agex 0.1.0

package/dist/chunk-KVAOIDJ4.js

@@ -0,0 +1,419 @@
+import {
+  buildEnv,
+  getAgentConfig,
+  getRequiredApiKey,
+  runClaude,
+  runCodex,
+  runCursor
+} from "./chunk-3FGK7LUI.js";
+
+// ../run/dist/run.js
+function buildDebugOutput(output) {
+  const lines = [
+    "agex-run",
+    `agent=${output.agent}`,
+    `model=${output.model ?? ""}`,
+    `status=${output.status}`,
+    `duration_ms=${output.durationMs}`,
+    `exit_code=${output.exitCode}`,
+    "---stdout---",
+    output.stdout.trimEnd(),
+    "---stderr---",
+    output.stderr.trimEnd()
+  ];
+  return lines.join("\n");
+}
+function normalizeOutput(output, formattedText) {
+  return {
+    json: output,
+    text: buildDebugOutput(output),
+    formattedText
+  };
+}
+async function runAgent(request) {
+  const config = getAgentConfig(request.agent);
+  const apiKeyEnv = getRequiredApiKey(request.agent);
+  const apiKey = request.env?.[apiKeyEnv] ?? process.env[apiKeyEnv];
+  if (!apiKey && request.requireApiKey) {
+    throw new Error(`Missing required API key: ${apiKeyEnv}`);
+  }
+  if (request.install) {
+    const { ensureAgentInstalled: ensureAgentInstalled2 } = await import("./agents-2OHWEGTE.js");
+    await ensureAgentInstalled2(request.agent, void 0, request.installCommand);
+  }
+  const args = config.args({
+    prompt: request.prompt,
+    model: request.model,
+    approveMcps: request.approveMcps,
+    browser: request.browser,
+    streamOutput: request.streamOutput ?? true,
+    approvalPolicy: request.approvalPolicy,
+    mcpConfigPath: request.mcpConfigPath,
+    workspacePath: request.workspacePath
+  });
+  if ((request.streamOutput ?? true) && !request.quietOutput) {
+    process.stdout.write("[agex-run] starting agent...\n");
+  }
+  const env = buildEnv({ apiKeyEnv, apiKey, request });
+  let result;
+  if (request.agent === "cursor") {
+    result = await runCursor(request, args, env);
+  } else if (request.agent === "claude") {
+    result = await runClaude(request, args, env);
+  } else {
+    result = await runCodex(request, args, env);
+  }
+  return normalizeOutput(result.output, result.formattedText);
+}
+
+// ../proov/dist/proov.js
+import * as fs2 from "fs";
+import * as os2 from "os";
+import * as path2 from "path";
+
+// ../browse/dist/index.js
+import * as fs from "fs";
+import * as os from "os";
+import * as path from "path";
+import { fileURLToPath } from "url";
+function getCursorScriptPath() {
+  const __dirname = path.dirname(fileURLToPath(import.meta.url));
+  return path.resolve(__dirname, "..", "assets", "cursor.js");
+}
+function buildMcpArgs(config) {
+  const args = ["@playwright/mcp@latest"];
+  const outputDir = path.resolve(config.output.dir);
+  if (config.storageStatePath && fs.existsSync(config.storageStatePath)) {
+    args.push("--storage-state", path.resolve(config.storageStatePath));
+  }
+  if (config.initScriptPath && fs.existsSync(config.initScriptPath)) {
+    args.push("--init-script", path.resolve(config.initScriptPath));
+  }
+  const showCursor = config.output.showCursor ?? config.output.video ?? true;
+  if (showCursor) {
+    const cursorScript = getCursorScriptPath();
+    if (fs.existsSync(cursorScript)) {
+      args.push("--init-script", cursorScript);
+    }
+  }
+  if (config.output.video) {
+    const viewport = config.output.viewport ?? { width: 1920, height: 1080 };
+    args.push(`--save-video=${viewport.width}x${viewport.height}`);
+  }
+  args.push("--output-dir", outputDir);
+  if (config.output.viewport) {
+    args.push("--viewport-size", `${config.output.viewport.width}x${config.output.viewport.height}`);
+  }
+  return args;
+}
+function createMcpJsonFile(config, agent = "cursor") {
+  const tempDir = fs.mkdtempSync(path.join(os.tmpdir(), "agex-"));
+  let mcpJsonPath;
+  if (agent === "cursor") {
+    const cursorDir = path.join(tempDir, ".cursor");
+    fs.mkdirSync(cursorDir, { recursive: true });
+    mcpJsonPath = path.join(cursorDir, "mcp.json");
+  } else {
+    mcpJsonPath = path.join(tempDir, "mcp.json");
+  }
+  const mcpJson = {
+    mcpServers: {
+      playwright: {
+        command: "npx",
+        args: buildMcpArgs(config)
+      }
+    }
+  };
+  fs.writeFileSync(mcpJsonPath, JSON.stringify(mcpJson, null, 2));
+  return {
+    path: mcpJsonPath,
+    workspacePath: tempDir,
+    cleanup: () => {
+      try {
+        fs.rmSync(tempDir, { recursive: true, force: true });
+      } catch {
+        return;
+      }
+    }
+  };
+}
+
+// ../proov/dist/prompt.js
+function buildProovPrompt(assertion, options) {
+  const recordingInfo = options.video !== false ? "Video recording is enabled." : "Screenshots only.";
+  const screenshotInfo = options.screenshots !== false ? "Take screenshots at key moments using 3-digit zero-padded numbering (e.g., 001-login-button.png). Only capture when it adds new evidence." : "";
+  const parts = [
+    "You have the Playwright MCP for browser automation.",
+    recordingInfo,
+    "",
+    `YOUR TASK: ${assertion}`
+  ];
+  if (options.url) {
+    parts.push(`Navigate to: ${options.url}`);
+  }
+  const screenshotInstruction = screenshotInfo ? `2. ${screenshotInfo}` : "2. Document key observations";
+  parts.push(`
+INSTRUCTIONS:
+1. Use the browser tools to navigate and interact
+${screenshotInstruction}
+3. Add brief waits between actions for video clarity
+4. Narrate what you observe
+
+FINAL VERDICT (required):
+End your response with exactly one of these lines (copy-paste exactly):
+  [PROOV_VERDICT:PASS] - <brief explanation>
+  [PROOV_VERDICT:FAIL] - <brief explanation>
+  [PROOV_VERDICT:SKIP] - <reason why it cannot be proven>`);
+  return parts.join("\n");
+}
+
+// ../proov/dist/verdict.js
+var VERDICT_REGEX = /\[PROOV_VERDICT:(PASS|FAIL|SKIP)\]\s*-?\s*([^"\n\r}]*)/i;
+function parseVerdict(output) {
+  const matches = [...output.matchAll(new RegExp(VERDICT_REGEX, "gi"))];
+  if (matches.length === 0) {
+    return { verdict: "fail", reason: "No verdict found in output" };
+  }
+  const lastMatch = matches[matches.length - 1];
+  return {
+    verdict: lastMatch[1].toLowerCase(),
+    reason: lastMatch[2]?.trim() || ""
+  };
+}
+
+// ../proov/dist/proov.js
+var DEFAULT_OPTIONS = {
+  agent: "cursor",
+  outputDir: "./proov-results",
+  video: true,
+  screenshots: true,
+  viewport: { width: 1920, height: 1080 },
+  headless: true
+};
+function createOutputDir(baseDir, assertion) {
+  const timestamp = (/* @__PURE__ */ new Date()).toISOString().replace(/[:.]/g, "-").slice(0, 19);
+  const slug = assertion.toLowerCase().replace(/[^a-z0-9]+/g, "-").replace(/^-|-$/g, "").slice(0, 50);
+  const dir = path2.resolve(baseDir, `${timestamp}_${slug}`);
+  fs2.mkdirSync(dir, { recursive: true });
+  return dir;
+}
+function copyPlaywrightArtifacts(outputDir) {
+  const tmpDir = os2.tmpdir();
+  const playwrightOutputBase = path2.join(tmpDir, "playwright-mcp-output");
+  if (!fs2.existsSync(playwrightOutputBase)) {
+    return;
+  }
+  const subdirs = fs2.readdirSync(playwrightOutputBase).map((name) => ({ name, path: path2.join(playwrightOutputBase, name) })).filter((item) => fs2.statSync(item.path).isDirectory()).sort((a, b) => {
+    const statA = fs2.statSync(a.path);
+    const statB = fs2.statSync(b.path);
+    return statB.mtime.getTime() - statA.mtime.getTime();
+  });
+  if (subdirs.length === 0) {
+    return;
+  }
+  const latestDir = subdirs[0].path;
+  copyFilesRecursively(latestDir, outputDir);
+}
+function copyFilesRecursively(srcDir, destDir) {
+  if (!fs2.existsSync(srcDir)) {
+    return;
+  }
+  const entries = fs2.readdirSync(srcDir, { withFileTypes: true });
+  for (const entry of entries) {
+    const srcPath = path2.join(srcDir, entry.name);
+    if (entry.isDirectory()) {
+      copyFilesRecursively(srcPath, destDir);
+    } else if (entry.isFile() && /\.(png|jpg|jpeg|webm|mp4)$/i.test(entry.name)) {
+      const destPath = path2.join(destDir, entry.name);
+      fs2.copyFileSync(srcPath, destPath);
+    }
+  }
+}
+async function proov(assertion, options = {}) {
+  const opts = { ...DEFAULT_OPTIONS, ...options };
+  const agent = opts.agent;
+  const outputDir = createOutputDir(opts.outputDir, assertion);
+  const browserConfig = {
+    output: {
+      dir: outputDir,
+      video: opts.video,
+      screenshots: opts.screenshots,
+      viewport: opts.viewport,
+      showCursor: true
+    }
+  };
+  const mcpFile = createMcpJsonFile(browserConfig, agent);
+  const prompt = buildProovPrompt(assertion, opts);
+  const startTime = Date.now();
+  try {
+    const result = await runAgent({
+      agent,
+      prompt,
+      model: opts.model,
+      approveMcps: true,
+      browser: true,
+      streamOutput: true,
+      approvalPolicy: "never",
+      mcpConfigPath: mcpFile.path,
+      workspacePath: mcpFile.workspacePath,
+      timeoutMs: opts.timeout
+    });
+    const durationMs = Date.now() - startTime;
+    const { verdict, reason } = parseVerdict(result.json.stdout);
+    copyPlaywrightArtifacts(outputDir);
+    return {
+      verdict,
+      reason,
+      agent,
+      durationMs,
+      outputDir,
+      exitCode: result.json.exitCode,
+      stdout: result.json.stdout,
+      stderr: result.json.stderr
+    };
+  } finally {
+    mcpFile.cleanup();
+  }
+}
+
+// ../review/dist/git.js
+import { execa } from "execa";
+async function buildGitContext(baseRef, includeWorktree, cwd) {
+  const exec = (command) => execa("bash", ["-lc", command], { cwd, reject: false }).then((result) => result.stdout.trim());
+  const currentBranch = await exec('git branch --show-current 2>/dev/null || echo "HEAD"');
+  const diffStat = await exec(`git diff ${baseRef}...HEAD --stat 2>/dev/null | head -100`);
+  const changedFiles = await exec(`git diff ${baseRef}...HEAD --name-only 2>/dev/null | head -60`);
+  const commitMessages = await exec(`git log ${baseRef}..HEAD --oneline 2>/dev/null | head -30`);
+  let worktreeUnstaged = "";
+  let worktreeStaged = "";
+  let worktreeFilesUnstaged = "";
+  let worktreeFilesStaged = "";
+  if (includeWorktree) {
+    worktreeUnstaged = await exec("git diff --stat 2>/dev/null | head -100");
+    worktreeStaged = await exec("git diff --cached --stat 2>/dev/null | head -100");
+    worktreeFilesUnstaged = await exec("git diff --name-only 2>/dev/null | head -60");
+    worktreeFilesStaged = await exec("git diff --cached --name-only 2>/dev/null | head -60");
+  }
+  return [
+    `BASE: ${baseRef}`,
+    `HEAD: ${currentBranch}`,
+    "",
+    "COMMITS (base..HEAD):",
+    commitMessages,
+    "",
+    "CHANGED FILES (base...HEAD):",
+    changedFiles,
+    "",
+    "DIFF STATS (base...HEAD):",
+    diffStat,
+    "",
+    "UNCOMMITTED (unstaged) DIFF STATS:",
+    worktreeUnstaged,
+    "",
+    "UNCOMMITTED (staged) DIFF STATS:",
+    worktreeStaged,
+    "",
+    "UNCOMMITTED FILES (unstaged):",
+    worktreeFilesUnstaged,
+    "",
+    "UNCOMMITTED FILES (staged):",
+    worktreeFilesStaged
+  ].join("\n");
+}
+
+// ../review/dist/prompt.js
+function buildPlanPrompt(gitContext, hypothesesCount, userHint) {
+  const hint = userHint && userHint.trim().length > 0 ? userHint : "NONE";
+  return `You are generating a detailed test specification based on git changes.
+
+RULES:
+- Do NOT use any tools. Do NOT browse. Do NOT run commands.
+- Output MUST be a complete, actionable proof spec for a human/agent to execute later.
+- Include Hypotheses H1.. up to H${hypothesesCount}. Each hypothesis should be short and testable.
+- Prefer UI/UX observable outcomes. Avoid implementation details unless necessary for test setup.
+
+CONTEXT (git summary):
+${gitContext}
+
+USER HINT (optional):
+${hint}
+
+OUTPUT FORMAT (exact headings, in this order):
+---GOAL---
+[1-3 sentences]
+---SCOPE---
+[bullets]
+---NON_GOALS---
+[bullets]
+---PRECONDITIONS---
+[bullets]
+---HYPOTHESES---
+H1: ...
+...
+H${hypothesesCount}: ...
+---PROOF_PLAN---
+1. ...
+2. ...
+[each step includes expected observation]
+---TEST_DATA---
+[bullets or NONE]
+---RISKS_AND_FLAKINESS---
+[bullets]
+---PASS_FAIL---
+PASS: ...
+FAIL: ...
+---END---`;
+}
+
+// ../review/dist/review.js
+function buildTextOutput(output) {
+  const lines = [
+    "agex-review",
+    `agent=${output.agent}`,
+    `model=${output.model ?? ""}`,
+    `base_ref=${output.baseRef}`,
+    "---plan---",
+    output.plan.stdout.trimEnd(),
+    "---execution---",
+    output.execution.stdout.trimEnd()
+  ];
+  return lines.join("\n");
+}
+function normalizeReviewOutput(output) {
+  return {
+    json: output,
+    text: buildTextOutput(output)
+  };
+}
+async function runReview(request) {
+  const hypothesesCount = request.hypotheses ?? 5;
+  const includeWorktree = request.includeWorktree ?? true;
+  const gitContext = await buildGitContext(request.baseRef, includeWorktree, request.cwd);
+  const planPrompt = buildPlanPrompt(gitContext, hypothesesCount, request.promptHint);
+  const plan = await runAgent({
+    agent: request.agent,
+    model: request.model,
+    prompt: planPrompt,
+    cwd: request.cwd
+  });
+  const execution = await runAgent({
+    agent: request.agent,
+    model: request.model,
+    prompt: plan.json.stdout,
+    cwd: request.cwd
+  });
+  const outputJson = {
+    baseRef: request.baseRef,
+    agent: request.agent,
+    model: request.model,
+    plan: plan.json,
+    execution: execution.json
+  };
+  return normalizeReviewOutput(outputJson);
+}
+
+export {
+  runAgent,
+  proov,
+  runReview
+};

package/dist/cli.d.ts

@@ -0,0 +1,4 @@
+#!/usr/bin/env node
+import 'agex-proov';
+import 'agex-run';
+import 'agex-review';