@przeprogramowani/10x-cli 1.10.0 → 1.11.0

package/dist/index.mjs

@@ -2271,7 +2271,7 @@ var cac = (name = "") => new CAC(name);
 // package.json
 var package_default = {
   name: "@przeprogramowani/10x-cli",
-  version: "1.10.0",
+  version: "1.11.0",
   description: "Open-source CLI for 10xDevs course content",
   repository: {
     type: "git",
@@ -3521,6 +3521,248 @@ function handleLoginError(ctx, status, code, error) {
   outputError(ctx, code || "auth_error", "Authentication failed.", ExitCodes.ERROR, error ? `Server said: ${error}` : undefined);
 }
 
+// src/lib/bench-client.ts
+var DEFAULT_BENCH_BASE = "https://10xbench.ai";
+var PROD_HOSTNAME2 = "10xbench.ai";
+var DEV_HOSTNAMES2 = new Set(["localhost", "127.0.0.1"]);
+function resolveBenchBase() {
+  const override = process.env["BENCH_BASE_URL"];
+  if (!override)
+    return DEFAULT_BENCH_BASE;
+  let url;
+  try {
+    url = new URL(override);
+  } catch {
+    throw new Error(`BENCH_BASE_URL is not a valid URL: ${JSON.stringify(override)}`);
+  }
+  if (url.pathname !== "/" && url.pathname !== "") {
+    throw new Error(`BENCH_BASE_URL must not include a path (got ${JSON.stringify(url.pathname)})`);
+  }
+  if (url.search !== "" || url.hash !== "") {
+    throw new Error("BENCH_BASE_URL must not include a query string or fragment");
+  }
+  if (url.protocol === "https:" && url.hostname === PROD_HOSTNAME2) {
+    return `${url.protocol}//${url.host}`;
+  }
+  if (url.protocol === "http:" && DEV_HOSTNAMES2.has(url.hostname)) {
+    return `${url.protocol}//${url.host}`;
+  }
+  throw new Error(`BENCH_BASE_URL must be ${DEFAULT_BENCH_BASE} or http://localhost[:port] (got ${JSON.stringify(override)})`);
+}
+var SUPPORTED_LEADERBOARD_SCHEMA = 1;
+var LEADERBOARD_PATH = "/api/leaderboard.json";
+var DEFAULT_TIMEOUT_MS3 = 30000;
+function isLeaderboardResponse(value) {
+  if (value === null || typeof value !== "object")
+    return false;
+  const v2 = value;
+  return typeof v2.schemaVersion === "number" && typeof v2.generatedAt === "string" && typeof v2.totalModels === "number" && Array.isArray(v2.leaderboard);
+}
+async function fetchLeaderboard(options = {}) {
+  const url = `${resolveBenchBase()}${LEADERBOARD_PATH}`;
+  let response;
+  try {
+    response = await fetch(url, {
+      method: "GET",
+      headers: {
+        accept: "application/json",
+        "user-agent": "10x-cli"
+      },
+      signal: options.signal ?? AbortSignal.timeout(DEFAULT_TIMEOUT_MS3)
+    });
+  } catch (err) {
+    const message = err instanceof Error ? err.message : String(err);
+    return { ok: false, status: 0, code: "network_error", error: message };
+  }
+  const text = await response.text();
+  if (!response.ok) {
+    return {
+      ok: false,
+      status: response.status,
+      code: `http_${response.status}`,
+      error: response.statusText || "Leaderboard request failed."
+    };
+  }
+  let parsed;
+  try {
+    parsed = JSON.parse(text);
+  } catch {
+    return {
+      ok: false,
+      status: response.status,
+      code: "bench_unavailable",
+      error: "The leaderboard endpoint returned a non-JSON response."
+    };
+  }
+  if (!isLeaderboardResponse(parsed)) {
+    return {
+      ok: false,
+      status: response.status,
+      code: "bench_unavailable",
+      error: "The leaderboard endpoint returned an unexpected shape."
+    };
+  }
+  return {
+    ok: true,
+    status: response.status,
+    data: parsed,
+    responseHeaders: response.headers,
+    rawBody: text
+  };
+}
+
+// src/lib/format.ts
+function formatReleaseAt(iso, now = new Date) {
+  const date = new Date(iso);
+  if (!Number.isFinite(date.getTime()))
+    return iso;
+  const absolute = date.toLocaleDateString("en-US", {
+    year: "numeric",
+    month: "long",
+    day: "numeric",
+    timeZone: "UTC"
+  });
+  const msPerDay = 24 * 60 * 60 * 1000;
+  const startOfTarget = Date.UTC(date.getUTCFullYear(), date.getUTCMonth(), date.getUTCDate());
+  const startOfNow = Date.UTC(now.getUTCFullYear(), now.getUTCMonth(), now.getUTCDate());
+  const days = Math.round((startOfTarget - startOfNow) / msPerDay);
+  if (days <= 0)
+    return absolute;
+  if (days === 1)
+    return `${absolute} (tomorrow)`;
+  return `${absolute} (in ${days} days)`;
+}
+
+// src/commands/bench.ts
+function registerBenchCommand(cli) {
+  cli.command("bench", "Show the 10xBench top-10 model leaderboard (10xbench.ai)").option("--limit <n>", "Show only the top N models").action(async (options) => {
+    const ctx = resolveContext(options);
+    await runBench(ctx, options);
+  });
+}
+async function runBench(ctx, options) {
+  const limit = parseLimit(ctx, options.limit);
+  verbose(ctx, `fetching leaderboard from ${resolveBenchBase()}`);
+  const result = await fetchLeaderboard();
+  if (!result.ok) {
+    handleBenchError(ctx, result.status, result.code, result.error);
+  }
+  const data = result.data;
+  if (data.schemaVersion !== SUPPORTED_LEADERBOARD_SCHEMA) {
+    outputError(ctx, "unsupported_schema", `The leaderboard now uses format version ${data.schemaVersion}, but this CLI only understands version ${SUPPORTED_LEADERBOARD_SCHEMA}.`, ExitCodes.ERROR, "Run 'npm install -g @przeprogramowani/10x-cli@latest' to update, or visit https://10xbench.ai.");
+  }
+  renderLeaderboard(ctx, data, limit);
+}
+function parseLimit(ctx, raw) {
+  if (raw === undefined)
+    return null;
+  const n = typeof raw === "number" ? raw : Number.parseInt(raw, 10);
+  if (!Number.isInteger(n) || n < 1) {
+    outputError(ctx, "invalid_limit", `'${raw}' is not a valid limit.`, ExitCodes.USAGE, "Pass a positive whole number, for example '10x bench --limit 5'.");
+  }
+  return n;
+}
+function handleBenchError(ctx, status, code, error) {
+  if (status === 0) {
+    outputError(ctx, "network_error", "Could not reach 10xbench.ai.", ExitCodes.ERROR, "Check your internet connection and run '10x bench' again.");
+  }
+  if (code === "bench_unavailable") {
+    outputError(ctx, "bench_unavailable", "The 10xBench leaderboard is temporarily unavailable.", ExitCodes.ERROR, "Try again in a few minutes, or visit https://10xbench.ai directly.");
+  }
+  outputError(ctx, code || "bench_failed", "Failed to load the leaderboard.", ExitCodes.ERROR, error ? `Server said: ${error}` : undefined);
+}
+function colorsEnabled() {
+  return process.env["NO_COLOR"] === undefined && process.stderr.isTTY === true;
+}
+var STYLE = {
+  bold: "1",
+  dim: "2",
+  red: "31",
+  green: "32",
+  yellow: "33",
+  cyan: "36"
+};
+function paint(style, text) {
+  if (!colorsEnabled())
+    return text;
+  return `\x1B[${style}m${text}\x1B[0m`;
+}
+function scoreStyle(percentage) {
+  if (percentage >= 90)
+    return STYLE.green;
+  if (percentage >= 60)
+    return STYLE.yellow;
+  return STYLE.red;
+}
+var BAR_WIDTH = 20;
+function scoreBar(percentage) {
+  const clamped = Math.min(Math.max(percentage, 0), 100);
+  const filled = Math.round(clamped / 100 * BAR_WIDTH);
+  return "█".repeat(filled) + "░".repeat(BAR_WIDTH - filled);
+}
+function renderLeaderboard(ctx, data, limit) {
+  const ranked = limit === null ? data.leaderboard : data.leaderboard.slice(0, limit);
+  if (ctx.json) {
+    output(ctx, "", {
+      source: resolveBenchBase(),
+      generatedAt: data.generatedAt,
+      totalAttempts: data.totalAttempts,
+      totalModels: data.totalModels,
+      models: ranked.map((e2, i) => ({
+        rank: i + 1,
+        model: e2.modelName,
+        modelBaseId: e2.modelBaseId,
+        score: e2.averageScore,
+        maxScore: e2.averageMaxScore,
+        percentage: e2.averagePercentage,
+        attempts: e2.attemptCount,
+        environment: e2.agentEnvironment,
+        averageCost: e2.averageCost
+      }))
+    });
+    return;
+  }
+  if (ranked.length === 0) {
+    output(ctx, "The leaderboard has no results yet.", undefined);
+    return;
+  }
+  const names = ranked.map((e2) => sanitize(e2.modelName));
+  const details = ranked.map((e2) => formatDetail(e2));
+  const nameWidth = Math.max(...names.map((n) => n.length));
+  const detailWidth = Math.max(...details.map((d2) => d2.length));
+  const rankWidth = String(ranked.length).length;
+  const lines = [];
+  lines.push(paint(STYLE.bold, `10xBench — Top ${ranked.length} of ${data.totalModels} models`) + paint(STYLE.dim, `  (${data.totalAttempts} attempts scored)`));
+  lines.push("");
+  for (let i = 0;i < ranked.length; i++) {
+    lines.push(formatRow(ranked[i], i + 1, names[i], details[i], rankWidth, nameWidth, detailWidth));
+  }
+  lines.push("");
+  lines.push(paint(STYLE.dim, `Data as of ${formatReleaseAt(data.generatedAt)} — ${resolveBenchBase()}`));
+  output(ctx, lines.join(`
+`), undefined);
+}
+function formatDetail(entry) {
+  const score = `${entry.averageScore.toFixed(1)}/${entry.averageMaxScore.toFixed(0)}`;
+  const runs = `${entry.attemptCount} run${entry.attemptCount === 1 ? "" : "s"}`;
+  const cost = entry.averageCost === null || entry.averageCost === undefined ? "" : `, ~$${entry.averageCost.toFixed(2)}/run`;
+  return `(${score}, ${runs}${cost})`;
+}
+function formatRow(entry, rank, name, detail, rankWidth, nameWidth, detailWidth) {
+  const style = scoreStyle(entry.averagePercentage);
+  const rankCell = String(rank).padStart(rankWidth);
+  const nameCell = name.padEnd(nameWidth);
+  const pctCell = `${entry.averagePercentage.toFixed(1)}%`.padStart(6);
+  return [
+    `  ${paint(rank <= 3 ? STYLE.bold : STYLE.dim, rankCell)}`,
+    rank <= 3 ? paint(STYLE.bold, nameCell) : nameCell,
+    paint(style, scoreBar(entry.averagePercentage)),
+    paint(style, pctCell),
+    detail.padEnd(detailWidth),
+    paint(STYLE.dim, sanitize(entry.agentEnvironment))
+  ].join("  ");
+}
+
 // src/commands/doctor.ts
 import { existsSync as existsSync2, accessSync, constants } from "node:fs";
 import { join as join2 } from "node:path";
@@ -3732,28 +3974,6 @@ function apiBaseUrl() {
   return resolveApiBase();
 }
 
-// src/lib/format.ts
-function formatReleaseAt(iso, now = new Date) {
-  const date = new Date(iso);
-  if (!Number.isFinite(date.getTime()))
-    return iso;
-  const absolute = date.toLocaleDateString("en-US", {
-    year: "numeric",
-    month: "long",
-    day: "numeric",
-    timeZone: "UTC"
-  });
-  const msPerDay = 24 * 60 * 60 * 1000;
-  const startOfTarget = Date.UTC(date.getUTCFullYear(), date.getUTCMonth(), date.getUTCDate());
-  const startOfNow = Date.UTC(now.getUTCFullYear(), now.getUTCMonth(), now.getUTCDate());
-  const days = Math.round((startOfTarget - startOfNow) / msPerDay);
-  if (days <= 0)
-    return absolute;
-  if (days === 1)
-    return `${absolute} (tomorrow)`;
-  return `${absolute} (in ${days} days)`;
-}
-
 // src/lib/tool-profile.ts
 var SENTINEL_BEGIN = "<!-- BEGIN @przeprogramowani/10x-cli -->";
 var SENTINEL_END = "<!-- END @przeprogramowani/10x-cli -->";
@@ -6297,6 +6517,7 @@ registerGetCommand(cli);
 registerListCommand(cli);
 registerSyncCommand(cli);
 registerDoctorCommand(cli);
+registerBenchCommand(cli);
 cli.help();
 cli.version(package_default.version);
 try {

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@przeprogramowani/10x-cli",
-  "version": "1.10.0",
+  "version": "1.11.0",
   "description": "Open-source CLI for 10xDevs course content",
   "repository": {
     "type": "git",