agent-gauntlet 0.13.0 → 0.13.1

package/dist/index.js

@@ -7,7 +7,7 @@ import { Command } from "commander";
 // package.json
 var package_default = {
   name: "agent-gauntlet",
-  version: "0.13.0",
+  version: "0.13.1",
   description: "A CLI tool for testing AI coding agents",
   license: "Apache-2.0",
   author: "Paul Caplan",
@@ -5244,7 +5244,7 @@ import fs19 from "node:fs";
 import fsPromises2 from "node:fs/promises";
 import path17 from "node:path";
 import { inspect } from "node:util";
-var ANSI_REGEX = /\x1b\[[0-9;]*m/g;
+var ANSI_REGEX = /\x1b(?:\[[0-9;?]*[A-Za-z]|[78])/g;
 function stripAnsi(text) {
   return text.replace(ANSI_REGEX, "");
 }
@@ -5700,16 +5700,16 @@ function registerCheckCommand(program) {
       await debugLogger?.logRunStart(runMode, changes.length, jobs.length);
       const reporter = new ConsoleReporter;
       const runner = new Runner(config, logger, reporter, failuresMap, changeOptions, effectiveBaseBranch, passedSlotsMap, debugLogger ?? undefined, isRerun);
-      const success = await runner.run(jobs);
-      await debugLogger?.logRunEnd(success ? "pass" : "fail", 0, 0, 0, logger.getRunNumber());
+      const outcome = await runner.run(jobs);
+      await debugLogger?.logRunEnd(outcome.allPassed ? "pass" : "fail", outcome.stats.fixed, outcome.stats.skipped, outcome.stats.failed, logger.getRunNumber());
       await writeExecutionState(config.project.log_dir);
-      if (success) {
+      if (outcome.allPassed) {
         await debugLogger?.logClean("auto", "all_passed");
         await cleanLogs(config.project.log_dir);
       }
       await releaseLock(config.project.log_dir);
       restoreConsole?.restore();
-      process.exit(success ? 0 : 1);
+      process.exit(outcome.allPassed ? 0 : 1);
     } catch (error) {
       if (config && lockAcquired) {
         try {
@@ -6760,7 +6760,7 @@ allowed-tools: Bash
   if (isRun) {
     steps.push(`3. If it fails:
    - Identify the failed gates from the console output.
-   - For CHECK failures: Read the \`.log\` file path provided in the output.
+   - For CHECK failures: Read the \`.log\` file path provided in the output. If the log contains a \`--- Fix Instructions ---\` section, follow those instructions to fix the issue. If it contains a \`--- Fix Skill: <name> ---\` section, invoke that skill.
    - For REVIEW failures: Read the \`.json\` file path provided in the "Review: <path>" output.
 4. Address the violations:
    - For REVIEW violations: You MUST update the \`"status"\` and \`"result"\` fields in the provided \`.json\` file for EACH violation.
@@ -6779,7 +6779,7 @@ allowed-tools: Bash
    - Outstanding Failures: (if retry limit exceeded, list unverified fixes and remaining issues)`);
   } else {
     steps.push(`3. If any checks fail:
-   - Read the \`.log\` file path provided in the output for each failed check.
+   - Read the \`.log\` file path provided in the output for each failed check. If the log contains a \`--- Fix Instructions ---\` section, follow those instructions. If it contains a \`--- Fix Skill: <name> ---\` section, invoke that skill.
    - Fix the issues found.
 4. Run \`${command}\` again to verify your fixes. Do NOT run \`agent-gauntlet clean\` between retries.
 5. Repeat steps 3-4 until all checks pass or you've made 3 attempts.
@@ -6837,6 +6837,7 @@ var HELP_SKILL_BUNDLE = {
 };
 var SETUP_SKILL_CONTENT = readSkillTemplate("setup-skill.md");
 var CHECK_CATALOG_REFERENCE = readSkillTemplate("check-catalog.md");
+var PROJECT_STRUCTURE_REFERENCE = readSkillTemplate("setup-ref-project-structure.md");
 var SKILL_DEFINITIONS = [
   { action: "run", content: GAUNTLET_RUN_SKILL_CONTENT },
   { action: "check", content: GAUNTLET_CHECK_SKILL_CONTENT },
@@ -6852,7 +6853,10 @@ var SKILL_DEFINITIONS = [
   {
     action: "setup",
     content: SETUP_SKILL_CONTENT,
-    references: { "check-catalog.md": CHECK_CATALOG_REFERENCE },
+    references: {
+      "check-catalog.md": CHECK_CATALOG_REFERENCE,
+      "project-structure.md": PROJECT_STRUCTURE_REFERENCE
+    },
     skillsOnly: true
   }
 ];
@@ -7006,14 +7010,29 @@ async function addToGitignore(projectRoot, entry) {
 `);
   console.log(chalk9.green(`Added ${entry} to .gitignore`));
 }
-async function detectBaseBranch() {
+function gitSilent(args, opts) {
+  const { execFileSync } = __require("node:child_process");
   try {
-    const { execSync } = await import("node:child_process");
-    const ref = execSync("git symbolic-ref refs/remotes/origin/HEAD 2>/dev/null", { encoding: "utf-8" }).trim();
-    if (ref) {
-      return ref.replace("refs/remotes/", "");
+    return execFileSync("git", args, {
+      encoding: "utf-8",
+      timeout: opts?.timeout,
+      stdio: ["pipe", "pipe", "ignore"]
+    }).trim();
+  } catch {
+    return null;
+  }
+}
+async function detectBaseBranch() {
+  gitSilent(["remote", "set-head", "origin", "--auto"], { timeout: 5000 });
+  const ref = gitSilent(["symbolic-ref", "refs/remotes/origin/HEAD"]);
+  if (ref) {
+    return ref.replace("refs/remotes/", "");
+  }
+  for (const candidate of ["origin/main", "origin/master"]) {
+    if (gitSilent(["rev-parse", "--verify", candidate]) !== null) {
+      return candidate;
     }
-  } catch {}
+  }
   return "origin/main";
 }
 function buildAdapterSettingsBlock(adapters3) {
@@ -7392,16 +7411,16 @@ function registerReviewCommand(program) {
       await debugLogger?.logRunStart(runMode, changes.length, jobs.length);
       const reporter = new ConsoleReporter;
       const runner = new Runner(config, logger, reporter, failuresMap, changeOptions, effectiveBaseBranch, passedSlotsMap, debugLogger ?? undefined, isRerun);
-      const success = await runner.run(jobs);
-      await debugLogger?.logRunEnd(success ? "pass" : "fail", 0, 0, 0, logger.getRunNumber());
+      const outcome = await runner.run(jobs);
+      await debugLogger?.logRunEnd(outcome.allPassed ? "pass" : "fail", outcome.stats.fixed, outcome.stats.skipped, outcome.stats.failed, logger.getRunNumber());
       await writeExecutionState(config.project.log_dir);
-      if (success) {
+      if (outcome.allPassed) {
         await debugLogger?.logClean("auto", "all_passed");
         await cleanLogs(config.project.log_dir);
       }
       await releaseLock(config.project.log_dir);
       restoreConsole?.restore();
-      process.exit(success ? 0 : 1);
+      process.exit(outcome.allPassed ? 0 : 1);
     } catch (error) {
       if (config && lockAcquired) {
         try {
@@ -8358,4 +8377,4 @@ if (process.argv.length < 3) {
 }
 program.parse(process.argv);
 
-//# debugId=D6CA917DC551041A64756E2164756E21
+//# debugId=61E76D84AB3F58F264756E2164756E21