self-bench 0.3.5 → 0.3.8

package/dist/review/assets/index-DgSLzX8l.css

@@ -1 +0,0 @@
-:root{--bg: #f4f5f7;--surface: #ffffff;--border: #d8dce3;--text: #1a1d23;--muted: #6b7280;--accent: #2563eb;--danger: #dc2626;--success: #059669;--code-bg: #f1f3f5;--font: -apple-system, BlinkMacSystemFont, "Segoe UI", system-ui, sans-serif;--mono: "SF Mono", "Cascadia Code", "Fira Code", monospace}@media(prefers-color-scheme:dark){:root{--bg: #0d1117;--surface: #161b22;--border: #30363d;--text: #e6edf3;--muted: #8b949e;--accent: #58a6ff;--danger: #f85149;--success: #3fb950;--code-bg: #1c2128}}*{margin:0;padding:0;box-sizing:border-box}body{font-family:var(--font);background:var(--bg);color:var(--text);font-size:14px;line-height:1.5}.app-shell{min-height:100vh;display:flex;flex-direction:column}.topbar{display:flex;justify-content:space-between;align-items:center;padding:16px 24px;border-bottom:1px solid var(--border);background:var(--surface)}.eyebrow{font-size:11px;font-weight:600;letter-spacing:.08em;text-transform:uppercase;color:var(--muted)}.topbar h1{font-size:18px;font-weight:600;margin-top:2px}.topbar-actions{display:flex;gap:8px;align-items:center}.token-input{padding:6px 10px;border:1px solid var(--border);border-radius:6px;background:var(--surface);color:var(--text);font-size:12px;width:180px}.error-banner{padding:12px 24px;background:var(--danger);color:#fff;font-size:13px}.workspace{flex:1;display:grid;grid-template-columns:280px 1fr;gap:16px;padding:16px 24px}.surface{background:var(--surface);border:1px solid var(--border);border-radius:8px}.run-list{padding:16px;overflow-y:auto;max-height:calc(100vh - 100px)}.section-heading{display:flex;justify-content:space-between;align-items:baseline;margin-bottom:12px;padding-bottom:8px;border-bottom:1px solid var(--border)}.section-heading h2{font-size:13px;font-weight:600}.section-heading span{font-size:11px;color:var(--muted)}.run-row{width:100%;display:flex;flex-direction:column;gap:2px;padding:10px 12px;margin-bottom:6px;border:1px solid var(--border);border-radius:6px;background:var(--surface);cursor:pointer;text-align:left;font:inherit;color:inherit}.run-row:hover{border-color:var(--accent)}.run-row.selected{border-color:var(--accent);box-shadow:0 0 0 1px var(--accent)}.run-row strong{font-size:13px}.run-row span{font-size:11px;color:var(--muted)}.run-row small{font-size:10px;color:var(--muted)}.content{display:flex;flex-direction:column;gap:16px}.empty{padding:64px 32px;text-align:center;display:flex;flex-direction:column;align-items:center;gap:8px;min-height:400px;justify-content:center}.empty h2{font-size:16px;font-weight:600}.empty p{color:var(--muted)}.summary{display:flex;justify-content:space-between;align-items:center;padding:16px 20px}.summary h2{font-size:16px;font-weight:600}.summary .muted{font-size:12px;color:var(--muted);margin-top:2px}.review-grid{display:grid;grid-template-columns:240px 1fr;gap:16px}.task-list{padding:16px;max-height:calc(100vh - 280px);overflow-y:auto}.task-row{width:100%;display:flex;justify-content:space-between;align-items:center;padding:8px 10px;margin-bottom:4px;border:1px solid var(--border);border-radius:5px;background:var(--surface);cursor:pointer;text-align:left;font:inherit;color:inherit;font-size:12px}.task-row:hover,.task-row.selected{border-color:var(--accent)}.task-row.removed{opacity:.5}.task-row.removed strong{text-decoration:line-through}.task-row span{font-size:10px;color:var(--muted)}.task-detail{padding:20px 24px;max-height:calc(100vh - 280px);overflow-y:auto}.detail-header{display:flex;justify-content:space-between;align-items:flex-start;margin-bottom:20px;padding-bottom:16px;border-bottom:1px solid var(--border)}.detail-header h2{font-size:18px;font-weight:600}.detail-block{margin-bottom:20px}.detail-block h3{font-size:12px;font-weight:600;letter-spacing:.04em;text-transform:uppercase;color:var(--muted);margin-bottom:6px}.detail-block pre{padding:12px 16px;background:var(--code-bg);border-radius:6px;font-family:var(--mono);font-size:12px;white-space:pre-wrap;word-break:break-all;max-height:400px;overflow-y:auto;line-height:1.6}.patch-files{display:grid;gap:10px;padding:8px;max-height:600px;overflow:auto;background:var(--code-bg);border-radius:6px}.patch-files>div{overflow:hidden;border:1px solid var(--border);border-radius:4px}.button{padding:7px 14px;border:1px solid var(--border);border-radius:6px;background:var(--surface);color:var(--text);font-size:12px;font-weight:500;cursor:pointer;font-family:var(--font)}.button:hover{border-color:var(--accent);color:var(--accent)}.button.primary{background:var(--accent);color:#fff;border-color:var(--accent)}.button.primary:hover{opacity:.9;color:#fff}.button.danger{color:var(--danger);border-color:var(--danger)}.button.danger:hover{background:var(--danger);color:#fff}.button.undo{color:var(--success);border-color:var(--success)}.button.undo:hover{background:var(--success);color:#fff}.muted{color:var(--muted)}

package/dist/sandbox-author.d.ts

@@ -1,3 +0,0 @@
-#!/usr/bin/env node
-export {};
-//# sourceMappingURL=sandbox-author.d.ts.map

package/dist/sandbox-author.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-author.d.ts","sourceRoot":"","sources":["../src/sandbox-author.ts"],"names":[],"mappings":""}

package/dist/sandbox-author.js

@@ -1,17 +0,0 @@
-#!/usr/bin/env node
-import { readdir } from "node:fs/promises";
-import { join } from "node:path";
-import { compileHarborTask } from "./harbor-task.js";
-const [tasksRoot, repositoryDirectory, outputDirectory] = process.argv.slice(2);
-if (!tasksRoot || !repositoryDirectory || !outputDirectory) {
-    throw new Error("usage: sandbox-author TASKS_ROOT REPOSITORY OUTPUT");
-}
-const entries = await readdir(tasksRoot, { withFileTypes: true });
-const taskDirectories = entries
-    .filter((entry) => entry.isDirectory())
-    .map((entry) => join(tasksRoot, entry.name));
-if (taskDirectories.length !== 1) {
-    throw new Error(`authoring must produce exactly one task; found ${taskDirectories.length}`);
-}
-await compileHarborTask(taskDirectories[0], repositoryDirectory, outputDirectory);
-//# sourceMappingURL=sandbox-author.js.map

package/dist/sandbox-author.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-author.js","sourceRoot":"","sources":["../src/sandbox-author.ts"],"names":[],"mappings":";AAEA,OAAO,EAAE,OAAO,EAAE,MAAM,kBAAkB,CAAC;AAC3C,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,iBAAiB,EAAE,MAAM,kBAAkB,CAAC;AAErD,MAAM,CAAC,SAAS,EAAE,mBAAmB,EAAE,eAAe,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;AAChF,IAAI,CAAC,SAAS,IAAI,CAAC,mBAAmB,IAAI,CAAC,eAAe,EAAE,CAAC;IAC3D,MAAM,IAAI,KAAK,CAAC,oDAAoD,CAAC,CAAC;AACxE,CAAC;AAED,MAAM,OAAO,GAAG,MAAM,OAAO,CAAC,SAAS,EAAE,EAAE,aAAa,EAAE,IAAI,EAAE,CAAC,CAAC;AAClE,MAAM,eAAe,GAAG,OAAO;KAC5B,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,WAAW,EAAE,CAAC;KACtC,GAAG,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,EAAE,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC;AAC/C,IAAI,eAAe,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;IACjC,MAAM,IAAI,KAAK,CAAC,kDAAkD,eAAe,CAAC,MAAM,EAAE,CAAC,CAAC;AAC9F,CAAC;AACD,MAAM,iBAAiB,CAAC,eAAe,CAAC,CAAC,CAAW,EAAE,mBAAmB,EAAE,eAAe,CAAC,CAAC"}

package/dist/sandbox-repair.bundle.js

@@ -1,315 +0,0 @@
-#!/usr/bin/env node
-
-// src/sandbox-repair.ts
-import { access, chmod, mkdir, readFile, writeFile } from "node:fs/promises";
-import { homedir, tmpdir } from "node:os";
-import { join } from "node:path";
-
-// src/hash.ts
-import { createHash } from "node:crypto";
-function sha256(value) {
-  return createHash("sha256").update(value).digest("hex");
-}
-
-// src/process.ts
-import { spawn } from "node:child_process";
-
-class InactivityTimeoutError extends Error {
-  scope;
-  timeoutMs;
-  constructor(scope, timeoutMs) {
-    super(`${scope} produced no output for ${timeoutMs}ms`);
-    this.scope = scope;
-    this.timeoutMs = timeoutMs;
-    this.name = "InactivityTimeoutError";
-  }
-}
-async function runCommand(command, args, options = {}) {
-  return await new Promise((resolve, reject) => {
-    const child = spawn(command, args, {
-      cwd: options.cwd,
-      env: options.env ?? process.env,
-      stdio: [options.input === undefined ? "ignore" : "pipe", "pipe", "pipe"]
-    });
-    const stdout = new RollingOutput;
-    const stderr = new RollingOutput;
-    if (!child.stdout || !child.stderr) {
-      reject(new Error(`failed to capture ${command} output`));
-      return;
-    }
-    let inactivityTimer;
-    let forceKillTimer;
-    let inactivityError;
-    const clearInactivityTimer = () => {
-      if (inactivityTimer) {
-        clearTimeout(inactivityTimer);
-        inactivityTimer = undefined;
-      }
-    };
-    const terminate = () => {
-      child.kill("SIGTERM");
-      forceKillTimer ??= setTimeout(() => child.kill("SIGKILL"), 5000);
-      forceKillTimer.unref();
-    };
-    const armInactivityTimer = () => {
-      clearInactivityTimer();
-      if (!options.inactivityTimeoutMs) {
-        return;
-      }
-      inactivityTimer = setTimeout(() => {
-        inactivityError = new InactivityTimeoutError(`${command} ${args.join(" ")}`, options.inactivityTimeoutMs ?? 0);
-        terminate();
-      }, options.inactivityTimeoutMs);
-      inactivityTimer.unref();
-    };
-    const capture = (stream, output, chunk) => {
-      output.push(chunk);
-      options.onOutput?.(stream, chunk);
-      armInactivityTimer();
-    };
-    child.stdout.on("data", (chunk) => capture("stdout", stdout, chunk));
-    child.stderr.on("data", (chunk) => capture("stderr", stderr, chunk));
-    child.on("error", reject);
-    const abort = () => terminate();
-    if (options.signal?.aborted) {
-      abort();
-    } else {
-      options.signal?.addEventListener("abort", abort, { once: true });
-    }
-    const timeout = options.timeoutMs ? setTimeout(() => terminate(), options.timeoutMs) : undefined;
-    timeout?.unref();
-    armInactivityTimer();
-    child.on("close", (exitCode, signal) => {
-      options.signal?.removeEventListener("abort", abort);
-      clearInactivityTimer();
-      if (forceKillTimer) {
-        clearTimeout(forceKillTimer);
-      }
-      if (timeout) {
-        clearTimeout(timeout);
-      }
-      const result = {
-        stdout: stdout.text(),
-        stderr: stderr.text(),
-        exitCode: exitCode ?? (signal ? 128 : 1)
-      };
-      if (inactivityError) {
-        reject(inactivityError);
-        return;
-      }
-      if (result.exitCode !== 0 && !options.allowFailure) {
-        const detail = result.stderr.trim() || result.stdout.trim();
-        reject(new Error(`${command} exited ${result.exitCode}${detail ? `: ${detail}` : ""}`));
-        return;
-      }
-      resolve(result);
-    });
-    if (child.stdin && options.input !== undefined) {
-      child.stdin.end(options.input);
-    }
-  });
-}
-
-class RollingOutput {
-  #limit = 8 * 1024 * 1024;
-  #chunks = [];
-  #size = 0;
-  #truncated = false;
-  push(chunk) {
-    if (chunk.byteLength >= this.#limit) {
-      this.#chunks.length = 0;
-      this.#chunks.push(chunk.subarray(chunk.byteLength - this.#limit));
-      this.#size = this.#limit;
-      this.#truncated = true;
-      return;
-    }
-    this.#chunks.push(chunk);
-    this.#size += chunk.byteLength;
-    while (this.#size > this.#limit) {
-      const first = this.#chunks.shift();
-      if (!first) {
-        break;
-      }
-      const overflow = this.#size - this.#limit;
-      if (first.byteLength > overflow) {
-        this.#chunks.unshift(first.subarray(overflow));
-        this.#size -= overflow;
-      } else {
-        this.#size -= first.byteLength;
-      }
-      this.#truncated = true;
-    }
-  }
-  text() {
-    const value = Buffer.concat(this.#chunks, this.#size).toString("utf8");
-    return this.#truncated ? `[selfbench: earlier output truncated]
-${value}` : value;
-  }
-}
-
-// src/repair.ts
-function patchPaths(patch) {
-  const paths = new Set;
-  for (const line of patch.split(`
-`)) {
-    const match = /^diff --git a\/(.+) b\/(.+)$/.exec(line);
-    if (match?.[2]) {
-      paths.add(match[2]);
-    }
-  }
-  return [...paths].sort();
-}
-function assertRepairPaths(originalTestPatch, changedPaths) {
-  const allowed = new Set(patchPaths(originalTestPatch));
-  if (allowed.size === 0) {
-    throw new Error("original held-out test patch changes no files");
-  }
-  const outside = changedPaths.filter((path) => !allowed.has(path));
-  if (outside.length > 0) {
-    throw new Error(`repair changed files outside the held-out tests: ${outside.join(", ")}`);
-  }
-}
-function repairPrompt(input) {
-  return `Repair the held-out tests for SelfBench task ${input.taskId}.
-
-The repository is the exact base snapshot with the current held-out test patch already applied. You may edit only these existing test files:
-
-${input.allowedPaths.map((path) => `- ${path}`).join(`
-`)}
-
-The authentic engineer request is:
-
-<authentic_request>
-${input.authenticRequest.trim()}
-</authentic_request>
-
-The independent coupling report is:
-
-<coupling_report>
-${input.couplingReport.trim()}
-</coupling_report>
-
-Rewrite the tests so they verify every material requested behavior through stable public boundaries without requiring the gold patch's exact internal names, helper structure, error prose, mock input shape, response presentation, or incidental implementation choices. Preserve meaningful negative, authorization, compatibility, and regression coverage. Do not edit application code, the request, or the gold patch. Do not delete assertions merely to silence the report. A coherent alternative implementation must be able to pass, while the unchanged base must still fail the fail-to-pass tests and the gold implementation must still pass.
-
-The final working tree must retain a non-empty test patch relative to HEAD and every originally added fail-to-pass test must still have an equivalent behavioral test. Do not reset, revert, delete, or commit the test patch. If an exact public response field is the only stable seam, prefer asserting the underlying externally visible behavior through existing endpoints, persistence, or follow-up actions. If no rigorous uncoupled test is possible within the allowed files, leave the current tests in place and explain the blocker in your final message instead of removing coverage.
-
-Inspect /work/task/solution/gold.patch only to understand the intended behavior and available seams, never to copy its private structure into assertions. Inspect /work/task/tests/test.sh for the verifier command. Run focused tests when feasible. Finish only after the working tree contains the repaired test files and no other changes.`;
-}
-
-// src/sandbox-repair.ts
-var [archivePath, reviewPath, outputArchive, outputReport] = process.argv.slice(2);
-if (!archivePath || !reviewPath || !outputArchive || !outputReport) {
-  throw new Error("usage: sandbox-repair TASK.tar.gz REVIEW.json OUTPUT.tar.gz OUTPUT-REPORT.json");
-}
-var extractedDirectory = "/work/task";
-var repositoryDirectory = "/work/repo";
-await Promise.all([mkdir(extractedDirectory, { recursive: true }), mkdir(repositoryDirectory)]);
-await runCommand("tar", ["-xzf", archivePath, "-C", extractedDirectory]);
-var taskDirectory = await access(join(extractedDirectory, "instruction.md")).then(() => extractedDirectory, () => join(extractedDirectory, "harbor-task"));
-var [instruction, originalPatch, review] = await Promise.all([
-  readFile(join(taskDirectory, "instruction.md"), "utf8"),
-  readFile(join(taskDirectory, "tests/test.patch"), "utf8"),
-  readFile(reviewPath, "utf8")
-]);
-var manifestPath = join(taskDirectory, ".selfbench-manifest.json");
-var manifest = JSON.parse(await readFile(manifestPath, "utf8"));
-var taskId = typeof manifest.taskId === "string" ? manifest.taskId : "unknown-task";
-var allowedPaths = patchPaths(originalPatch);
-await runCommand("tar", [
-  "-xzf",
-  join(taskDirectory, "tests/repo.tar.gz"),
-  "-C",
-  repositoryDirectory
-]);
-await runCommand("git", ["-C", repositoryDirectory, "init", "-q"]);
-await runCommand("git", ["-C", repositoryDirectory, "config", "user.name", "SelfBench"]);
-await runCommand("git", ["-C", repositoryDirectory, "config", "user.email", "selfbench@local"]);
-await runCommand("git", ["-C", repositoryDirectory, "add", "-A"]);
-await runCommand("git", ["-C", repositoryDirectory, "commit", "-qm", "base"]);
-await runCommand("git", [
-  "-C",
-  repositoryDirectory,
-  "apply",
-  join(taskDirectory, "tests/test.patch")
-]);
-await runCommand("git", ["-C", repositoryDirectory, "add", "-N", "--all"]);
-var codexAuth = process.env.SELFBENCH_CODEX_AUTH_JSON;
-var apiKey = process.env.OPENAI_API_KEY;
-if (!apiKey && !codexAuth) {
-  fail("OPENAI_API_KEY or SELFBENCH_CODEX_AUTH_JSON is required");
-}
-var codexHome = join(homedir(), ".codex");
-var authPath = join(codexHome, "auth.json");
-await mkdir(codexHome, { recursive: true });
-await writeFile(authPath, codexAuth ?? `${JSON.stringify({ OPENAI_API_KEY: apiKey })}
-`);
-await chmod(authPath, 384);
-var promptPath = join(tmpdir(), `selfbench-repair-${taskId}.md`);
-await writeFile(promptPath, repairPrompt({
-  taskId,
-  authenticRequest: instruction,
-  couplingReport: review,
-  allowedPaths
-}));
-var codex = await runCommand("codex", [
-  "exec",
-  "--model",
-  process.env.SELFBENCH_REPAIR_MODEL ?? "gpt-5.6-sol",
-  "--dangerously-bypass-approvals-and-sandbox",
-  "--ephemeral",
-  "--ignore-user-config",
-  "--json",
-  "-C",
-  repositoryDirectory,
-  "-"
-], {
-  allowFailure: true,
-  timeoutMs: 90 * 60 * 1000,
-  env: process.env,
-  input: await readFile(promptPath, "utf8"),
-  onOutput: (stream, chunk) => {
-    (stream === "stdout" ? process.stdout : process.stderr).write(chunk);
-  }
-});
-if (codex.exitCode !== 0) {
-  throw new Error(`Codex repair exited ${codex.exitCode}: ${codex.stderr.slice(-2000)}`);
-}
-var [tracked, untracked] = await Promise.all([
-  runCommand("git", ["-C", repositoryDirectory, "diff", "--name-only", "HEAD"]),
-  runCommand("git", ["-C", repositoryDirectory, "ls-files", "--others", "--exclude-standard"])
-]);
-var changedPaths = [...tracked.stdout.split(`
-`), ...untracked.stdout.split(`
-`)].filter(Boolean).sort();
-assertRepairPaths(originalPatch, changedPaths);
-var repaired = await runCommand("git", ["-C", repositoryDirectory, "diff", "--binary", "HEAD"]);
-if (!repaired.stdout.startsWith("diff --git ")) {
-  throw new Error(`repair produced no held-out test patch; status=${JSON.stringify(changedPaths)}; Codex tail=${codex.stdout.slice(-4000)}`);
-}
-if (repaired.stdout === originalPatch) {
-  throw new Error("repair left the held-out test patch unchanged");
-}
-await writeFile(join(taskDirectory, "tests/test.patch"), repaired.stdout);
-await writeFile(manifestPath, `${JSON.stringify({
-  ...manifest,
-  testPatchSha256: sha256(repaired.stdout),
-  repair: {
-    model: process.env.SELFBENCH_REPAIR_MODEL ?? "gpt-5.6-sol",
-    originalTestPatchSha256: sha256(originalPatch)
-  }
-}, null, 2)}
-`);
-await runCommand("tar", ["-czf", outputArchive, "-C", extractedDirectory, "."]);
-await writeFile(outputReport, `${JSON.stringify({
-  schemaVersion: 1,
-  taskId,
-  model: process.env.SELFBENCH_REPAIR_MODEL ?? "gpt-5.6-sol",
-  changedPaths,
-  originalTestPatchSha256: sha256(originalPatch),
-  repairedTestPatchSha256: sha256(repaired.stdout),
-  codexOutputTail: codex.stdout.slice(-4000)
-}, null, 2)}
-`);
-function fail(message) {
-  throw new Error(message);
-}

package/dist/sandbox-repair.d.ts

@@ -1,3 +0,0 @@
-#!/usr/bin/env node
-export {};
-//# sourceMappingURL=sandbox-repair.d.ts.map

package/dist/sandbox-repair.d.ts.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-repair.d.ts","sourceRoot":"","sources":["../src/sandbox-repair.ts"],"names":[],"mappings":""}

package/dist/sandbox-repair.js

@@ -1,121 +0,0 @@
-#!/usr/bin/env node
-import { access, chmod, mkdir, readFile, writeFile } from "node:fs/promises";
-import { homedir, tmpdir } from "node:os";
-import { join } from "node:path";
-import { sha256 } from "./hash.js";
-import { runCommand } from "./process.js";
-import { assertRepairPaths, patchPaths, repairPrompt } from "./repair.js";
-const [archivePath, reviewPath, outputArchive, outputReport] = process.argv.slice(2);
-if (!archivePath || !reviewPath || !outputArchive || !outputReport) {
-    throw new Error("usage: sandbox-repair TASK.tar.gz REVIEW.json OUTPUT.tar.gz OUTPUT-REPORT.json");
-}
-const extractedDirectory = "/work/task";
-const repositoryDirectory = "/work/repo";
-await Promise.all([mkdir(extractedDirectory, { recursive: true }), mkdir(repositoryDirectory)]);
-await runCommand("tar", ["-xzf", archivePath, "-C", extractedDirectory]);
-const taskDirectory = await access(join(extractedDirectory, "instruction.md")).then(() => extractedDirectory, () => join(extractedDirectory, "harbor-task"));
-const [instruction, originalPatch, review] = await Promise.all([
-    readFile(join(taskDirectory, "instruction.md"), "utf8"),
-    readFile(join(taskDirectory, "tests/test.patch"), "utf8"),
-    readFile(reviewPath, "utf8"),
-]);
-const manifestPath = join(taskDirectory, ".selfbench-manifest.json");
-const manifest = JSON.parse(await readFile(manifestPath, "utf8"));
-const taskId = typeof manifest.taskId === "string" ? manifest.taskId : "unknown-task";
-const allowedPaths = patchPaths(originalPatch);
-await runCommand("tar", [
-    "-xzf",
-    join(taskDirectory, "tests/repo.tar.gz"),
-    "-C",
-    repositoryDirectory,
-]);
-await runCommand("git", ["-C", repositoryDirectory, "init", "-q"]);
-await runCommand("git", ["-C", repositoryDirectory, "config", "user.name", "SelfBench"]);
-await runCommand("git", ["-C", repositoryDirectory, "config", "user.email", "selfbench@local"]);
-await runCommand("git", ["-C", repositoryDirectory, "add", "-A"]);
-await runCommand("git", ["-C", repositoryDirectory, "commit", "-qm", "base"]);
-await runCommand("git", [
-    "-C",
-    repositoryDirectory,
-    "apply",
-    join(taskDirectory, "tests/test.patch"),
-]);
-await runCommand("git", ["-C", repositoryDirectory, "add", "-N", "--all"]);
-const codexAuth = process.env.SELFBENCH_CODEX_AUTH_JSON;
-const apiKey = process.env.OPENAI_API_KEY;
-if (!apiKey && !codexAuth) {
-    fail("OPENAI_API_KEY or SELFBENCH_CODEX_AUTH_JSON is required");
-}
-const codexHome = join(homedir(), ".codex");
-const authPath = join(codexHome, "auth.json");
-await mkdir(codexHome, { recursive: true });
-await writeFile(authPath, codexAuth ?? `${JSON.stringify({ OPENAI_API_KEY: apiKey })}\n`);
-await chmod(authPath, 0o600);
-const promptPath = join(tmpdir(), `selfbench-repair-${taskId}.md`);
-await writeFile(promptPath, repairPrompt({
-    taskId,
-    authenticRequest: instruction,
-    couplingReport: review,
-    allowedPaths,
-}));
-const codex = await runCommand("codex", [
-    "exec",
-    "--model",
-    process.env.SELFBENCH_REPAIR_MODEL ?? "gpt-5.6-sol",
-    "--dangerously-bypass-approvals-and-sandbox",
-    "--ephemeral",
-    "--ignore-user-config",
-    "--json",
-    "-C",
-    repositoryDirectory,
-    "-",
-], {
-    allowFailure: true,
-    timeoutMs: 90 * 60 * 1000,
-    env: process.env,
-    input: await readFile(promptPath, "utf8"),
-    onOutput: (stream, chunk) => {
-        (stream === "stdout" ? process.stdout : process.stderr).write(chunk);
-    },
-});
-if (codex.exitCode !== 0) {
-    throw new Error(`Codex repair exited ${codex.exitCode}: ${codex.stderr.slice(-2_000)}`);
-}
-const [tracked, untracked] = await Promise.all([
-    runCommand("git", ["-C", repositoryDirectory, "diff", "--name-only", "HEAD"]),
-    runCommand("git", ["-C", repositoryDirectory, "ls-files", "--others", "--exclude-standard"]),
-]);
-const changedPaths = [...tracked.stdout.split("\n"), ...untracked.stdout.split("\n")]
-    .filter(Boolean)
-    .sort();
-assertRepairPaths(originalPatch, changedPaths);
-const repaired = await runCommand("git", ["-C", repositoryDirectory, "diff", "--binary", "HEAD"]);
-if (!repaired.stdout.startsWith("diff --git ")) {
-    throw new Error(`repair produced no held-out test patch; status=${JSON.stringify(changedPaths)}; Codex tail=${codex.stdout.slice(-4_000)}`);
-}
-if (repaired.stdout === originalPatch) {
-    throw new Error("repair left the held-out test patch unchanged");
-}
-await writeFile(join(taskDirectory, "tests/test.patch"), repaired.stdout);
-await writeFile(manifestPath, `${JSON.stringify({
-    ...manifest,
-    testPatchSha256: sha256(repaired.stdout),
-    repair: {
-        model: process.env.SELFBENCH_REPAIR_MODEL ?? "gpt-5.6-sol",
-        originalTestPatchSha256: sha256(originalPatch),
-    },
-}, null, 2)}\n`);
-await runCommand("tar", ["-czf", outputArchive, "-C", extractedDirectory, "."]);
-await writeFile(outputReport, `${JSON.stringify({
-    schemaVersion: 1,
-    taskId,
-    model: process.env.SELFBENCH_REPAIR_MODEL ?? "gpt-5.6-sol",
-    changedPaths,
-    originalTestPatchSha256: sha256(originalPatch),
-    repairedTestPatchSha256: sha256(repaired.stdout),
-    codexOutputTail: codex.stdout.slice(-4_000),
-}, null, 2)}\n`);
-function fail(message) {
-    throw new Error(message);
-}
-//# sourceMappingURL=sandbox-repair.js.map

package/dist/sandbox-repair.js.map

@@ -1 +0,0 @@
-{"version":3,"file":"sandbox-repair.js","sourceRoot":"","sources":["../src/sandbox-repair.ts"],"names":[],"mappings":";AAEA,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,KAAK,EAAE,QAAQ,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAC7E,OAAO,EAAE,OAAO,EAAE,MAAM,EAAE,MAAM,SAAS,CAAC;AAC1C,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,MAAM,EAAE,MAAM,WAAW,CAAC;AACnC,OAAO,EAAE,UAAU,EAAE,MAAM,cAAc,CAAC;AAC1C,OAAO,EAAE,iBAAiB,EAAE,UAAU,EAAE,YAAY,EAAE,MAAM,aAAa,CAAC;AAE1E,MAAM,CAAC,WAAW,EAAE,UAAU,EAAE,aAAa,EAAE,YAAY,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC;AACrF,IAAI,CAAC,WAAW,IAAI,CAAC,UAAU,IAAI,CAAC,aAAa,IAAI,CAAC,YAAY,EAAE,CAAC;IACnE,MAAM,IAAI,KAAK,CAAC,gFAAgF,CAAC,CAAC;AACpG,CAAC;AAED,MAAM,kBAAkB,GAAG,YAAY,CAAC;AACxC,MAAM,mBAAmB,GAAG,YAAY,CAAC;AACzC,MAAM,OAAO,CAAC,GAAG,CAAC,CAAC,KAAK,CAAC,kBAAkB,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,EAAE,KAAK,CAAC,mBAAmB,CAAC,CAAC,CAAC,CAAC;AAChG,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,MAAM,EAAE,WAAW,EAAE,IAAI,EAAE,kBAAkB,CAAC,CAAC,CAAC;AACzE,MAAM,aAAa,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,kBAAkB,EAAE,gBAAgB,CAAC,CAAC,CAAC,IAAI,CACjF,GAAG,EAAE,CAAC,kBAAkB,EACxB,GAAG,EAAE,CAAC,IAAI,CAAC,kBAAkB,EAAE,aAAa,CAAC,CAC9C,CAAC;AAEF,MAAM,CAAC,WAAW,EAAE,aAAa,EAAE,MAAM,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;IAC7D,QAAQ,CAAC,IAAI,CAAC,aAAa,EAAE,gBAAgB,CAAC,EAAE,MAAM,CAAC;IACvD,QAAQ,CAAC,IAAI,CAAC,aAAa,EAAE,kBAAkB,CAAC,EAAE,MAAM,CAAC;IACzD,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;CAC7B,CAAC,CAAC;AACH,MAAM,YAAY,GAAG,IAAI,CAAC,aAAa,EAAE,0BAA0B,CAAC,CAAC;AACrE,MAAM,QAAQ,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,QAAQ,CAAC,YAAY,EAAE,MAAM,CAAC,CAA4B,CAAC;AAC7F,MAAM,MAAM,GAAG,OAAO,QAAQ,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,QAAQ,CAAC,MAAM,CAAC,CAAC,CAAC,cAAc,CAAC;AACtF,MAAM,YAAY,GAAG,UAAU,CAAC,aAAa,CAAC,CAAC;AAE/C,MAAM,UAAU,CAAC,KAAK,EAAE;IACtB,MAAM;IACN,IAAI,CAAC,aAAa,EAAE,mBAAmB,CAAC;IACxC,IAAI;IACJ,mBAAmB;CACpB,CAAC,CAAC;AACH,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,MAAM,EAAE,IAAI,CAAC,CAAC,CAAC;AACnE,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,QAAQ,EAAE,WAAW,EAAE,WAAW,CAAC,CAAC,CAAC;AACzF,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,QAAQ,EAAE,YAAY,EAAE,iBAAiB,CAAC,CAAC,CAAC;AAChG,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,KAAK,EAAE,IAAI,CAAC,CAAC,CAAC;AAClE,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,QAAQ,EAAE,KAAK,EAAE,MAAM,CAAC,CAAC,CAAC;AAC9E,MAAM,UAAU,CAAC,KAAK,EAAE;IACtB,IAAI;IACJ,mBAAmB;IACnB,OAAO;IACP,IAAI,CAAC,aAAa,EAAE,kBAAkB,CAAC;CACxC,CAAC,CAAC;AACH,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,KAAK,EAAE,IAAI,EAAE,OAAO,CAAC,CAAC,CAAC;AAE3E,MAAM,SAAS,GAAG,OAAO,CAAC,GAAG,CAAC,yBAAyB,CAAC;AACxD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,cAAc,CAAC;AAC1C,IAAI,CAAC,MAAM,IAAI,CAAC,SAAS,EAAE,CAAC;IAC1B,IAAI,CAAC,yDAAyD,CAAC,CAAC;AAClE,CAAC;AACD,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,EAAE,EAAE,QAAQ,CAAC,CAAC;AAC5C,MAAM,QAAQ,GAAG,IAAI,CAAC,SAAS,EAAE,WAAW,CAAC,CAAC;AAC9C,MAAM,KAAK,CAAC,SAAS,EAAE,EAAE,SAAS,EAAE,IAAI,EAAE,CAAC,CAAC;AAC5C,MAAM,SAAS,CAAC,QAAQ,EAAE,SAAS,IAAI,GAAG,IAAI,CAAC,SAAS,CAAC,EAAE,cAAc,EAAE,MAAM,EAAE,CAAC,IAAI,CAAC,CAAC;AAC1F,MAAM,KAAK,CAAC,QAAQ,EAAE,KAAK,CAAC,CAAC;AAC7B,MAAM,UAAU,GAAG,IAAI,CAAC,MAAM,EAAE,EAAE,oBAAoB,MAAM,KAAK,CAAC,CAAC;AACnE,MAAM,SAAS,CACb,UAAU,EACV,YAAY,CAAC;IACX,MAAM;IACN,gBAAgB,EAAE,WAAW;IAC7B,cAAc,EAAE,MAAM;IACtB,YAAY;CACb,CAAC,CACH,CAAC;AAEF,MAAM,KAAK,GAAG,MAAM,UAAU,CAC5B,OAAO,EACP;IACE,MAAM;IACN,SAAS;IACT,OAAO,CAAC,GAAG,CAAC,sBAAsB,IAAI,aAAa;IACnD,4CAA4C;IAC5C,aAAa;IACb,sBAAsB;IACtB,QAAQ;IACR,IAAI;IACJ,mBAAmB;IACnB,GAAG;CACJ,EACD;IACE,YAAY,EAAE,IAAI;IAClB,SAAS,EAAE,EAAE,GAAG,EAAE,GAAG,IAAI;IACzB,GAAG,EAAE,OAAO,CAAC,GAAG;IAChB,KAAK,EAAE,MAAM,QAAQ,CAAC,UAAU,EAAE,MAAM,CAAC;IACzC,QAAQ,EAAE,CAAC,MAAM,EAAE,KAAK,EAAE,EAAE;QAC1B,CAAC,MAAM,KAAK,QAAQ,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,KAAK,CAAC,KAAK,CAAC,CAAC;IACvE,CAAC;CACF,CACF,CAAC;AACF,IAAI,KAAK,CAAC,QAAQ,KAAK,CAAC,EAAE,CAAC;IACzB,MAAM,IAAI,KAAK,CAAC,uBAAuB,KAAK,CAAC,QAAQ,KAAK,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,CAAC,CAAC;AAC1F,CAAC;AAED,MAAM,CAAC,OAAO,EAAE,SAAS,CAAC,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC;IAC7C,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,MAAM,EAAE,aAAa,EAAE,MAAM,CAAC,CAAC;IAC7E,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,UAAU,EAAE,UAAU,EAAE,oBAAoB,CAAC,CAAC;CAC7F,CAAC,CAAC;AACH,MAAM,YAAY,GAAG,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,GAAG,SAAS,CAAC,MAAM,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC;KAClF,MAAM,CAAC,OAAO,CAAC;KACf,IAAI,EAAE,CAAC;AACV,iBAAiB,CAAC,aAAa,EAAE,YAAY,CAAC,CAAC;AAC/C,MAAM,QAAQ,GAAG,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,IAAI,EAAE,mBAAmB,EAAE,MAAM,EAAE,UAAU,EAAE,MAAM,CAAC,CAAC,CAAC;AAClG,IAAI,CAAC,QAAQ,CAAC,MAAM,CAAC,UAAU,CAAC,aAAa,CAAC,EAAE,CAAC;IAC/C,MAAM,IAAI,KAAK,CACb,kDAAkD,IAAI,CAAC,SAAS,CAAC,YAAY,CAAC,gBAAgB,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,EAAE,CAC3H,CAAC;AACJ,CAAC;AACD,IAAI,QAAQ,CAAC,MAAM,KAAK,aAAa,EAAE,CAAC;IACtC,MAAM,IAAI,KAAK,CAAC,+CAA+C,CAAC,CAAC;AACnE,CAAC;AAED,MAAM,SAAS,CAAC,IAAI,CAAC,aAAa,EAAE,kBAAkB,CAAC,EAAE,QAAQ,CAAC,MAAM,CAAC,CAAC;AAC1E,MAAM,SAAS,CACb,YAAY,EACZ,GAAG,IAAI,CAAC,SAAS,CACf;IACE,GAAG,QAAQ;IACX,eAAe,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;IACxC,MAAM,EAAE;QACN,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,sBAAsB,IAAI,aAAa;QAC1D,uBAAuB,EAAE,MAAM,CAAC,aAAa,CAAC;KAC/C;CACF,EACD,IAAI,EACJ,CAAC,CACF,IAAI,CACN,CAAC;AACF,MAAM,UAAU,CAAC,KAAK,EAAE,CAAC,MAAM,EAAE,aAAa,EAAE,IAAI,EAAE,kBAAkB,EAAE,GAAG,CAAC,CAAC,CAAC;AAChF,MAAM,SAAS,CACb,YAAY,EACZ,GAAG,IAAI,CAAC,SAAS,CACf;IACE,aAAa,EAAE,CAAC;IAChB,MAAM;IACN,KAAK,EAAE,OAAO,CAAC,GAAG,CAAC,sBAAsB,IAAI,aAAa;IAC1D,YAAY;IACZ,uBAAuB,EAAE,MAAM,CAAC,aAAa,CAAC;IAC9C,uBAAuB,EAAE,MAAM,CAAC,QAAQ,CAAC,MAAM,CAAC;IAChD,eAAe,EAAE,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC;CAC5C,EACD,IAAI,EACJ,CAAC,CACF,IAAI,CACN,CAAC;AAEF,SAAS,IAAI,CAAC,OAAe;IAC3B,MAAM,IAAI,KAAK,CAAC,OAAO,CAAC,CAAC;AAC3B,CAAC"}