
+        state.status !== "finished" &&
+        state.status !== "game_over" &&
+        state.status !== "error";
+    const elapsedLabel = showElapsed && startTime !== null ? formatElapsed(startTime) : null;
+    const agents = state.config.participants.map((p) => p.agent.split(":").pop() ?? p.agent).join(", ");
+    const summary = `${state.config.challenge_slug} (${agents})`;
+    const maxSummaryLength = getVisibleColumns() - indexLabel.length - statusLabel.length - (elapsedLabel ? elapsedLabel.length : 0) - 4; // 4 for the spaces and emoji
+    // biome-ignore lint/style/useTemplate: template literal would be less readable
+    const summaryText = summary.length > maxSummaryLength ? summary.slice(0, maxSummaryLength - 1) + "…" : summary;
+    return (_jsxs(Text, { inverse: isSelected, children: [_jsx(Text, { color: color, children: icon }), " ", indexLabel, " ", _jsx(Text, { color: color, children: statusLabel }), elapsedLabel ? (_jsxs(_Fragment, { children: [" ", _jsx(Text, { dimColor: true, children: elapsedLabel })] })) : null, " ", _jsx(Text, { dimColor: true, children: summaryText })] }));
+};
+const EvaluationUI = ({ evaluationName, states, statusCounts, onQuit, onOpenRun }) => {
+    const [selectedIndex, setSelectedIndex] = useState(0);
+    const [scrollOffset, setScrollOffset] = useState(0);
+    const [tick, setTick] = useState(0); // force elapsed-time updates
+    useEffect(() => {
+        if (states.length === 0) {
+            setSelectedIndex(0);
+            setScrollOffset(0);
+            return;
+        }
+        setSelectedIndex((current) => Math.min(current, states.length - 1));
+    }, [states.length]);
+    useEffect(() => {
+        const rows = getVisibleRows();
+        setScrollOffset((offset) => {
+            if (selectedIndex < offset)
+                return selectedIndex;
+            if (selectedIndex >= offset + rows)
+                return Math.min(selectedIndex - rows + 1, Math.max(0, states.length - rows));
+            return Math.min(offset, Math.max(0, states.length - rows));
+        });
+    }, [selectedIndex, states.length]);
+    useEffect(() => {
+        const interval = setInterval(() => setTick((value) => value + 1), 1000);
+        return () => clearInterval(interval);
+    }, []);
+    useInput((input, key) => {
+        if (input === "q" || (key.ctrl && input === "c")) {
+            onQuit();
+            return;
+        }
+        if (states.length === 0)
+            return;
+        if (key.upArrow || input === "k") {
+            setSelectedIndex((current) => Math.max(0, current - 1));
+        }
+        else if (key.downArrow || input === "j") {
+            setSelectedIndex((current) => Math.min(states.length - 1, current + 1));
+        }
+        else if (input === "o") {
+            onOpenRun(selectedIndex);
+        }
+    });
+    const rowsAvailable = getVisibleRows();
+    const visibleRuns = useMemo(() => states.slice(scrollOffset, scrollOffset + rowsAvailable), [states, scrollOffset, rowsAvailable]);
+    const showScroll = states.length > rowsAvailable;
+    const rangeStart = states.length === 0 ? 0 : scrollOffset + 1;
+    const rangeEnd = Math.min(scrollOffset + rowsAvailable, states.length);
+    const horizontalRule = "─".repeat(Math.min(process.stdout.columns || 80, 80));
+    const padding = states.length.toString().length;
+    return (_jsxs(Box, { flexDirection: "column", children: [_jsx(Text, { bold: true, children: `Evaluation: ${evaluationName}` }), _jsx(Text, { dimColor: true, children: horizontalRule }), _jsx(Text, { dimColor: true, children: "q:quit \u2191/\u2193/j/k:select o:open in browser" }), _jsx(Text, { children: " " }), _jsxs(Box, { flexDirection: "column", children: [visibleRuns.map((state, index) => (_jsx(RenderRunLine, { state: state, total: states.length, isSelected: scrollOffset + index === selectedIndex, padding: padding }, state.index))), visibleRuns.length < rowsAvailable
+                        ? Array.from({ length: rowsAvailable - visibleRuns.length }).map((_, index) => (_jsx(Text, { children: " " }, `empty-${index}`)))
+                        : null] }), showScroll ? _jsx(Text, { dimColor: true, children: `[${rangeStart}-${rangeEnd} of ${states.length}]` }) : _jsx(Text, { children: " " }), _jsx(Text, { dimColor: true, children: horizontalRule }), _jsxs(Text, { children: [_jsx(Text, { children: "Completed: " }), _jsx(Text, { color: "green", children: statusCounts.completed }), _jsx(Text, { children: ` | Active: ` }), _jsx(Text, { color: "yellow", children: statusCounts.active }), _jsx(Text, { children: ` | Queued: ` }), _jsx(Text, { dimColor: true, children: statusCounts.queued }), statusCounts.errors > 0 ? (_jsxs(_Fragment, { children: [_jsx(Text, { children: ` | Errors: ` }), _jsx(Text, { color: "red", children: statusCounts.errors })] })) : null] })] }));
+};
+export class TUI {
+    options;
+    states = [];
+    statusCounts = { completed: 0, active: 0, queued: 0, errors: 0 };
+    app;
+    running = false;
+    constructor(options) {
+        this.options = options;
+    }
+    start() {
+        this.running = true;
+        this.app = render(this.renderApp());
+    }
+    stop() {
+        this.running = false;
+        this.app?.unmount();
+        this.app = undefined;
+    }
+    updateStates(states) {
+        this.states = states;
+        this.rerender();
+    }
+    updateStatusCounts(counts) {
+        this.statusCounts = counts;
+        this.rerender();
+    }
+    rerender() {
+        if (!this.running || !this.app)
+            return;
+        this.app.rerender(this.renderApp());
+    }
+    renderApp() {
+        return (_jsx(EvaluationUI, { evaluationName: this.options.evaluationName, states: this.states, statusCounts: this.statusCounts, onQuit: this.options.onQuit, onOpenRun: this.options.onOpenRun }));
+    }
+}

package/dist/lib/evaluation/types.d.ts

@@ -0,0 +1,128 @@
+import { z } from "zod";
+export declare const ParticipantSchema: z.ZodObject<{
+    agent: z.ZodString;
+    role: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+export type Participant = z.infer<typeof ParticipantSchema>;
+export declare const RunConfigSchema: z.ZodObject<{
+    challenge_slug: z.ZodString;
+    participants: z.ZodArray<z.ZodObject<{
+        agent: z.ZodString;
+        role: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>>;
+}, z.core.$strip>;
+export type RunConfig = z.infer<typeof RunConfigSchema>;
+export declare const ManifestSchema: z.ZodObject<{
+    runs: z.ZodArray<z.ZodObject<{
+        challenge_slug: z.ZodString;
+        participants: z.ZodArray<z.ZodObject<{
+            agent: z.ZodString;
+            role: z.ZodOptional<z.ZodString>;
+        }, z.core.$strip>>;
+    }, z.core.$strip>>;
+    tags: z.ZodOptional<z.ZodArray<z.ZodString>>;
+}, z.core.$strip>;
+export type Manifest = z.infer<typeof ManifestSchema>;
+export type RunStatus = "queued" | "initializing" | "watcher_connected" | "participants_connected" | "started" | "running" | "recovering" | "completed" | "game_over" | "finished" | "error";
+export declare const ProgressEntrySchema: z.ZodObject<{
+    index: z.ZodNumber;
+    status: z.ZodEnum<{
+        error: "error";
+        queued: "queued";
+        initializing: "initializing";
+        watcher_connected: "watcher_connected";
+        participants_connected: "participants_connected";
+        started: "started";
+        running: "running";
+        recovering: "recovering";
+        completed: "completed";
+        game_over: "game_over";
+        finished: "finished";
+    }>;
+    runId: z.ZodOptional<z.ZodString>;
+    startTime: z.ZodOptional<z.ZodNumber>;
+    endTime: z.ZodOptional<z.ZodNumber>;
+    error: z.ZodOptional<z.ZodString>;
+}, z.core.$strip>;
+export type ProgressEntry = z.infer<typeof ProgressEntrySchema>;
+export declare const ProgressSchema: z.ZodObject<{
+    entries: z.ZodArray<z.ZodObject<{
+        index: z.ZodNumber;
+        status: z.ZodEnum<{
+            error: "error";
+            queued: "queued";
+            initializing: "initializing";
+            watcher_connected: "watcher_connected";
+            participants_connected: "participants_connected";
+            started: "started";
+            running: "running";
+            recovering: "recovering";
+            completed: "completed";
+            game_over: "game_over";
+            finished: "finished";
+        }>;
+        runId: z.ZodOptional<z.ZodString>;
+        startTime: z.ZodOptional<z.ZodNumber>;
+        endTime: z.ZodOptional<z.ZodNumber>;
+        error: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>>;
+    lastUpdated: z.ZodNumber;
+}, z.core.$strip>;
+export type Progress = z.infer<typeof ProgressSchema>;
+export declare const RunResultSchema: z.ZodObject<{
+    index: z.ZodNumber;
+    runId: z.ZodString;
+    challenge_slug: z.ZodString;
+    participants: z.ZodArray<z.ZodObject<{
+        agent: z.ZodString;
+        role: z.ZodOptional<z.ZodString>;
+    }, z.core.$strip>>;
+    status: z.ZodString;
+    startTime: z.ZodNumber;
+    endTime: z.ZodNumber;
+    duration: z.ZodNumber;
+    logs: z.ZodOptional<z.ZodArray<z.ZodUnknown>>;
+    summary: z.ZodOptional<z.ZodString>;
+    error: z.ZodOptional<z.ZodString>;
+    outcome: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+}, z.core.$strip>;
+export type RunResult = z.infer<typeof RunResultSchema>;
+export interface RunState {
+    index: number;
+    config: RunConfig;
+    status: RunStatus;
+    runId?: string;
+    startTime?: number;
+    error?: string;
+}
+export interface StatusCounts {
+    completed: number;
+    active: number;
+    queued: number;
+    errors: number;
+}
+export declare const RunStatusResponseSchema: z.ZodObject<{
+    id: z.ZodString;
+    status: z.ZodString;
+    createdAt: z.ZodOptional<z.ZodString>;
+    updatedAt: z.ZodOptional<z.ZodString>;
+    outcome: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+}, z.core.$strip>;
+export type RunStatusResponse = z.infer<typeof RunStatusResponseSchema>;
+export declare const RunLogsResponseSchema: z.ZodObject<{
+    logs: z.ZodArray<z.ZodUnknown>;
+}, z.core.$strip>;
+export type RunLogsResponse = z.infer<typeof RunLogsResponseSchema>;
+export declare const EvaluationMetadataSchema: z.ZodObject<{
+    currentIteration: z.ZodNumber;
+}, z.core.$strip>;
+export type EvaluationMetadata = z.infer<typeof EvaluationMetadataSchema>;
+export interface EvaluationOptions {
+    new: boolean;
+    maxConcurrent: number;
+    openMetabase?: boolean;
+}
+export declare const STATUS_ICONS: Record<RunStatus, {
+    icon: string;
+    color: "white" | "yellow" | "blue" | "magenta" | "cyan" | "green" | "red";
+}>;

package/dist/lib/evaluation/types.js

@@ -0,0 +1,86 @@
+import { z } from "zod";
+// Participant in a run
+export const ParticipantSchema = z.object({
+    agent: z.string(),
+    role: z.string().optional(),
+});
+// Single run configuration
+export const RunConfigSchema = z.object({
+    challenge_slug: z.string(),
+    participants: z.array(ParticipantSchema),
+});
+// Manifest returned by config.ts main()
+export const ManifestSchema = z.object({
+    runs: z.array(RunConfigSchema),
+    tags: z.array(z.string()).optional(),
+});
+// Progress entry for a single run
+export const ProgressEntrySchema = z.object({
+    index: z.number(),
+    status: z.enum([
+        "queued",
+        "initializing",
+        "watcher_connected",
+        "participants_connected",
+        "started",
+        "running",
+        "recovering",
+        "completed",
+        "game_over",
+        "finished",
+        "error",
+    ]),
+    runId: z.string().optional(),
+    startTime: z.number().optional(),
+    endTime: z.number().optional(),
+    error: z.string().optional(),
+});
+// Progress file schema
+export const ProgressSchema = z.object({
+    entries: z.array(ProgressEntrySchema),
+    lastUpdated: z.number(),
+});
+// Run result with logs and summary
+export const RunResultSchema = z.object({
+    index: z.number(),
+    runId: z.string(),
+    challenge_slug: z.string(),
+    participants: z.array(ParticipantSchema),
+    status: z.string(),
+    startTime: z.number(),
+    endTime: z.number(),
+    duration: z.number(),
+    logs: z.array(z.unknown()).optional(),
+    summary: z.string().optional(),
+    error: z.string().optional(),
+    outcome: z.record(z.string(), z.unknown()).optional(),
+});
+// API response schemas for run status
+export const RunStatusResponseSchema = z.object({
+    id: z.string(),
+    status: z.string(),
+    createdAt: z.string().optional(),
+    updatedAt: z.string().optional(),
+    outcome: z.record(z.string(), z.unknown()).optional(),
+});
+export const RunLogsResponseSchema = z.object({
+    logs: z.array(z.unknown()),
+});
+// Evaluation metadata stored in .evaluation.json
+export const EvaluationMetadataSchema = z.object({
+    currentIteration: z.number(),
+});
+// Icons and colors for TUI
+export const STATUS_ICONS = {
+    queued: { icon: "·", color: "white" },
+    initializing: { icon: "○", color: "yellow" },
+    watcher_connected: { icon: "◐", color: "blue" },
+    participants_connected: { icon: "◉", color: "blue" },
+    started: { icon: "▶", color: "magenta" },
+    running: { icon: "▶", color: "magenta" },
+    recovering: { icon: "⟳", color: "cyan" },
+    completed: { icon: "✓", color: "green" },
+    game_over: { icon: "✓", color: "green" },
+    finished: { icon: "✓", color: "green" },
+    error: { icon: "✗", color: "red" },
+};

package/dist/lib/schemas.d.ts

@@ -20,10 +20,13 @@ export declare const ChallengeSchema: z.ZodObject<{
             spectator: "spectator";
         }>;
     }, z.core.$strip>;
+    description: z.ZodOptional<z.ZodString>;
     task: z.ZodString;
     roles: z.ZodRecord<z.ZodString, z.ZodObject<{
         description: z.ZodString;
         specificTask: z.ZodString;
+        minParticipants: z.ZodOptional<z.ZodNumber>;
+        maxParticipants: z.ZodOptional<z.ZodNumber>;
     }, z.core.$strip>>;
     objective: z.ZodObject<{
         fieldName: z.ZodString;
@@ -32,6 +35,7 @@ export declare const ChallengeSchema: z.ZodObject<{
             minimize: "minimize";
         }>;
     }, z.core.$strip>;
+    endStates: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
     creationTime: z.ZodOptional<z.ZodString>;
     updateTime: z.ZodOptional<z.ZodString>;
     creator: z.ZodOptional<z.ZodString>;
@@ -58,10 +62,13 @@ export declare const ChallengesResponseSchema: z.ZodObject<{
                 spectator: "spectator";
             }>;
         }, z.core.$strip>;
+        description: z.ZodOptional<z.ZodString>;
         task: z.ZodString;
         roles: z.ZodRecord<z.ZodString, z.ZodObject<{
             description: z.ZodString;
             specificTask: z.ZodString;
+            minParticipants: z.ZodOptional<z.ZodNumber>;
+            maxParticipants: z.ZodOptional<z.ZodNumber>;
         }, z.core.$strip>>;
         objective: z.ZodObject<{
             fieldName: z.ZodString;
@@ -70,6 +77,7 @@ export declare const ChallengesResponseSchema: z.ZodObject<{
                 minimize: "minimize";
             }>;
         }, z.core.$strip>;
+        endStates: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodString>>;
         creationTime: z.ZodOptional<z.ZodString>;
         updateTime: z.ZodOptional<z.ZodString>;
         creator: z.ZodOptional<z.ZodString>;
@@ -82,6 +90,13 @@ export declare const HumanSchema: z.ZodObject<{
 export declare const RunResponseSchema: z.ZodObject<{
     runIds: z.ZodOptional<z.ZodArray<z.ZodString>>;
 }, z.core.$strip>;
+export declare const RunStatusSchema: z.ZodObject<{
+    id: z.ZodString;
+    status: z.ZodString;
+    createdAt: z.ZodOptional<z.ZodString>;
+    updatedAt: z.ZodOptional<z.ZodString>;
+    outcome: z.ZodOptional<z.ZodRecord<z.ZodString, z.ZodUnknown>>;
+}, z.core.$strip>;
 export declare const UploadUrlResponseSchema: z.ZodObject<{
     uploadUrl: z.ZodString;
     expiresAt: z.ZodString;
@@ -123,5 +138,6 @@ export type ChallengeSchemaType = z.infer<typeof ChallengeSchema>;
 export type ChallengesResponseType = z.infer<typeof ChallengesResponseSchema>;
 export type HumanSchemaType = z.infer<typeof HumanSchema>;
 export type RunResponseType = z.infer<typeof RunResponseSchema>;
+export type RunStatusSchemaType = z.infer<typeof RunStatusSchema>;
 export type AgentSchemaType = z.infer<typeof AgentSchema>;
 export type AgentsResponseType = z.infer<typeof AgentsResponseSchema>;

package/dist/lib/schemas.js

@@ -11,15 +11,19 @@ export const ChallengeSchema = z.object({
         datapack: z.boolean(),
         gameMode: z.enum(["survival", "creative", "adventure", "spectator"]),
     }),
+    description: z.string().optional(),
     task: z.string(),
     roles: z.record(z.string(), z.object({
         description: z.string(),
         specificTask: z.string(),
+        minParticipants: z.number().optional(),
+        maxParticipants: z.number().optional(),
     })),
     objective: z.object({
         fieldName: z.string(),
         direction: z.enum(["maximize", "minimize"]),
     }),
+    endStates: z.record(z.string(), z.string()).optional(),
     creationTime: z.string().optional(),
     updateTime: z.string().optional(),
     creator: z.string().optional(),
@@ -34,6 +38,13 @@ export const HumanSchema = z.object({
 export const RunResponseSchema = z.object({
     runIds: z.array(z.string()).optional(),
 });
+export const RunStatusSchema = z.object({
+    id: z.string(),
+    status: z.string(),
+    createdAt: z.string().optional(),
+    updatedAt: z.string().optional(),
+    outcome: z.record(z.string(), z.unknown()).optional(),
+});
 export const UploadUrlResponseSchema = z.object({
     uploadUrl: z.string(),
     expiresAt: z.string(),

package/dist/lib/utils.d.ts

@@ -87,3 +87,10 @@ export declare function executeCommand(command: string, args: string[], options?
  * @returns A promise that resolves with the stdout of the command.
  */
 export declare function executeNodeCommand(args: string[], config: Config): Promise<string>;
+/**
+ * Open a URL in the default browser.
+ * This is fire-and-forget, so we don't wait for it to complete.
+ *
+ * @param url The URL to open.
+ */
+export declare function openInBrowser(url: string): void;

package/dist/lib/utils.js

@@ -1,4 +1,4 @@
-import { fork, spawn } from "node:child_process";
+import { exec, fork, spawn } from "node:child_process";
 import fs from "node:fs/promises";
 import os from "node:os";
 import path from "node:path";
@@ -168,3 +168,23 @@ export async function executeCommand(command, args, options) {
 export async function executeNodeCommand(args, config) {
     return executeCommand(process.execPath, args, { env: config });
 }
+/**
+ * Open a URL in the default browser.
+ * This is fire-and-forget, so we don't wait for it to complete.
+ *
+ * @param url The URL to open.
+ */
+export function openInBrowser(url) {
+    const platform = process.platform;
+    let command;
+    if (platform === "darwin") {
+        command = `open "${url}"`;
+    }
+    else if (platform === "win32") {
+        command = `start "${url}"`;
+    }
+    else {
+        command = `xdg-open "${url}"`;
+    }
+    exec(command);
+}

package/oclif.manifest.json

@@ -304,7 +304,110 @@
         "challenge",
         "watch.js"
       ]
+    },
+    "evaluation:init": {
+      "aliases": [],
+      "args": {
+        "name": {
+          "description": "Name of the evaluation",
+          "name": "name",
+          "required": true
+        }
+      },
+      "description": "Initialize a new evaluation",
+      "examples": [
+        "<%= config.bin %> <%= command.id %> my-evaluation"
+      ],
+      "flags": {},
+      "hasDynamicHelp": false,
+      "hiddenAliases": [],
+      "id": "evaluation:init",
+      "pluginAlias": "@kradle/cli",
+      "pluginName": "@kradle/cli",
+      "pluginType": "core",
+      "strict": true,
+      "enableJsonFlag": false,
+      "isESM": true,
+      "relativePath": [
+        "dist",
+        "commands",
+        "evaluation",
+        "init.js"
+      ]
+    },
+    "evaluation:list": {
+      "aliases": [],
+      "args": {},
+      "description": "List all evaluations",
+      "examples": [
+        "<%= config.bin %> <%= command.id %>"
+      ],
+      "flags": {},
+      "hasDynamicHelp": false,
+      "hiddenAliases": [],
+      "id": "evaluation:list",
+      "pluginAlias": "@kradle/cli",
+      "pluginName": "@kradle/cli",
+      "pluginType": "core",
+      "strict": true,
+      "enableJsonFlag": false,
+      "isESM": true,
+      "relativePath": [
+        "dist",
+        "commands",
+        "evaluation",
+        "list.js"
+      ]
+    },
+    "evaluation:run": {
+      "aliases": [],
+      "args": {
+        "name": {
+          "description": "Name of the evaluation to run",
+          "name": "name",
+          "required": true
+        }
+      },
+      "description": "Run an evaluation. If the evaluation had an ongoing iteration, it will resume from the last state.",
+      "examples": [
+        "<%= config.bin %> <%= command.id %> my-evaluation",
+        "<%= config.bin %> <%= command.id %> my-evaluation --new",
+        "<%= config.bin %> <%= command.id %> my-evaluation --max-concurrent 10"
+      ],
+      "flags": {
+        "new": {
+          "char": "n",
+          "description": "Start a new iteration of the evaluation",
+          "name": "new",
+          "allowNo": false,
+          "type": "boolean"
+        },
+        "max-concurrent": {
+          "char": "m",
+          "description": "Maximum concurrent runs",
+          "name": "max-concurrent",
+          "default": 5,
+          "hasDynamicHelp": false,
+          "multiple": false,
+          "type": "option"
+        }
+      },
+      "hasDynamicHelp": false,
+      "hiddenAliases": [],
+      "id": "evaluation:run",
+      "pluginAlias": "@kradle/cli",
+      "pluginName": "@kradle/cli",
+      "pluginType": "core",
+      "strict": true,
+      "enableJsonFlag": false,
+      "isESM": true,
+      "relativePath": [
+        "dist",
+        "commands",
+        "evaluation",
+        "run.js"
+      ]
     }
   },
-  "version": "0.0.4"
+  "version": "0.0.6"
 }

package/package.json

@@ -1,6 +1,6 @@
 {
 	"name": "@kradle/cli",
-	"version": "0.0.4",
+	"version": "0.0.6",
 	"description": "Kradle's CLI. Manage challenges, evaluations, agents and more!",
 	"keywords": [
 		"cli"
@@ -24,11 +24,10 @@
 		"build": "rm -rf dist && tsc",
 		"watch": "rm -rf dist && tsc --watch",
 		"lint": "biome check .",
-		"lint:fix": "biome check --write .",
-		"format": "biome format --write .",
+		"format": "biome format --write . && biome check --write .",
 		"prepack": "sh scripts/prepack.sh",
 		"postpack": "sh scripts/postpack.sh",
-		"version": "oclif readme && git add README.md"
+		"version": "oclif manifest && oclif readme && git add README.md"
 	},
 	"dependencies": {
 		"@google-cloud/storage": "^7.17.3",
@@ -38,8 +37,10 @@
 		"chokidar": "^4.0.3",
 		"dotenv": "^17.2.3",
 		"enquirer": "^2.4.1",
+		"ink": "^4.4.1",
 		"listr2": "^9.0.5",
 		"picocolors": "^1.1.1",
+		"react": "^18.2.0",
 		"tar": "^7.5.2",
 		"zod": "^4.1.12"
 	},
@@ -48,6 +49,8 @@
 		"@oclif/test": "^4",
 		"@types/chai": "^4",
 		"@types/node": "^18",
+		"@types/react": "^19.2.7",
+		"@types/react-dom": "^19.2.3",
 		"@types/tar": "^6.1.13",
 		"chai": "^4",
 		"oclif": "^4",
@@ -72,6 +75,9 @@
 			},
 			"agent": {
 				"description": "Manage agents"
+			},
+			"evaluation": {
+				"description": "Manage and run evaluations"
 			}
 		}
 	}