agent-gauntlet 0.2.2 → 0.4.0

package/src/commands/review.test.ts

@@ -1,31 +0,0 @@
-import { beforeEach, describe, expect, it } from "bun:test";
-import { Command } from "commander";
-import { registerReviewCommand } from "./review.js";
-
-describe("Review Command", () => {
-	let program: Command;
-
-	beforeEach(() => {
-		program = new Command();
-		registerReviewCommand(program);
-	});
-
-	it("should register the review command", () => {
-		const reviewCmd = program.commands.find((cmd) => cmd.name() === "review");
-		expect(reviewCmd).toBeDefined();
-		expect(reviewCmd?.description()).toBe(
-			"Run only applicable reviews for detected changes",
-		);
-	});
-
-	it("should have correct options", () => {
-		const reviewCmd = program.commands.find((cmd) => cmd.name() === "review");
-		expect(reviewCmd?.options.some((opt) => opt.long === "--gate")).toBe(true);
-		expect(reviewCmd?.options.some((opt) => opt.long === "--commit")).toBe(
-			true,
-		);
-		expect(reviewCmd?.options.some((opt) => opt.long === "--uncommitted")).toBe(
-			true,
-		);
-	});
-});

package/src/commands/run.test.ts

@@ -1,27 +0,0 @@
-import { beforeEach, describe, expect, it } from "bun:test";
-import { Command } from "commander";
-import { registerRunCommand } from "./run.js";
-
-describe("Run Command", () => {
-	let program: Command;
-
-	beforeEach(() => {
-		program = new Command();
-		registerRunCommand(program);
-	});
-
-	it("should register the run command", () => {
-		const runCmd = program.commands.find((cmd) => cmd.name() === "run");
-		expect(runCmd).toBeDefined();
-		expect(runCmd?.description()).toBe("Run gates for detected changes");
-	});
-
-	it("should have correct options", () => {
-		const runCmd = program.commands.find((cmd) => cmd.name() === "run");
-		expect(runCmd?.options.some((opt) => opt.long === "--gate")).toBe(true);
-		expect(runCmd?.options.some((opt) => opt.long === "--commit")).toBe(true);
-		expect(runCmd?.options.some((opt) => opt.long === "--uncommitted")).toBe(
-			true,
-		);
-	});
-});

package/src/config/loader.test.ts

@@ -1,151 +0,0 @@
-import { afterAll, beforeAll, describe, expect, it } from "bun:test";
-import fs from "node:fs/promises";
-import path from "node:path";
-import { loadConfig } from "./loader.js";
-
-const TEST_DIR = path.join(process.cwd(), `test-env-${Date.now()}`);
-const GAUNTLET_DIR = path.join(TEST_DIR, ".gauntlet");
-const CHECKS_DIR = path.join(GAUNTLET_DIR, "checks");
-const REVIEWS_DIR = path.join(GAUNTLET_DIR, "reviews");
-
-describe("Config Loader", () => {
-	beforeAll(async () => {
-		// Setup directory structure
-		await fs.mkdir(TEST_DIR);
-		await fs.mkdir(GAUNTLET_DIR);
-		await fs.mkdir(CHECKS_DIR);
-		await fs.mkdir(REVIEWS_DIR);
-
-		// Write config.yml
-		await fs.writeFile(
-			path.join(GAUNTLET_DIR, "config.yml"),
-			`
-base_branch: origin/dev
-log_dir: test_logs
-cli:
-  default_preference:
-    - claude
-    - gemini
-  check_usage_limit: false
-entry_points:
-  - path: src/
-    checks:
-      - lint
-    reviews:
-      - security
-`,
-		);
-
-		// Write a check definition
-		await fs.writeFile(
-			path.join(CHECKS_DIR, "lint.yml"),
-			`
-name: lint
-command: npm run lint
-working_directory: .
-`,
-		);
-
-		// Write a review definition
-		await fs.writeFile(
-			path.join(REVIEWS_DIR, "security.md"),
-			`---
-cli_preference:
-  - gemini
----
-
-# Security Review
-Check for vulnerabilities.
-`,
-		);
-
-		// Write a review definition without preference
-		await fs.writeFile(
-			path.join(REVIEWS_DIR, "style.md"),
-			`---
-num_reviews: 1
----
-
-# Style Review
-Check style.
-`,
-		);
-	});
-
-	afterAll(async () => {
-		// Cleanup
-		await fs.rm(TEST_DIR, { recursive: true, force: true });
-	});
-
-	it("should load project configuration correctly", async () => {
-		const config = await loadConfig(TEST_DIR);
-
-		expect(config.project.base_branch).toBe("origin/dev");
-		expect(config.project.log_dir).toBe("test_logs");
-		expect(config.project.entry_points).toHaveLength(1);
-		expect(config.project.entry_points[0].path).toBe("src/");
-	});
-
-	it("should load check gates correctly", async () => {
-		const config = await loadConfig(TEST_DIR);
-
-		expect(Object.keys(config.checks)).toContain("lint");
-		expect(config.checks.lint.command).toBe("npm run lint");
-	});
-
-	it("should load review gates correctly", async () => {
-		const config = await loadConfig(TEST_DIR);
-
-		expect(Object.keys(config.reviews)).toContain("security");
-		expect(config.reviews.security.name).toBe("security");
-		expect(config.reviews.security.cli_preference).toEqual(["gemini"]);
-		expect(config.reviews.security.promptContent).toContain(
-			"Check for vulnerabilities.",
-		);
-	});
-
-	it("should merge default cli preference", async () => {
-		const config = await loadConfig(TEST_DIR);
-
-		expect(Object.keys(config.reviews)).toContain("style");
-		expect(config.reviews.style.cli_preference).toEqual(["claude", "gemini"]);
-	});
-
-	it("should reject check gate with fail_fast when parallel is true", async () => {
-		await fs.writeFile(
-			path.join(CHECKS_DIR, "invalid.yml"),
-			`
-name: invalid
-command: echo test
-parallel: true
-fail_fast: true
-`,
-		);
-
-		await expect(loadConfig(TEST_DIR)).rejects.toThrow(
-			/fail_fast can only be used when parallel is false/,
-		);
-	});
-
-	it("should accept check gate with fail_fast when parallel is false", async () => {
-		// Clean up the invalid file first
-		try {
-			await fs.unlink(path.join(CHECKS_DIR, "invalid.yml"));
-		} catch {}
-
-		await fs.writeFile(
-			path.join(CHECKS_DIR, "valid.yml"),
-			`
-name: valid
-command: echo test
-parallel: false
-fail_fast: true
-`,
-		);
-
-		const config = await loadConfig(TEST_DIR);
-		expect(config.checks.valid).toBeDefined();
-		expect(config.checks.valid.fail_fast).toBe(true);
-		expect(config.checks.valid.parallel).toBe(false);
-	});
-});

package/src/core/entry-point.test.ts

@@ -1,61 +0,0 @@
-import { describe, expect, it } from "bun:test";
-import type { EntryPointConfig } from "../config/types.js";
-import { EntryPointExpander } from "./entry-point.js";
-
-describe("EntryPointExpander", () => {
-	const expander = new EntryPointExpander();
-
-	it("should include root entry point if there are any changes", async () => {
-		const entryPoints: EntryPointConfig[] = [{ path: "." }];
-		const changes = ["some/file.ts"];
-
-		const result = await expander.expand(entryPoints, changes);
-
-		expect(result).toHaveLength(1);
-		expect(result[0].path).toBe(".");
-	});
-
-	it("should match fixed directory entry points", async () => {
-		const entryPoints: EntryPointConfig[] = [
-			{ path: "apps/api" },
-			{ path: "apps/web" },
-		];
-		const changes = ["apps/api/src/index.ts"];
-
-		const result = await expander.expand(entryPoints, changes);
-
-		// Result should have root (implicit or explicit fallback in code) + matched
-		// Looking at code: "if (changedFiles.length > 0) ... results.push({ path: '.', ... })"
-		// Wait, the code creates a default root config if one isn't provided in the list?
-		// Code: "const rootConfig = rootEntryPoint ?? { path: '.' }; results.push({ path: '.', config: rootConfig });"
-		// Yes, it always pushes root if changes > 0.
-
-		expect(result.some((r) => r.path === "apps/api")).toBe(true);
-		expect(result.some((r) => r.path === "apps/web")).toBe(false);
-	});
-
-	it("should match wildcard entry points", async () => {
-		const entryPoints: EntryPointConfig[] = [{ path: "packages/*" }];
-		const changes = [
-			"packages/ui/button.ts",
-			"packages/utils/helper.ts",
-			"other/file.ts",
-		];
-
-		const result = await expander.expand(entryPoints, changes);
-
-		const paths = result.map((r) => r.path);
-		expect(paths).toContain("packages/ui");
-		expect(paths).toContain("packages/utils");
-		expect(paths).not.toContain("packages/other");
-	});
-
-	it("should handle no changes", async () => {
-		const entryPoints: EntryPointConfig[] = [{ path: "." }];
-		const changes: string[] = [];
-
-		const result = await expander.expand(entryPoints, changes);
-
-		expect(result).toHaveLength(0);
-	});
-});

package/src/gates/review.test.ts

@@ -1,291 +0,0 @@
-import { afterEach, beforeEach, describe, expect, it, mock } from "bun:test";
-import fs from "node:fs/promises";
-import path from "node:path";
-import type { CLIAdapter } from "../cli-adapters/index.js";
-import type {
-	ReviewGateConfig,
-	ReviewPromptFrontmatter,
-} from "../config/types.js";
-import { Logger } from "../output/logger.js";
-import type { ReviewGateExecutor } from "./review.js";
-
-const TEST_DIR = path.join(process.cwd(), `test-review-logs-${Date.now()}`);
-
-describe("ReviewGateExecutor Logging", () => {
-	let logger: Logger;
-	let executor: ReviewGateExecutor;
-	let originalCI: string | undefined;
-	let originalGithubActions: string | undefined;
-	let originalCwd: string;
-
-	beforeEach(async () => {
-		await fs.mkdir(TEST_DIR, { recursive: true });
-
-		// Save and disable CI mode for this test to avoid complex git ref issues
-		originalCI = process.env.CI;
-		originalGithubActions = process.env.GITHUB_ACTIONS;
-		originalCwd = process.cwd();
-		delete process.env.CI;
-		delete process.env.GITHUB_ACTIONS;
-
-		// Change to test directory with its own git repo to avoid issues with the main repo
-		process.chdir(TEST_DIR);
-		// Initialize a minimal git repo for the test
-		const { exec } = await import("node:child_process");
-		const { promisify } = await import("node:util");
-		const execAsync = promisify(exec);
-		await execAsync("git init");
-		await execAsync('git config user.email "test@test.com"');
-		await execAsync('git config user.name "Test"');
-		// Create an initial commit so we have a history
-		await fs.writeFile("test.txt", "initial");
-		await execAsync("git add test.txt");
-		await execAsync('git commit -m "initial"');
-		// Create a "main" branch
-		await execAsync("git branch -M main");
-		// Create src directory for our test
-		await fs.mkdir("src", { recursive: true });
-		await fs.writeFile("src/test.ts", "test content");
-		await execAsync("git add src/test.ts");
-		await execAsync('git commit -m "add src"');
-
-		// Make uncommitted changes so the diff isn't empty
-		await fs.writeFile("src/test.ts", "modified test content");
-
-		// Now create the log directory and logger in the test directory
-		await fs.mkdir("logs", { recursive: true });
-		logger = new Logger(path.join(process.cwd(), "logs"));
-
-		// Create a factory function for mock adapters that returns the correct name
-		const createMockAdapter = (name: string): CLIAdapter =>
-			({
-				name,
-				isAvailable: async () => true,
-				checkHealth: async () => ({ status: "healthy" }),
-				// execute returns the raw string output from the LLM, which is then parsed by the executor.
-				// The real adapter returns a string. In this test, we return a JSON string to simulate
-				// the LLM returning structured data. This IS intentional and matches the expected contract
-				// where execute() -> Promise<string>.
-				execute: async () => {
-					await new Promise((r) => setTimeout(r, 1)); // Simulate async work
-					return JSON.stringify({ status: "pass", message: "OK" });
-				},
-				getProjectCommandDir: () => null,
-				getUserCommandDir: () => null,
-				getCommandExtension: () => "md",
-				canUseSymlink: () => false,
-				transformCommand: (c: string) => c,
-			}) as unknown as CLIAdapter;
-
-		// Mock getAdapter and other exports that may be imported by other modules
-		mock.module("../cli-adapters/index.js", () => ({
-			getAdapter: (name: string) => createMockAdapter(name),
-			getAllAdapters: () => [
-				createMockAdapter("codex"),
-				createMockAdapter("claude"),
-			],
-			getProjectCommandAdapters: () => [
-				createMockAdapter("codex"),
-				createMockAdapter("claude"),
-			],
-			getUserCommandAdapters: () => [
-				createMockAdapter("codex"),
-				createMockAdapter("claude"),
-			],
-			getValidCLITools: () => ["codex", "claude", "gemini"],
-		}));
-
-		const { ReviewGateExecutor } = await import("./review.js");
-		executor = new ReviewGateExecutor();
-	});
-
-	afterEach(async () => {
-		// Restore working directory first
-		process.chdir(originalCwd);
-
-		await fs.rm(TEST_DIR, { recursive: true, force: true });
-		mock.restore();
-
-		// Restore CI env vars
-		if (originalCI !== undefined) {
-			process.env.CI = originalCI;
-		}
-		if (originalGithubActions !== undefined) {
-			process.env.GITHUB_ACTIONS = originalGithubActions;
-		}
-	});
-
-	it("should only create adapter-specific logs and no generic log", async () => {
-		const jobId = "review:src:code-quality";
-		const config: ReviewGateConfig & ReviewPromptFrontmatter = {
-			name: "code-quality",
-			cli_preference: ["codex", "claude"],
-			num_reviews: 2,
-		};
-
-		const loggerFactory = logger.createLoggerFactory(jobId);
-
-		// We need to mock getDiff since it uses execAsync which we mocked
-		// Actually ReviewGateExecutor is a class, we can mock its private method if needed
-		// or just let it run if the mock promisify works.
-
-		const result = await executor.execute(
-			jobId,
-			config,
-			"src/",
-			loggerFactory,
-			"main",
-		);
-
-		// Enhanced error messages for better debugging
-		if (result.status !== "pass") {
-			throw new Error(
-				`Expected result.status to be "pass" but got "${result.status}". Message: ${result.message || "none"}. Duration: ${result.duration}ms`,
-			);
-		}
-
-		if (!result.logPaths) {
-			throw new Error(
-				`Expected result.logPaths to be defined but got ${JSON.stringify(result.logPaths)}`,
-			);
-		}
-
-		if (result.logPaths.length !== 2) {
-			throw new Error(
-				`Expected result.logPaths to have length 2 but got ${result.logPaths.length}. Paths: ${JSON.stringify(result.logPaths)}`,
-			);
-		}
-
-		if (!result.logPaths[0]?.includes("review_src_code-quality_codex.log")) {
-			throw new Error(
-				`Expected result.logPaths[0] to contain "review_src_code-quality_codex.log" but got "${result.logPaths[0]}"`,
-			);
-		}
-
-		if (!result.logPaths[1]?.includes("review_src_code-quality_claude.log")) {
-			throw new Error(
-				`Expected result.logPaths[1] to contain "review_src_code-quality_claude.log" but got "${result.logPaths[1]}"`,
-			);
-		}
-
-		const files = await fs.readdir("logs");
-		const filesList = files.join(", ");
-
-		if (!files.includes("review_src_code-quality_codex.log")) {
-			throw new Error(
-				`Expected log directory to contain "review_src_code-quality_codex.log" but only found: [${filesList}]`,
-			);
-		}
-
-		if (!files.includes("review_src_code-quality_claude.log")) {
-			throw new Error(
-				`Expected log directory to contain "review_src_code-quality_claude.log" but only found: [${filesList}]`,
-			);
-		}
-
-		if (files.includes("review_src_code-quality.log")) {
-			throw new Error(
-				`Expected log directory NOT to contain generic log "review_src_code-quality.log" but it was found. All files: [${filesList}]`,
-			);
-		}
-
-		// Verify multiplexed content
-		const codexLog = await fs.readFile(
-			"logs/review_src_code-quality_codex.log",
-			"utf-8",
-		);
-		if (!codexLog.includes("Starting review: code-quality")) {
-			throw new Error(
-				`Expected codex log to contain "Starting review: code-quality" but got: ${codexLog.substring(0, 200)}...`,
-			);
-		}
-		if (!codexLog.includes("Review result (codex): pass")) {
-			throw new Error(
-				`Expected codex log to contain "Review result (codex): pass" but got: ${codexLog.substring(0, 200)}...`,
-			);
-		}
-
-		const claudeLog = await fs.readFile(
-			"logs/review_src_code-quality_claude.log",
-			"utf-8",
-		);
-		if (!claudeLog.includes("Starting review: code-quality")) {
-			throw new Error(
-				`Expected claude log to contain "Starting review: code-quality" but got: ${claudeLog.substring(0, 200)}...`,
-			);
-		}
-		if (!claudeLog.includes("Review result (claude): pass")) {
-			throw new Error(
-				`Expected claude log to contain "Review result (claude): pass" but got: ${claudeLog.substring(0, 200)}...`,
-			);
-		}
-	});
-
-	it("should be handled correctly by ConsoleReporter", async () => {
-		const jobId = "review:src:code-quality";
-		const codexPath = "logs/review_src_code-quality_codex.log";
-		const claudePath = "logs/review_src_code-quality_claude.log";
-
-		await fs.writeFile(
-			codexPath,
-			`
-[2026-01-14T10:00:00.000Z] Starting review: code-quality
---- Parsed Result (codex) ---
-Status: FAIL
-Violations:
-1. src/index.ts:10 - Security risk
-   Fix: Use a safer method
-`,
-		);
-
-		await fs.writeFile(
-			claudePath,
-			`
-[2026-01-14T10:00:00.000Z] Starting review: code-quality
---- Parsed Result (claude) ---
-Status: FAIL
-Violations:
-1. src/main.ts:20 - Style issue
-   Fix: Rename variable
-`,
-		);
-
-		const result = {
-			jobId,
-			status: "fail" as const,
-			duration: 1000,
-			message: "Found violations",
-			logPaths: [codexPath, claudePath],
-		};
-
-		const { ConsoleReporter } = await import("../output/console.js");
-		const reporter = new ConsoleReporter();
-
-		// We can access extractFailureDetails directly as it is public
-		const details = await reporter.extractFailureDetails(result);
-
-		// Check for presence of key information rather than exact counts
-		expect(
-			details.some(
-				(d: string) =>
-					d.includes("src/index.ts") &&
-					d.includes("10") &&
-					d.includes("Security risk"),
-			),
-		).toBe(true);
-		expect(details.some((d: string) => d.includes("Use a safer method"))).toBe(
-			true,
-		);
-		expect(
-			details.some(
-				(d: string) =>
-					d.includes("src/main.ts") &&
-					d.includes("20") &&
-					d.includes("Style issue"),
-			),
-		).toBe(true);
-		expect(details.some((d: string) => d.includes("Rename variable"))).toBe(
-			true,
-		);
-	});
-});