agent-gauntlet 0.2.2 → 0.4.0

package/README.md

@@ -51,9 +51,9 @@ The use cases below illustrate when each of these patterns may be used.
 2. Run `/gauntlet` from chat
 3. Gauntlet detects changed files and runs configured checks (linter, tests, type checking, etc.)
 4. Simultaneously, Gauntlet invokes AI CLIs for code review
-5. Assistant reviews results, fixes identified issues, and runs `agent-gauntlet rerun`
-6. Gauntlet verifies fixes and checks for new issues
-7. Process repeats automatically (up to 3 reruns) until all gates pass
+5. Assistant reviews results, fixes identified issues, and runs `agent-gauntlet run` again
+6. Gauntlet detects existing logs, switches to verification mode, and checks fixes
+7. Process repeats automatically (up to 3 iterations) until all gates pass
 
 ### 3. Agentic Implementation
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "agent-gauntlet",
-  "version": "0.2.2",
+  "version": "0.4.0",
   "description": "A CLI tool for testing AI coding agents",
   "license": "Apache-2.0",
   "author": "Paul Caplan",

package/src/cli-adapters/claude.ts

@@ -40,7 +40,7 @@ export class ClaudeAdapter implements CLIAdapter {
 				// We use a simple "hello" prompt to avoid "No messages returned" errors from empty input
 				const { stdout, stderr } = await execAsync(
 					'echo "hello" | claude -p --max-turns 1',
-					{ timeout: 10000 },
+					{ timeout: 30000 },
 				);
 
 				const combined = (stdout || "") + (stderr || "");
@@ -58,7 +58,19 @@ export class ClaudeAdapter implements CLIAdapter {
 					stderr?: string;
 					stdout?: string;
 					message?: string;
+					code?: number | string;
+					signal?: string;
 				};
+
+				// Check for timeout
+				if (execError.signal === "SIGTERM" && execError.code === null) {
+					return {
+						available: true,
+						status: "unhealthy",
+						message: "Error: Health check timed out",
+					};
+				}
+
 				const stderr = execError.stderr || "";
 				const stdout = execError.stdout || "";
 				const combined = stderr + stdout;

package/src/cli-adapters/gemini.ts

@@ -38,7 +38,7 @@ export class GeminiAdapter implements CLIAdapter {
 			try {
 				const { stdout, stderr } = await execAsync(
 					'echo "hello" | gemini --sandbox --output-format text',
-					{ timeout: 10000 },
+					{ timeout: 30000 },
 				);
 
 				const combined = (stdout || "") + (stderr || "");
@@ -56,7 +56,19 @@ export class GeminiAdapter implements CLIAdapter {
 					stderr?: string;
 					stdout?: string;
 					message?: string;
+					code?: number | string;
+					signal?: string;
 				};
+
+				// Check for timeout
+				if (execError.signal === "SIGTERM" && execError.code === null) {
+					return {
+						available: true,
+						status: "unhealthy",
+						message: "Error: Health check timed out",
+					};
+				}
+
 				const stderr = execError.stderr || "";
 				const stdout = execError.stdout || "";
 				const combined = stderr + stdout;
@@ -159,7 +171,10 @@ ${escapedBody}
 
 		// Write to a temporary file to avoid shell escaping issues
 		const tmpDir = os.tmpdir();
-		const tmpFile = path.join(tmpDir, `gauntlet-gemini-${Date.now()}.txt`);
+		const tmpFile = path.join(
+			tmpDir,
+			`gauntlet-gemini-${process.pid}-${Date.now()}.txt`,
+		);
 		await fs.writeFile(tmpFile, fullContent);
 
 		try {

package/src/commands/check.ts

@@ -6,8 +6,19 @@ import { EntryPointExpander } from "../core/entry-point.js";
 import { JobGenerator } from "../core/job.js";
 import { Runner } from "../core/runner.js";
 import { ConsoleReporter } from "../output/console.js";
+import { startConsoleLog } from "../output/console-log.js";
 import { Logger } from "../output/logger.js";
-import { rotateLogs } from "./shared.js";
+import {
+	findPreviousFailures,
+	type PreviousViolation,
+} from "../utils/log-parser.js";
+import { readSessionRef, writeSessionRef } from "../utils/session-ref.js";
+import {
+	acquireLock,
+	cleanLogs,
+	hasExistingLogs,
+	releaseLock,
+} from "./shared.js";
 
 export function registerCheckCommand(program: Command): void {
 	program
@@ -24,14 +35,16 @@ export function registerCheckCommand(program: Command): void {
 			"Use diff for current uncommitted changes (staged and unstaged)",
 		)
 		.action(async (options) => {
+			let config: Awaited<ReturnType<typeof loadConfig>> | undefined;
+			let lockAcquired = false;
+			let restoreConsole: (() => void) | undefined;
 			try {
-				const config = await loadConfig();
-
-				// Rotate logs before starting
-				await rotateLogs(config.project.log_dir);
+				config = await loadConfig();
+				restoreConsole = await startConsoleLog(config.project.log_dir);
+				await acquireLock(config.project.log_dir);
+				lockAcquired = true;
 
 				// Determine effective base branch
-				// Priority: CLI override > CI env var > config
 				const effectiveBaseBranch =
 					options.baseBranch ||
 					(process.env.GITHUB_BASE_REF &&
@@ -40,10 +53,76 @@ export function registerCheckCommand(program: Command): void {
 						: null) ||
 					config.project.base_branch;
 
-				const changeDetector = new ChangeDetector(effectiveBaseBranch, {
-					commit: options.commit,
-					uncommitted: options.uncommitted,
-				});
+				// Detect rerun mode: if logs exist and not targeting a specific commit, enter verification mode
+				const logsExist = await hasExistingLogs(config.project.log_dir);
+				const isRerun = logsExist && !options.commit;
+
+				let failuresMap:
+					| Map<string, Map<string, PreviousViolation[]>>
+					| undefined;
+				let changeOptions:
+					| { commit?: string; uncommitted?: boolean; fixBase?: string }
+					| undefined;
+
+				if (isRerun) {
+					console.log(
+						chalk.dim(
+							"Existing logs detected — running in verification mode...",
+						),
+					);
+					const previousFailures = await findPreviousFailures(
+						config.project.log_dir,
+						options.gate,
+					);
+
+					failuresMap = new Map();
+					for (const gateFailure of previousFailures) {
+						const adapterMap = new Map<string, PreviousViolation[]>();
+						for (const af of gateFailure.adapterFailures) {
+							const key = af.reviewIndex
+								? String(af.reviewIndex)
+								: af.adapterName;
+							adapterMap.set(key, af.violations);
+						}
+						failuresMap.set(gateFailure.jobId, adapterMap);
+					}
+
+					if (previousFailures.length > 0) {
+						const totalViolations = previousFailures.reduce(
+							(sum, gf) =>
+								sum +
+								gf.adapterFailures.reduce(
+									(s, af) => s + af.violations.length,
+									0,
+								),
+							0,
+						);
+						console.log(
+							chalk.yellow(
+								`Found ${previousFailures.length} gate(s) with ${totalViolations} previous violation(s)`,
+							),
+						);
+					}
+
+					changeOptions = { uncommitted: true };
+					const fixBase = await readSessionRef(config.project.log_dir);
+					if (fixBase) {
+						changeOptions.fixBase = fixBase;
+					}
+				} else if (options.commit || options.uncommitted) {
+					changeOptions = {
+						commit: options.commit,
+						uncommitted: options.uncommitted,
+					};
+				}
+
+				const changeDetector = new ChangeDetector(
+					effectiveBaseBranch,
+					changeOptions || {
+						commit: options.commit,
+						uncommitted: options.uncommitted,
+					},
+				);
 				const expander = new EntryPointExpander();
 				const jobGen = new JobGenerator(config);
 
@@ -52,6 +131,8 @@ export function registerCheckCommand(program: Command): void {
 
 				if (changes.length === 0) {
 					console.log(chalk.green("No changes detected."));
+					await releaseLock(config.project.log_dir);
+					restoreConsole?.();
 					process.exit(0);
 				}
 
@@ -72,6 +153,8 @@ export function registerCheckCommand(program: Command): void {
 
 				if (jobs.length === 0) {
 					console.log(chalk.yellow("No applicable checks for these changes."));
+					await releaseLock(config.project.log_dir);
+					restoreConsole?.();
 					process.exit(0);
 				}
 
@@ -83,16 +166,29 @@ export function registerCheckCommand(program: Command): void {
 					config,
 					logger,
 					reporter,
-					undefined,
-					undefined,
+					failuresMap,
+					changeOptions,
 					effectiveBaseBranch,
 				);
 
 				const success = await runner.run(jobs);
+
+				if (success) {
+					await cleanLogs(config.project.log_dir);
+				} else {
+					await writeSessionRef(config.project.log_dir);
+				}
+
+				await releaseLock(config.project.log_dir);
+				restoreConsole?.();
 				process.exit(success ? 0 : 1);
 			} catch (error: unknown) {
+				if (config && lockAcquired) {
+					await releaseLock(config.project.log_dir);
+				}
 				const err = error as { message?: string };
 				console.error(chalk.red("Error:"), err.message);
+				restoreConsole?.();
 				process.exit(1);
 			}
 		});

package/src/commands/ci/list-jobs.ts

@@ -34,8 +34,9 @@ export async function listJobs(): Promise<void> {
 						}
 
 						const workingDirectory = checkDef.working_directory || ep.path;
-						// Include entry point in key to ensure each entry point/check pair is distinct
-						const jobKey = `${ep.path}:${check.name}:${workingDirectory}`;
+						// Dedupe by check name + working directory only - if two entry points
+						// both trigger e.g. "test" with working_directory: ".", run it once
+						const jobKey = `${check.name}:${workingDirectory}`;
 
 						// Skip if we've already created a job for this exact entry point/check combination
 						if (seenJobs.has(jobKey)) {

package/src/commands/clean.ts

@@ -0,0 +1,29 @@
+import chalk from "chalk";
+import type { Command } from "commander";
+import { loadConfig } from "../config/loader.js";
+import { acquireLock, cleanLogs, releaseLock } from "./shared.js";
+
+export function registerCleanCommand(program: Command): void {
+	program
+		.command("clean")
+		.description("Archive logs (move current logs into previous/)")
+		.action(async () => {
+			let config: Awaited<ReturnType<typeof loadConfig>> | undefined;
+			let lockAcquired = false;
+			try {
+				config = await loadConfig();
+				await acquireLock(config.project.log_dir);
+				lockAcquired = true;
+				await cleanLogs(config.project.log_dir);
+				await releaseLock(config.project.log_dir);
+				console.log(chalk.green("Logs archived successfully."));
+			} catch (error: unknown) {
+				if (config && lockAcquired) {
+					await releaseLock(config.project.log_dir);
+				}
+				const err = error as { message?: string };
+				console.error(chalk.red("Error:"), err.message);
+				process.exit(1);
+			}
+		});
+}

package/src/commands/help.ts

@@ -15,9 +15,9 @@ export function registerHelpCommand(program: Command): void {
 			);
 			console.log(chalk.bold("Commands:\n"));
 			console.log("  run      Run gates for detected changes");
-			console.log("  rerun    Rerun gates with previous failure context");
 			console.log("  check    Run only applicable checks");
 			console.log("  review   Run only applicable reviews");
+			console.log("  clean    Archive logs (move current logs into previous/)");
 			console.log(
 				"  detect   Show what gates would run (without executing them)",
 			);

package/src/commands/index.ts

@@ -1,10 +1,11 @@
 export { registerCheckCommand } from "./check.js";
 export { registerCICommand } from "./ci/index.js";
+export { registerCleanCommand } from "./clean.js";
 export { registerDetectCommand } from "./detect.js";
 export { registerHealthCommand } from "./health.js";
 export { registerHelpCommand } from "./help.js";
 export { registerInitCommand } from "./init.js";
 export { registerListCommand } from "./list.js";
-export { registerRerunCommand } from "./rerun.js";
 export { registerReviewCommand } from "./review.js";
 export { registerRunCommand } from "./run.js";
+export { registerValidateCommand } from "./validate.js";

package/src/commands/init.ts

@@ -35,12 +35,12 @@ Execute the autonomous verification suite.
    - If fix instructions are available, they will be in the log file under a "--- Fix Instructions ---" section—carefully read and apply them FIRST before attempting other fixes.
 3. Fix any code or logic errors found by the tools or AI reviewers, prioritizing higher-priority violations (critical > high > medium > low).
 4. Apply the trust level above when deciding whether to act on AI reviewer feedback. If you skip an issue due to the trust threshold, report it with a brief explanation (e.g., "Skipped: [issue summary] — reason: [stylistic/subjective/disagree]").
-5. Do NOT commit your changes yet—keep them uncommitted so the rerun command can review them.
-6. Run \`agent-gauntlet rerun\` to verify your fixes. The rerun command reviews only uncommitted changes and uses previous failures as context.
+5. Do NOT commit your changes yet—keep them uncommitted so the next run can verify them.
+6. Run \`agent-gauntlet run\` again to verify your fixes. It will detect existing logs and automatically switch to verification mode (uncommitted changes + previous failure context).
 7. Repeat steps 2-6 until one of the following termination conditions is met:
-   - All gates pass
+   - All gates pass (logs are automatically archived)
    - You disagree with remaining failures (ask the human how to proceed)
-   - Still failing after 3 rerun attempts
+   - Still failing after 3 attempts
 8. Once all gates pass, do NOT commit or push your changes—await the human's review and explicit instruction to commit.
 `;
 

package/src/commands/review.ts

@@ -6,8 +6,19 @@ import { EntryPointExpander } from "../core/entry-point.js";
 import { JobGenerator } from "../core/job.js";
 import { Runner } from "../core/runner.js";
 import { ConsoleReporter } from "../output/console.js";
+import { startConsoleLog } from "../output/console-log.js";
 import { Logger } from "../output/logger.js";
-import { rotateLogs } from "./shared.js";
+import {
+	findPreviousFailures,
+	type PreviousViolation,
+} from "../utils/log-parser.js";
+import { readSessionRef, writeSessionRef } from "../utils/session-ref.js";
+import {
+	acquireLock,
+	cleanLogs,
+	hasExistingLogs,
+	releaseLock,
+} from "./shared.js";
 
 export function registerReviewCommand(program: Command): void {
 	program
@@ -24,14 +35,16 @@ export function registerReviewCommand(program: Command): void {
 			"Use diff for current uncommitted changes (staged and unstaged)",
 		)
 		.action(async (options) => {
+			let config: Awaited<ReturnType<typeof loadConfig>> | undefined;
+			let lockAcquired = false;
+			let restoreConsole: (() => void) | undefined;
 			try {
-				const config = await loadConfig();
-
-				// Rotate logs before starting
-				await rotateLogs(config.project.log_dir);
+				config = await loadConfig();
+				restoreConsole = await startConsoleLog(config.project.log_dir);
+				await acquireLock(config.project.log_dir);
+				lockAcquired = true;
 
 				// Determine effective base branch
-				// Priority: CLI override > CI env var > config
 				const effectiveBaseBranch =
 					options.baseBranch ||
 					(process.env.GITHUB_BASE_REF &&
@@ -40,10 +53,76 @@ export function registerReviewCommand(program: Command): void {
 						: null) ||
 					config.project.base_branch;
 
-				const changeDetector = new ChangeDetector(effectiveBaseBranch, {
-					commit: options.commit,
-					uncommitted: options.uncommitted,
-				});
+				// Detect rerun mode: if logs exist and not targeting a specific commit, enter verification mode
+				const logsExist = await hasExistingLogs(config.project.log_dir);
+				const isRerun = logsExist && !options.commit;
+
+				let failuresMap:
+					| Map<string, Map<string, PreviousViolation[]>>
+					| undefined;
+				let changeOptions:
+					| { commit?: string; uncommitted?: boolean; fixBase?: string }
+					| undefined;
+
+				if (isRerun) {
+					console.log(
+						chalk.dim(
+							"Existing logs detected — running in verification mode...",
+						),
+					);
+					const previousFailures = await findPreviousFailures(
+						config.project.log_dir,
+						options.gate,
+					);
+
+					failuresMap = new Map();
+					for (const gateFailure of previousFailures) {
+						const adapterMap = new Map<string, PreviousViolation[]>();
+						for (const af of gateFailure.adapterFailures) {
+							const key = af.reviewIndex
+								? String(af.reviewIndex)
+								: af.adapterName;
+							adapterMap.set(key, af.violations);
+						}
+						failuresMap.set(gateFailure.jobId, adapterMap);
+					}
+
+					if (previousFailures.length > 0) {
+						const totalViolations = previousFailures.reduce(
+							(sum, gf) =>
+								sum +
+								gf.adapterFailures.reduce(
+									(s, af) => s + af.violations.length,
+									0,
+								),
+							0,
+						);
+						console.log(
+							chalk.yellow(
+								`Found ${previousFailures.length} gate(s) with ${totalViolations} previous violation(s)`,
+							),
+						);
+					}
+
+					changeOptions = { uncommitted: true };
+					const fixBase = await readSessionRef(config.project.log_dir);
+					if (fixBase) {
+						changeOptions.fixBase = fixBase;
+					}
+				} else if (options.commit || options.uncommitted) {
+					changeOptions = {
+						commit: options.commit,
+						uncommitted: options.uncommitted,
+					};
+				}
+
+				const changeDetector = new ChangeDetector(
+					effectiveBaseBranch,
+					changeOptions || {
+						commit: options.commit,
+						uncommitted: options.uncommitted,
+					},
+				);
 				const expander = new EntryPointExpander();
 				const jobGen = new JobGenerator(config);
 
@@ -52,6 +131,8 @@ export function registerReviewCommand(program: Command): void {
 
 				if (changes.length === 0) {
 					console.log(chalk.green("No changes detected."));
+					await releaseLock(config.project.log_dir);
+					restoreConsole?.();
 					process.exit(0);
 				}
 
@@ -72,6 +153,8 @@ export function registerReviewCommand(program: Command): void {
 
 				if (jobs.length === 0) {
 					console.log(chalk.yellow("No applicable reviews for these changes."));
+					await releaseLock(config.project.log_dir);
+					restoreConsole?.();
 					process.exit(0);
 				}
 
@@ -83,16 +166,29 @@ export function registerReviewCommand(program: Command): void {
 					config,
 					logger,
 					reporter,
-					undefined,
-					undefined,
+					failuresMap,
+					changeOptions,
 					effectiveBaseBranch,
 				);
 
 				const success = await runner.run(jobs);
+
+				if (success) {
+					await cleanLogs(config.project.log_dir);
+				} else {
+					await writeSessionRef(config.project.log_dir);
+				}
+
+				await releaseLock(config.project.log_dir);
+				restoreConsole?.();
 				process.exit(success ? 0 : 1);
 			} catch (error: unknown) {
+				if (config && lockAcquired) {
+					await releaseLock(config.project.log_dir);
+				}
 				const err = error as { message?: string };
 				console.error(chalk.red("Error:"), err.message);
+				restoreConsole?.();
 				process.exit(1);
 			}
 		});