@garuda-agent/garuda 0.18.1 → 0.18.2

package/dist/chunk-CWVXOSAA.js

@@ -1,894 +0,0 @@
-#!/usr/bin/env node
-import {
-  LINK_CANDIDATES,
-  cleanArgs
-} from "./chunk-WK5NE6RY.js";
-import {
-  MANIFEST,
-  TEST_PATH,
-  tail
-} from "./chunk-AOGFYUKG.js";
-import {
-  git,
-  shellWord
-} from "./chunk-QXTHEFKP.js";
-import {
-  DEFAULT_ENV_ALLOWLIST,
-  HostExecutor,
-  denyAround
-} from "./chunk-LMKGSYJC.js";
-
-// src/evals/stats.ts
-var BOOTSTRAP_DRAWS = 1e4;
-var DEFAULT_SEED = 1;
-function seeded(seed) {
-  let a = seed >>> 0;
-  return () => {
-    a = a + 1831565813 >>> 0;
-    let t = a;
-    t = Math.imul(t ^ t >>> 15, t | 1);
-    t ^= t + Math.imul(t ^ t >>> 7, t | 61);
-    return ((t ^ t >>> 14) >>> 0) / 4294967296;
-  };
-}
-function shuffled(items, seed) {
-  const out = [...items];
-  const random = seeded(seed);
-  for (let i = out.length - 1; i > 0; i--) {
-    const j = Math.floor(random() * (i + 1));
-    [out[i], out[j]] = [out[j], out[i]];
-  }
-  return out;
-}
-function byTask(results) {
-  const map = /* @__PURE__ */ new Map();
-  for (const r of results) {
-    let t = map.get(r.id);
-    if (t === void 0) {
-      t = { id: r.id, runs: 0, passes: 0, errors: 0, results: [] };
-      map.set(r.id, t);
-    }
-    t.runs++;
-    if (r.passed) t.passes++;
-    if (r.stopReason === "error") t.errors++;
-    t.results.push(r);
-  }
-  return [...map.values()];
-}
-function passInterval(results, options = {}) {
-  const tasks = byTask(results);
-  const runs = tasks.reduce((s, t) => s + t.runs, 0);
-  if (tasks.length === 0 || runs === 0) return void 0;
-  const seed = options.seed ?? DEFAULT_SEED;
-  const draws = options.draws ?? BOOTSTRAP_DRAWS;
-  const level = options.level ?? 0.95;
-  const random = seeded(seed);
-  const rates = new Float64Array(draws);
-  for (let d = 0; d < draws; d++) {
-    let passes = 0;
-    let n = 0;
-    for (let i = 0; i < tasks.length; i++) {
-      const t = tasks[Math.floor(random() * tasks.length)];
-      passes += t.passes;
-      n += t.runs;
-    }
-    rates[d] = passes / n;
-  }
-  rates.sort();
-  const tail2 = (1 - level) / 2;
-  return {
-    rate: tasks.reduce((s, t) => s + t.passes, 0) / runs,
-    low: quantile(rates, tail2),
-    high: quantile(rates, 1 - tail2),
-    level,
-    draws,
-    seed,
-    tasks: tasks.length
-  };
-}
-function consistency(results) {
-  const repeated = byTask(results).filter((t) => t.runs >= 2);
-  return {
-    repeated: repeated.length,
-    allPassed: repeated.filter((t) => t.passes === t.runs).map((t) => t.id),
-    nonePassed: repeated.filter((t) => t.passes === 0 && t.errors < t.runs).map((t) => t.id),
-    onlyErrors: repeated.filter((t) => t.errors === t.runs).map((t) => t.id)
-  };
-}
-function spread(results) {
-  const scored = results.filter((r) => r.stopReason !== "error");
-  if (scored.length === 0) return void 0;
-  const of = (f) => {
-    const sorted = Float64Array.from(scored.map(f)).sort();
-    return { median: quantile(sorted, 0.5), p95: quantile(sorted, 0.95) };
-  };
-  return {
-    runs: scored.length,
-    steps: of((r) => r.steps),
-    tokens: of((r) => r.tokens),
-    durationMs: of((r) => r.durationMs)
-  };
-}
-function stepsSd(runs) {
-  const steps = runs.filter((r) => r.stopReason !== "error").map((r) => r.steps);
-  if (steps.length < 2) return void 0;
-  const mean = steps.reduce((s, x) => s + x, 0) / steps.length;
-  return Math.sqrt(steps.reduce((s, x) => s + (x - mean) ** 2, 0) / (steps.length - 1));
-}
-function costPerPass(results) {
-  if (results.some((r) => r.costUsd === void 0)) return { kind: "unknown" };
-  const totalUsd = results.reduce((s, r) => s + (r.costUsd ?? 0), 0);
-  const passes = results.filter((r) => r.passed).length;
-  if (passes === 0) return { kind: "none", totalUsd };
-  return {
-    kind: "value",
-    usd: totalUsd / passes,
-    minimum: results.some((r) => r.costIncomplete === true),
-    totalUsd,
-    passes
-  };
-}
-function unexpectedChanges(results) {
-  const measured = results.filter((r) => r.unexpected !== void 0);
-  return {
-    runs: measured.filter((r) => (r.unexpected?.length ?? 0) > 0).length,
-    files: measured.reduce((s, r) => s + (r.unexpected?.length ?? 0), 0),
-    measured: measured.length
-  };
-}
-function quantile(sorted, q) {
-  if (sorted.length === 0) return Number.NaN;
-  const pos = (sorted.length - 1) * q;
-  const lo = Math.floor(pos);
-  const hi = Math.ceil(pos);
-  const a = sorted[lo];
-  const b = sorted[hi];
-  return a + (b - a) * (pos - lo);
-}
-function mcpUse(results) {
-  const measured = results.filter((r) => r.mcpCalls !== void 0);
-  if (measured.length === 0) return void 0;
-  return {
-    calls: measured.reduce((sum, r) => sum + (r.mcpCalls ?? 0), 0),
-    runs: measured.filter((r) => (r.mcpCalls ?? 0) > 0).length,
-    measured: measured.length
-  };
-}
-function mcpUseText(use) {
-  return `${use.calls} in ${use.runs} of ${use.measured} runs${use.runs === 0 ? " (the agent never used the server)" : ""}`;
-}
-
-// src/evals/compare.ts
-import { readFile, stat } from "fs/promises";
-import { join } from "path";
-async function loadReport(path) {
-  const file = (await stat(path)).isDirectory() ? join(path, "report.json") : path;
-  const data = JSON.parse(await readFile(file, "utf8"));
-  return parseReport(data, file);
-}
-function parseReport(data, path) {
-  if (!Array.isArray(data.results)) throw new Error(`${path} is not an eval report (no results).`);
-  const schema = typeof data.schema === "number" ? data.schema : 1;
-  const tasks = schema >= 2 && data.tasks !== null && typeof data.tasks === "object" ? data.tasks : void 0;
-  return {
-    path,
-    schema,
-    model: typeof data.spec === "string" ? data.spec : typeof data.model === "string" ? data.model : void 0,
-    seed: typeof data.seed === "number" ? data.seed : void 0,
-    tasks,
-    results: data.results,
-    mcp: mcpServerOf(data.mcp)
-  };
-}
-function mcpServerOf(value) {
-  if (value === null || typeof value !== "object") return void 0;
-  const server = value.server;
-  return typeof server === "string" ? server : void 0;
-}
-function compareReports(a, b, options = {}) {
-  const problems = [];
-  const notes = [];
-  const idsA = [...new Set(a.results.map((r) => r.id))].sort();
-  const idsB = [...new Set(b.results.map((r) => r.id))].sort();
-  if (idsA.join("\n") !== idsB.join("\n")) {
-    const onlyA = idsA.filter((id) => !idsB.includes(id));
-    const onlyB = idsB.filter((id) => !idsA.includes(id));
-    problems.push(
-      `The reports ran different tasks${onlyA.length > 0 ? `; only in A: ${onlyA.join(", ")}` : ""}${onlyB.length > 0 ? `; only in B: ${onlyB.join(", ")}` : ""}.`
-    );
-  }
-  if (a.tasks !== void 0 && b.tasks !== void 0) {
-    const changed = idsA.filter((id) => idsB.includes(id) && a.tasks?.[id] !== b.tasks?.[id]);
-    if (changed.length > 0) {
-      problems.push(
-        `These tasks changed between the runs (their hashes differ): ${changed.join(", ")}. Run both arms on the same tasks.`
-      );
-    }
-  } else {
-    const old = [a, b].filter((r) => r.tasks === void 0).map((r) => r.path);
-    notes.push(
-      `${old.join(" and ")} ${old.length === 1 ? "has" : "have"} no task hashes (a report before 0.17): Garuda cannot check that the tasks were the same.`
-    );
-  }
-  if (problems.length > 0) return { problems };
-  if (idsA.length === 0) return { problems: ["The reports have no runs."] };
-  const tasksA = new Map(byTask(a.results).map((t) => [t.id, t]));
-  const tasksB = new Map(byTask(b.results).map((t) => [t.id, t]));
-  const rows = idsA.map((id) => {
-    const ta = tasksA.get(id);
-    const tb = tasksB.get(id);
-    const arm = (t) => {
-      const results = t?.results ?? [];
-      return {
-        passes: t?.passes ?? 0,
-        runs: t?.runs ?? 0,
-        costUsd: results.some((r) => r.costUsd === void 0) ? void 0 : results.reduce((s, r) => s + (r.costUsd ?? 0), 0),
-        minimum: results.some((r) => r.costIncomplete === true)
-      };
-    };
-    return { id, a: arm(ta), b: arm(tb) };
-  });
-  const rate = (side, list) => list.reduce((s, r) => s + r[side].passes, 0) / list.reduce((s, r) => s + r[side].runs, 0);
-  const rateA = rate("a", rows);
-  const rateB = rate("b", rows);
-  const seed = options.seed ?? DEFAULT_SEED;
-  const draws = options.draws ?? BOOTSTRAP_DRAWS;
-  const level = options.level ?? 0.95;
-  const random = seeded(seed);
-  const diffs = new Float64Array(draws);
-  for (let d = 0; d < draws; d++) {
-    let pa = 0;
-    let na = 0;
-    let pb = 0;
-    let nb = 0;
-    for (let i = 0; i < rows.length; i++) {
-      const r = rows[Math.floor(random() * rows.length)];
-      pa += r.a.passes;
-      na += r.a.runs;
-      pb += r.b.passes;
-      nb += r.b.runs;
-    }
-    diffs[d] = pb / nb - pa / na;
-  }
-  diffs.sort();
-  const tail2 = (1 - level) / 2;
-  const low = quantile(diffs, tail2);
-  const high = quantile(diffs, 1 - tail2);
-  const verdict = low > 0 ? "B passes more" : high < 0 ? "B passes less" : "no clear difference";
-  if (low === high) {
-    notes.push(
-      "Every draw gave the same pass-rate difference, so the interval shows no uncertainty: more tasks, or tasks that sometimes fail, are needed to measure it."
-    );
-  }
-  if (rows.length < 10) notes.push(roughNote(rows.length));
-  if (a.seed !== void 0 && b.seed !== void 0 && a.seed !== b.seed) {
-    notes.push(
-      `The arms ran with different seeds (${a.seed} and ${b.seed}): their run orders differ.`
-    );
-  }
-  if (a.mcp !== b.mcp) {
-    notes.push(
-      `The arms differ in the MCP server: A ${a.mcp ?? "none"}, B ${b.mcp ?? "none"}. For an MCP A/B, keep every other option the same.`
-    );
-  }
-  const cost = costRatio(rows, seed, draws, level);
-  if (typeof cost === "string") notes.push(cost);
-  else if (cost.skipped > draws * 0.05) {
-    notes.push(
-      `The cost interval skipped ${cost.skipped} of ${draws} draws (an arm had no pass in them).`
-    );
-  }
-  if (rows.some((r) => r.a.minimum || r.b.minimum)) {
-    notes.push("A run's cost is a minimum (no reported usage): the cost ratio is approximate.");
-  }
-  const runsA = rows.map((r) => r.a.runs);
-  const runsB = rows.map((r) => r.b.runs);
-  if (Math.min(...runsA, ...runsB) < 3) {
-    notes.push(
-      "Some tasks have fewer than 3 runs in an arm: repeat each task 3 times or more (--repeat 3)."
-    );
-  }
-  return {
-    comparison: {
-      rows,
-      rateA,
-      rateB,
-      difference: { value: rateB - rateA, low, high, level, draws, seed },
-      verdict,
-      ...typeof cost === "string" ? {} : { costRatio: cost },
-      notes
-    }
-  };
-}
-function costRatio(rows, seed, draws, level) {
-  if (rows.some((r) => r.a.costUsd === void 0 || r.b.costUsd === void 0)) {
-    return "A run's cost is unknown: no cost comparison.";
-  }
-  const per = (list, side) => {
-    const passes = list.reduce((s, r) => s + r[side].passes, 0);
-    const cost = list.reduce((s, r) => s + (r[side].costUsd ?? 0), 0);
-    return passes === 0 ? void 0 : cost / passes;
-  };
-  const a = per(rows, "a");
-  const b = per(rows, "b");
-  if (a === void 0 || b === void 0 || a === 0) {
-    return "An arm has no pass (or no cost): no cost ratio.";
-  }
-  const random = seeded(seed);
-  const ratios = [];
-  const drawn = new Array(rows.length);
-  for (let d = 0; d < draws; d++) {
-    for (let i = 0; i < rows.length; i++) {
-      drawn[i] = rows[Math.floor(random() * rows.length)];
-    }
-    const ca = per(drawn, "a");
-    const cb = per(drawn, "b");
-    if (ca !== void 0 && cb !== void 0 && ca > 0) ratios.push(cb / ca);
-  }
-  const sorted = Float64Array.from(ratios).sort();
-  const tail2 = (1 - level) / 2;
-  const low = quantile(sorted, tail2);
-  const high = quantile(sorted, 1 - tail2);
-  return {
-    value: b / a,
-    low,
-    high,
-    draws,
-    skipped: draws - ratios.length,
-    verdict: low > 1 ? "B costs more per accepted change" : high < 1 ? "B costs less per accepted change" : "no clear difference"
-  };
-}
-function roughNote(tasks) {
-  return `Only ${tasks} task${tasks === 1 ? "" : "s"}: with fewer than 10 tasks the interval is a rough guide.`;
-}
-var pct = (x) => `${Math.round(x * 100)}%`;
-var signed = (x) => {
-  const points = Math.round(Math.abs(x) * 100);
-  return points === 0 ? "0 pts" : `${x > 0 ? "+" : "\u2212"}${points} pts`;
-};
-function formatComparison(a, b, c) {
-  const width = Math.max(4, ...c.rows.map((r) => r.id.length));
-  const lines2 = [
-    `A: ${a.path}${a.model === void 0 ? "" : ` (${a.model})`}`,
-    `B: ${b.path}${b.model === void 0 ? "" : ` (${b.model})`}`,
-    "",
-    `${"Task".padEnd(width)}  A        B`,
-    ...c.rows.map(
-      (r) => `${r.id.padEnd(width)}  ${`${r.a.passes}/${r.a.runs}`.padEnd(7)}  ${r.b.passes}/${r.b.runs}`
-    ),
-    "",
-    `Pass rate over all attempts: A ${pct(c.rateA)}, B ${pct(c.rateB)}; B \u2212 A ${signed(c.difference.value)} (95% paired interval ${signed(c.difference.low)} to ${signed(c.difference.high)}; ${c.difference.draws} draws, seed ${c.difference.seed})`,
-    `Verdict: ${c.verdict}${c.verdict === "no clear difference" ? " (the interval contains 0)" : ""}.`
-  ];
-  for (const [name, report] of [
-    ["A", a],
-    ["B", b]
-  ]) {
-    const cost = costPerPass(report.results);
-    const u = unexpectedChanges(report.results);
-    const s = spread(report.results);
-    const mcp = mcpUse(report.results);
-    lines2.push(
-      `${name}: cost per accepted change ${cost.kind === "unknown" ? "unknown" : cost.kind === "none" ? "no passes" : `${cost.minimum ? "\u2265 " : ""}$${cost.usd.toFixed(4)}`} \xB7 unexpected changes ${u.measured === 0 ? "not measured" : `${u.runs} of ${u.measured} runs`} \xB7 median time ${s === void 0 ? "\u2013" : `${Math.round(s.durationMs.median / 1e3)} s`}${mcp === void 0 ? "" : ` \xB7 MCP tool calls ${mcpUseText(mcp)}`}`
-    );
-  }
-  if (c.costRatio !== void 0) {
-    const x = (v) => `\xD7${v.toFixed(2)}`;
-    lines2.push(
-      `Cost per accepted change: B/A ${x(c.costRatio.value)} (95% paired interval ${x(c.costRatio.low)} to ${x(c.costRatio.high)}). Verdict: ${c.costRatio.verdict}${c.costRatio.verdict === "no clear difference" ? " (the interval contains \xD71.00)" : ""}.`
-    );
-  }
-  for (const note of c.notes) lines2.push(`Note: ${note}`);
-  return lines2.join("\n");
-}
-
-// src/evals/toolchains.ts
-import { mkdtemp, rm } from "fs/promises";
-import { tmpdir } from "os";
-import { join as join3 } from "path";
-
-// src/evals/files.ts
-import { mkdirSync } from "fs";
-import { writeFile } from "fs/promises";
-import { dirname, join as join2 } from "path";
-async function writeFiles(root, files) {
-  for (const [path, content] of Object.entries(files)) {
-    const full = join2(root, path);
-    mkdirSync(dirname(full), { recursive: true });
-    await writeFile(full, content);
-  }
-}
-
-// src/evals/projects.ts
-var MAVEN_POM = `<?xml version="1.0" encoding="UTF-8"?>
-<project xmlns="http://maven.apache.org/POM/4.0.0"
-         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
-         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 https://maven.apache.org/xsd/maven-4.0.0.xsd">
-  <modelVersion>4.0.0</modelVersion>
-  <groupId>dev.garuda.eval</groupId>
-  <artifactId>eval</artifactId>
-  <version>1.0.0</version>
-
-  <properties>
-    <maven.compiler.release>17</maven.compiler.release>
-    <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
-  </properties>
-
-  <dependencies>
-    <dependency>
-      <groupId>org.junit.jupiter</groupId>
-      <artifactId>junit-jupiter</artifactId>
-      <version>5.11.4</version>
-      <scope>test</scope>
-    </dependency>
-  </dependencies>
-
-  <build>
-    <plugins>
-      <plugin>
-        <groupId>org.apache.maven.plugins</groupId>
-        <artifactId>maven-resources-plugin</artifactId>
-        <version>3.3.1</version>
-      </plugin>
-      <plugin>
-        <groupId>org.apache.maven.plugins</groupId>
-        <artifactId>maven-compiler-plugin</artifactId>
-        <version>3.13.0</version>
-      </plugin>
-      <plugin>
-        <groupId>org.apache.maven.plugins</groupId>
-        <artifactId>maven-surefire-plugin</artifactId>
-        <version>3.5.2</version>
-      </plugin>
-    </plugins>
-  </build>
-</project>
-`;
-var MAVEN_CHECK = "test ! -e .mvn && mvn -B -q -o test";
-var PYPROJECT = `[project]
-name = "evalpkg"
-version = "0.1.0"
-requires-python = ">=3.9"
-
-[tool.pytest.ini_options]
-pythonpath = ["src"]
-testpaths = ["tests"]
-`;
-var PYTEST_CHECK = "python3 -m pytest -q -c pyproject.toml --noconftest -p no:cacheprovider";
-function isProtectedPath(path) {
-  return /^(test|tests|src\/test)\//.test(path) || path === "pom.xml" || path === "pyproject.toml";
-}
-var lines = (...l) => `${l.join("\n")}
-`;
-
-// src/evals/toolchains.ts
-var TOOLCHAIN_ENV = ["JAVA_HOME", "GRADLE_USER_HOME", "VIRTUAL_ENV"];
-var TOOLCHAINS = {
-  maven: {
-    id: "maven",
-    title: "Maven with JUnit 5 in ~/.m2",
-    files: {
-      "pom.xml": MAVEN_POM,
-      "src/main/java/probe/Probe.java": "package probe;\n\npublic final class Probe {\n  public static int one() {\n    return 1;\n  }\n}\n",
-      "src/test/java/probe/ProbeTest.java": "package probe;\n\nimport static org.junit.jupiter.api.Assertions.assertEquals;\n\nimport org.junit.jupiter.api.Test;\n\nclass ProbeTest {\n  @Test\n  void one() {\n    assertEquals(1, Probe.one());\n  }\n}\n"
-    },
-    probe: "mvn -B -q -o test",
-    prepare: "mvn -B -q test",
-    hint: "The Java tasks need a JDK (17 or later), Maven, and JUnit 5 in ~/.m2. Run `garuda eval --prepare java` once: it downloads the plugins and JUnit with Maven (network)."
-  },
-  pytest: {
-    id: "pytest",
-    title: "Python 3 with pytest",
-    files: {
-      "pyproject.toml": PYPROJECT,
-      "src/probe/__init__.py": "def one():\n    return 1\n",
-      "tests/test_probe.py": "from probe import one\n\n\ndef test_one():\n    assert one() == 1\n"
-    },
-    probe: "python3 -m pytest -q -p no:cacheprovider",
-    hint: "The Python tasks need python3 with pytest 7 or later. Install it, for example with `python3 -m pip install --user pytest`."
-  }
-};
-var PROBE_TIMEOUT_MS = 12e4;
-var PREPARE_TIMEOUT_MS = 10 * 6e4;
-async function checkToolchains(ids) {
-  return Promise.all(
-    [...new Set(ids)].map(async (id) => {
-      const toolchain = TOOLCHAINS[id];
-      const r = await inScratch(toolchain, toolchain.probe, PROBE_TIMEOUT_MS);
-      return { id, ...r };
-    })
-  );
-}
-async function prepareToolchain(id) {
-  const toolchain = TOOLCHAINS[id];
-  if (toolchain.prepare === void 0) return { ok: false, output: toolchain.hint };
-  return inScratch(toolchain, toolchain.prepare, PREPARE_TIMEOUT_MS);
-}
-function toolchainId(name) {
-  const aliases = {
-    java: "maven",
-    maven: "maven",
-    python: "pytest",
-    pytest: "pytest"
-  };
-  return aliases[name.toLowerCase()];
-}
-async function inScratch(toolchain, command, timeoutMs) {
-  const root = await mkdtemp(join3(tmpdir(), `garuda-toolchain-${toolchain.id}-`));
-  try {
-    await writeFiles(root, toolchain.files);
-    const r = await new HostExecutor().run(command, {
-      root,
-      sandbox: false,
-      writePaths: [root],
-      denyWritePaths: [],
-      denyReadPaths: [],
-      network: true,
-      envAllowlist: [...DEFAULT_ENV_ALLOWLIST, ...TOOLCHAIN_ENV],
-      timeoutMs,
-      maxOutputBytes: 4e3
-    });
-    const output = `${r.stdout.text}${r.stderr.text}`.trim();
-    return { ok: r.exitCode === 0 && !r.timedOut, output };
-  } catch (error) {
-    return { ok: false, output: error.message };
-  } finally {
-    await rm(root, { recursive: true, force: true });
-  }
-}
-
-// src/evals/repoTasks.ts
-import { mkdtempSync, realpathSync } from "fs";
-import { mkdir, readFile as readFile2, rm as rm2, symlink, writeFile as writeFile2 } from "fs/promises";
-import { tmpdir as tmpdir2 } from "os";
-import { dirname as dirname2, join as join4, sep } from "path";
-import { z } from "zod";
-var REPO_SUITE_FILE = join4(".garuda", "evals", "repo-suite.json");
-var DEFAULT_COMMITS = 200;
-var DEFAULT_MAX_TASKS = 30;
-var MAX_TASK_FILES = 10;
-var BUILD_TEST_TIMEOUT_MS = 5 * 6e4;
-var testFile = z.object({ path: z.string(), content: z.string().nullable() });
-var taskSchema = z.object({
-  id: z.string(),
-  sha: z.string(),
-  base: z.string(),
-  subject: z.string(),
-  prompt: z.string(),
-  tests: z.array(testFile),
-  /** The check: the test command for this task's test files. Absent: the suite's test command. */
-  check: z.string().optional()
-});
-var suiteSchema = z.object({
-  version: z.literal(1),
-  root: z.string(),
-  head: z.string(),
-  createdAt: z.string(),
-  testCommand: z.string(),
-  tasks: z.array(taskSchema)
-});
-var DOC_PATH = /\.(md|mdx|rst|txt|adoc)$|(^|\/)docs?\//i;
-var PACKAGE_JSON = /(^|\/)package\.json$/;
-function skipReason(commit) {
-  const { files } = commit;
-  if (files.length === 0) return "no files";
-  const code = files.filter((f) => !DOC_PATH.test(f.path));
-  if (code.length > MAX_TASK_FILES) return `more than ${MAX_TASK_FILES} files`;
-  if (files.some((f) => f.added === void 0)) return "a binary file";
-  if (files.some((f) => MANIFEST.test(f.path) && !PACKAGE_JSON.test(f.path))) {
-    return "dependencies changed";
-  }
-  const tests = code.filter((f) => TEST_PATH.test(f.path));
-  if (tests.length === 0) return "no test change";
-  if (tests.length === code.length) return "only tests changed";
-  return void 0;
-}
-function dependencyChange(diff) {
-  for (const line of diff.split("\n")) {
-    if (!/^[+-]/.test(line) || /^(\+\+\+|---)/.test(line)) continue;
-    if (/[dD]ependencies"/.test(line)) return true;
-    const entry = /^[+-]\s*"([^"]+)":\s*"[~^<>=]*\d/.exec(line);
-    if (entry !== null && entry[1] !== "version") return true;
-  }
-  return false;
-}
-function scopedTestCommand(command, files) {
-  const list = files.map(shellWord).join(" ");
-  if (command.includes("{files}")) return command.replaceAll("{files}", list);
-  if (files.length === 0) return command;
-  const c = command.trim();
-  if (/^npm (run )?test$/.test(c)) return `${c} -- ${list}`;
-  if (/^(pnpm|yarn)( run)? test$/.test(c) || /^node --test$/.test(c) || /^(npx |pnpm exec )?(vitest|jest)( run)?$/.test(c) || /-m pytest( -q)?$|^pytest( -q)?$/.test(c)) {
-    return `${c} ${list}`;
-  }
-  return c;
-}
-function repoPrompt(commit, tests) {
-  return [
-    commit.subject.trim(),
-    ...commit.body.trim() === "" ? [] : ["", commit.body.trim()],
-    "",
-    `The tests for this change are already in the project: ${tests.join(", ")}. Make the change so that they pass. Do not change these test files.`
-  ].join("\n");
-}
-async function recentCommits(executor, root, options = {}) {
-  const log = await git(executor, root, [
-    "log",
-    "--no-merges",
-    "--format=%H%x00%s%x00%b%x1e",
-    `-n${options.commits ?? DEFAULT_COMMITS}`,
-    ...options.since === void 0 ? [] : [`--since=${options.since}`],
-    "HEAD"
-  ]);
-  const out = [];
-  for (const entry of log.stdout.split("")) {
-    const [sha = "", subject = "", body = ""] = entry.replace(/^\n/, "").split("\0");
-    if (!/^[0-9a-f]{40,64}$/.test(sha)) continue;
-    const stat2 = await git(executor, root, ["show", "--no-renames", "--numstat", "--format=", sha]);
-    const files = stat2.stdout.split("\n").map((line) => /^(\d+|-)\t(\d+|-)\t(.+)$/.exec(line)).filter((m) => m !== null).map(
-      (m) => m[1] === "-" ? { path: m[3] } : { path: m[3], added: Number(m[1]), removed: Number(m[2]) }
-    );
-    out.push({ sha, subject, body, files });
-  }
-  return out;
-}
-async function taskWorktree(executor, root, commit) {
-  const dir = realpathSync(mkdtempSync(join4(tmpdir2(), "garuda-repo-task-")));
-  await git(executor, root, ["worktree", "add", "--detach", "--quiet", dir, commit]);
-  for (const name of LINK_CANDIDATES) {
-    try {
-      await symlink(join4(root, name), join4(dir, name), "dir");
-    } catch {
-    }
-  }
-  return dir;
-}
-async function isolatedTaskCopy(executor, root, commit) {
-  const dir = realpathSync(mkdtempSync(join4(tmpdir2(), "garuda-repo-task-")));
-  const indexDir = mkdtempSync(join4(tmpdir2(), "garuda-repo-index-"));
-  try {
-    const env = { GIT_INDEX_FILE: join4(indexDir, "index") };
-    await git(executor, root, ["read-tree", commit], { env });
-    await git(executor, root, ["checkout-index", "--all", "--force", `--prefix=${dir}/`], { env });
-  } finally {
-    await rm2(indexDir, { recursive: true, force: true });
-  }
-  await git(executor, dir, ["init", "--quiet"]);
-  await git(executor, dir, ["add", "--all"]);
-  await git(executor, dir, [
-    "-c",
-    "user.name=Garuda eval",
-    "-c",
-    "user.email=eval@garuda.invalid",
-    "commit",
-    "--quiet",
-    "--no-verify",
-    "--message",
-    "The task's start"
-  ]);
-  for (const name of LINK_CANDIDATES) {
-    try {
-      await symlink(join4(root, name), join4(dir, name), "dir");
-    } catch {
-    }
-  }
-  return dir;
-}
-function repoTaskReadDenials(root, home) {
-  const links = LINK_CANDIDATES.map((name) => join4(root, name));
-  const inHome = root === home || root.startsWith(`${home}${sep}`);
-  return [
-    ...inHome ? denyAround(home, links, true) : [],
-    ...inHome ? [] : denyAround(root, links)
-  ];
-}
-async function removeTaskWorktree(executor, root, dir) {
-  await git(executor, root, ["worktree", "remove", "--force", dir], { check: false });
-  await rm2(dir, { recursive: true, force: true });
-  await git(executor, root, ["worktree", "prune"], { check: false });
-}
-async function applyTests(dir, tests) {
-  for (const t of tests) {
-    const file = join4(dir, t.path);
-    if (t.content === null) await rm2(file, { force: true });
-    else {
-      await mkdir(dirname2(file), { recursive: true });
-      await writeFile2(file, t.content);
-    }
-  }
-}
-async function runTests(executor, dir, command) {
-  const result = await executor.run(command, {
-    root: dir,
-    sandbox: false,
-    writePaths: [dir],
-    denyWritePaths: [],
-    denyReadPaths: [],
-    network: false,
-    envAllowlist: [...DEFAULT_ENV_ALLOWLIST, ...TOOLCHAIN_ENV],
-    timeoutMs: BUILD_TEST_TIMEOUT_MS,
-    maxOutputBytes: 2e4
-  });
-  const output = `${result.stdout.text}${result.stderr.text}`;
-  return {
-    ok: result.exitCode === 0 && !result.timedOut,
-    tail: result.timedOut ? "(stopped: too long)" : tail(output, 15)
-  };
-}
-async function buildRepoSuite(executor, root, options) {
-  const say = options.onProgress ?? (() => {
-  });
-  const head = (await git(executor, root, ["rev-parse", "HEAD"])).stdout.trim();
-  const commits = await recentCommits(executor, root, {
-    ...options.commits === void 0 ? {} : { commits: options.commits },
-    ...options.since === void 0 ? {} : { since: options.since }
-  });
-  const skipped = {};
-  const skip = (why) => {
-    skipped[why] = (skipped[why] ?? 0) + 1;
-  };
-  const tasks = [];
-  let checked = 0;
-  let failure;
-  const max = options.maxTasks ?? DEFAULT_MAX_TASKS;
-  for (const commit of commits) {
-    if (tasks.length >= max) break;
-    const why = skipReason(commit);
-    if (why !== void 0) {
-      skip(why);
-      continue;
-    }
-    const parent = await git(
-      executor,
-      root,
-      ["rev-parse", "--verify", "--quiet", `${commit.sha}^`],
-      {
-        check: false
-      }
-    );
-    if (parent.exitCode !== 0) {
-      skip("the first commit (no parent)");
-      continue;
-    }
-    if (commit.files.some((f) => PACKAGE_JSON.test(f.path))) {
-      const diff = await git(
-        executor,
-        root,
-        ["show", "--format=", commit.sha, "--", "package.json"],
-        {
-          check: false
-        }
-      );
-      if (dependencyChange(diff.stdout)) {
-        skip("dependencies changed");
-        continue;
-      }
-    }
-    const base = parent.stdout.trim();
-    const testPaths = commit.files.filter((f) => TEST_PATH.test(f.path) && !DOC_PATH.test(f.path)).map((f) => f.path);
-    const tests = [];
-    for (const path of testPaths) {
-      const shown = await git(executor, root, ["show", `${commit.sha}:${path}`], { check: false });
-      tests.push({ path, content: shown.exitCode === 0 ? shown.stdout : null });
-    }
-    const runnable = tests.filter((t) => t.content !== null).map((t) => t.path);
-    if (runnable.length === 0) {
-      skip("only deletes tests");
-      continue;
-    }
-    const check = scopedTestCommand(options.testCommand, runnable);
-    checked++;
-    say(`checking ${commit.sha.slice(0, 7)} ${commit.subject}`);
-    const dir = await taskWorktree(executor, root, base);
-    try {
-      await applyTests(dir, tests);
-      if ((await runTests(executor, dir, check)).ok) {
-        skip("the tests already pass at the parent");
-        continue;
-      }
-      await git(executor, dir, ["checkout", "--quiet", "--", "."], { check: false });
-      await git(executor, dir, cleanArgs(), { check: false });
-      await git(executor, dir, ["checkout", "--quiet", "--detach", commit.sha]);
-      const atCommit = await runTests(executor, dir, check);
-      if (!atCommit.ok) {
-        skip("the tests fail at the commit");
-        failure ??= { sha: commit.sha, command: check, tail: atCommit.tail };
-        continue;
-      }
-    } finally {
-      await removeTaskWorktree(executor, root, dir);
-    }
-    tasks.push({
-      id: `repo-${commit.sha.slice(0, 7)}`,
-      sha: commit.sha,
-      base,
-      subject: commit.subject,
-      prompt: repoPrompt(commit, testPaths),
-      tests,
-      check
-    });
-    say(`kept ${commit.sha.slice(0, 7)} (${tasks.length}/${max})`);
-  }
-  return {
-    suite: {
-      version: 1,
-      root,
-      head,
-      createdAt: (/* @__PURE__ */ new Date()).toISOString(),
-      testCommand: options.testCommand,
-      tasks
-    },
-    checked,
-    skipped,
-    ...failure === void 0 ? {} : { failure }
-  };
-}
-async function saveRepoSuite(suite) {
-  const file = join4(suite.root, REPO_SUITE_FILE);
-  await mkdir(dirname2(file), { recursive: true });
-  await writeFile2(file, `${JSON.stringify(suite, null, 2)}
-`, { mode: 384 });
-  return file;
-}
-async function loadRepoSuite(root) {
-  let text;
-  try {
-    text = await readFile2(join4(root, REPO_SUITE_FILE), "utf8");
-  } catch {
-    return void 0;
-  }
-  const parsed = suiteSchema.safeParse(JSON.parse(text));
-  if (!parsed.success) throw new Error(`${REPO_SUITE_FILE}: ${z.prettifyError(parsed.error)}`);
-  return parsed.data;
-}
-function repoEvalTasks(suite) {
-  return suite.tasks.map((t) => ({
-    id: t.id,
-    title: t.subject,
-    prompt: t.prompt,
-    files: {},
-    check: t.check ?? suite.testCommand,
-    protect: t.tests.filter((f) => f.content !== null).map((f) => f.path),
-    solution: {},
-    repo: { root: suite.root, base: t.base, sha: t.sha, tests: t.tests }
-  }));
-}
-
-export {
-  DEFAULT_SEED,
-  shuffled,
-  byTask,
-  passInterval,
-  consistency,
-  spread,
-  stepsSd,
-  costPerPass,
-  unexpectedChanges,
-  mcpUse,
-  mcpUseText,
-  loadReport,
-  compareReports,
-  roughNote,
-  formatComparison,
-  writeFiles,
-  MAVEN_POM,
-  MAVEN_CHECK,
-  PYPROJECT,
-  PYTEST_CHECK,
-  isProtectedPath,
-  lines,
-  TOOLCHAIN_ENV,
-  TOOLCHAINS,
-  checkToolchains,
-  prepareToolchain,
-  toolchainId,
-  DEFAULT_COMMITS,
-  isolatedTaskCopy,
-  repoTaskReadDenials,
-  applyTests,
-  buildRepoSuite,
-  saveRepoSuite,
-  loadRepoSuite,
-  repoEvalTasks
-};
-//# sourceMappingURL=chunk-CWVXOSAA.js.map