agent-gauntlet 0.15.3 → 0.15.4

package/dist/index.js

@@ -299,7 +299,7 @@ import { Command } from "commander";
 // package.json
 var package_default = {
   name: "agent-gauntlet",
-  version: "0.15.3",
+  version: "0.15.4",
   description: "A CLI tool for testing AI coding agents",
   license: "MIT",
   author: "Paul Caplan",
@@ -882,15 +882,15 @@ class ChangeDetector {
       }
       return this.getCommitChangedFiles(this.options.commit);
     }
-    if (this.options.uncommitted) {
-      return this.getUncommittedChangedFiles();
-    }
     if (this.options.fixBase) {
       if (!isValidGitRef(this.options.fixBase)) {
         throw new Error(`Invalid fixBase ref: ${this.options.fixBase}`);
       }
       return this.getFixBaseChangedFiles(this.options.fixBase);
     }
+    if (this.options.uncommitted) {
+      return this.getUncommittedChangedFiles();
+    }
     const isCI = process.env.CI === "true" || process.env.GITHUB_ACTIONS === "true";
     if (isCI) {
       return this.getCIChangedFiles();
@@ -2855,7 +2855,7 @@ function classifyBlock(block) {
   if (block.includes("descriptor:") && block.includes("dataPointType:") && block.includes("dataPoints:")) {
     return "metric";
   }
-  if (block.includes("resource:") && /body:\s*'claude_code\.\w+'/.test(block)) {
+  if (block.includes("resource:") && /body:\s*['"]claude_code\.\w+['"]/.test(block)) {
     return "log";
   }
   return "other";
@@ -2941,13 +2941,13 @@ function parseOtelMetrics(blocks) {
   return usage;
 }
 var OTEL_ATTR_RE = {
-  body: /body:\s*'([^']*)'/,
-  tool_result_size_bytes: /tool_result_size_bytes:\s*'([^']*)'/,
-  input_tokens: /input_tokens:\s*'([^']*)'/,
-  output_tokens: /output_tokens:\s*'([^']*)'/,
-  cache_read_tokens: /cache_read_tokens:\s*'([^']*)'/,
-  cache_creation_tokens: /cache_creation_tokens:\s*'([^']*)'/,
-  cost_usd: /cost_usd:\s*'([^']*)'/
+  body: /body:\s*['"]([^'"]*)['"]/,
+  tool_result_size_bytes: /tool_result_size_bytes:\s*['"]([^'"]*)['"]/,
+  input_tokens: /input_tokens:\s*['"]([^'"]*)['"]/,
+  output_tokens: /output_tokens:\s*['"]([^'"]*)['"]/,
+  cache_read_tokens: /cache_read_tokens:\s*['"]([^'"]*)['"]/,
+  cache_creation_tokens: /cache_creation_tokens:\s*['"]([^'"]*)['"]/,
+  cost_usd: /cost_usd:\s*['"]([^'"]*)['"]/
 };
 var API_REQUEST_FIELDS = [
   [OTEL_ATTR_RE.input_tokens, "input"],
@@ -4491,31 +4491,7 @@ If NO violations are found:
 }
 `;
 
-// src/gates/review-eval.ts
-var log5 = getCategoryLogger("gate", "review");
-function logDiffStats(diff, mainLogger) {
-  const diffLines = diff.split(`
-`).length;
-  const diffChars = diff.length;
-  const diffEstTokens = Math.ceil(diffChars / CHARS_PER_TOKEN);
-  const diffFileRanges = parseDiff(diff);
-  const diffFiles = diffFileRanges.size;
-  const msg = `[diff-stats] files=${diffFiles} lines=${diffLines} chars=${diffChars} est_tokens=${diffEstTokens}`;
-  log5.debug(msg);
-  mainLogger(`${msg}
-`);
-}
-function logInputStats(prompt, diff, adapterLogger) {
-  const promptChars = prompt.length;
-  const diffChars = diff.length;
-  const totalInputChars = promptChars + diffChars;
-  const promptEstTokens = Math.ceil(promptChars / CHARS_PER_TOKEN);
-  const diffEstTokens = Math.ceil(diffChars / CHARS_PER_TOKEN);
-  const totalEstTokens = promptEstTokens + diffEstTokens;
-  const msg = `[input-stats] prompt_chars=${promptChars} diff_chars=${diffChars} total_chars=${totalInputChars} prompt_est_tokens=${promptEstTokens} diff_est_tokens=${diffEstTokens} total_est_tokens=${totalEstTokens}`;
-  adapterLogger(`${msg}
-`);
-}
+// src/gates/review-prompt.ts
 function buildReviewPrompt(config, previousViolations = []) {
   const baseContent = config.promptContent || "";
   if (previousViolations.length > 0) {
@@ -4533,34 +4509,43 @@ function buildPreviousFailuresSection(violations) {
   const affectedFiles = [...new Set(violations.map((v) => v.file))];
   const lines = [];
   lines.push(buildRerunHeader());
-  if (toVerify.length === 0) {
-    lines.push(`(No violations were marked as FIXED for verification)
+  lines.push(...formatVerifySection(toVerify));
+  lines.push(...formatUnaddressedSection(unaddressed));
+  lines.push(buildRerunInstructions(affectedFiles));
+  return lines.join(`
 `);
-  } else {
-    for (const [i, v] of toVerify.entries()) {
-      lines.push(`${i + 1}. ${v.file}:${v.line} - ${v.issue}`);
-      if (v.fix)
-        lines.push(`   Suggested fix: ${v.fix}`);
-      if (v.result)
-        lines.push(`   Agent result: ${v.result}`);
-      lines.push("");
-    }
+}
+function formatVerifySection(toVerify) {
+  if (toVerify.length === 0) {
+    return [`(No violations were marked as FIXED for verification)
+`];
   }
-  if (unaddressed.length > 0) {
-    lines.push(buildUnaddressedHeader());
-    for (const [i, v] of unaddressed.entries()) {
-      lines.push(`${i + 1}. ${v.file}:${v.line} - ${v.issue}`);
-    }
+  const lines = [];
+  for (const [i, v] of toVerify.entries()) {
+    lines.push(`${i + 1}. ${v.file}:${v.line} - ${v.issue}`);
+    if (v.fix)
+      lines.push(`   Suggested fix: ${v.fix}`);
+    if (v.result)
+      lines.push(`   Agent result: ${v.result}`);
     lines.push("");
   }
-  lines.push(buildRerunInstructions(affectedFiles));
-  return lines.join(`
-`);
+  return lines;
 }
+function formatUnaddressedSection(unaddressed) {
+  if (unaddressed.length === 0)
+    return [];
+  const lines = [buildUnaddressedHeader()];
+  for (const [i, v] of unaddressed.entries()) {
+    lines.push(`${i + 1}. ${v.file}:${v.line} - ${v.issue}`);
+  }
+  lines.push("");
+  return lines;
+}
+var RERUN_SEPARATOR = "━".repeat(46);
 function buildRerunHeader() {
-  return `━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+  return `${RERUN_SEPARATOR}
 RERUN MODE: VERIFY PREVIOUS FIXES ONLY
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
+${RERUN_SEPARATOR}
 
 This is a RERUN review. The agent attempted to fix some of the violations listed below.
 Your task is STRICTLY LIMITED to verifying the fixes for violations marked as FIXED.
@@ -4574,39 +4559,74 @@ The following violations were NOT marked as fixed or skipped and are still activ
 `;
 }
 function buildRerunInstructions(affectedFiles) {
-  return `STRICT INSTRUCTIONS FOR RERUN MODE:
-
-1. VERIFY FIXES: Check if each violation marked as FIXED above has been addressed
-   - For violations that are fixed, confirm they no longer appear
-   - For violations that remain unfixed, include them in your violations array (status: "new")
-
-2. UNADDRESSED VIOLATIONS: You MUST include all UNADDRESSED violations listed above in your output array if they still exist.
-
-3. CHECK FOR REGRESSIONS ONLY: You may ONLY report NEW violations if they:
-   - Are in FILES that were modified to fix the above violations: ${affectedFiles.join(", ")}
-   - Are DIRECTLY caused by the fix changes (e.g., a fix introduced a new bug)
-   - Are in the same function/region that was modified to address a previous violation
-
-4. Return status "pass" ONLY if ALL previous violations (including unaddressed ones) are now fixed AND no regressions were introduced.
-   Otherwise, return status "fail" and list all remaining violations.
-
-━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━`;
+  const files = affectedFiles.join(", ");
+  return [
+    "STRICT INSTRUCTIONS FOR RERUN MODE:",
+    "",
+    "1. VERIFY FIXES: Check if each violation marked as FIXED above has been addressed",
+    "   - For violations that are fixed, confirm they no longer appear",
+    '   - For violations that remain unfixed, include them in your violations array (status: "new")',
+    "",
+    "2. UNADDRESSED VIOLATIONS: You MUST include all UNADDRESSED violations listed above in your output array if they still exist.",
+    "",
+    "3. CHECK FOR REGRESSIONS ONLY: You may ONLY report NEW violations if they:",
+    `   - Are in FILES that were modified to fix the above violations: ${files}`,
+    "   - Are DIRECTLY caused by the fix changes (e.g., a fix introduced a new bug)",
+    "   - Are in the same function/region that was modified to address a previous violation",
+    "",
+    '4. Return status "pass" ONLY if ALL previous violations (including unaddressed ones) are now fixed AND no regressions were introduced.',
+    '   Otherwise, return status "fail" and list all remaining violations.',
+    "",
+    RERUN_SEPARATOR
+  ].join(`
+`);
 }
+// src/gates/review-eval.ts
+var log5 = getCategoryLogger("gate", "review");
+function logDiffStats(diff, mainLogger) {
+  const diffLines = diff.split(`
+`).length;
+  const diffChars = diff.length;
+  const diffEstTokens = Math.ceil(diffChars / CHARS_PER_TOKEN);
+  const diffFileRanges = parseDiff(diff);
+  const diffFiles = diffFileRanges.size;
+  const msg = `[diff-stats] files=${diffFiles} lines=${diffLines} chars=${diffChars} est_tokens=${diffEstTokens}`;
+  log5.debug(msg);
+  mainLogger(`${msg}
+`);
+}
+function logInputStats(prompt, diff, adapterLogger) {
+  const promptChars = prompt.length;
+  const diffChars = diff.length;
+  const totalInputChars = promptChars + diffChars;
+  const promptEstTokens = Math.ceil(promptChars / CHARS_PER_TOKEN);
+  const diffEstTokens = Math.ceil(diffChars / CHARS_PER_TOKEN);
+  const totalEstTokens = promptEstTokens + diffEstTokens;
+  const msg = `[input-stats] prompt_chars=${promptChars} diff_chars=${diffChars} total_chars=${totalInputChars} prompt_est_tokens=${promptEstTokens} diff_est_tokens=${diffEstTokens} total_est_tokens=${totalEstTokens}`;
+  adapterLogger(`${msg}
+`);
+}
+var MAX_OUTPUT_SIZE_FOR_JSON_PROBE = 1e5;
 function evaluateOutput(output, diff) {
   const diffRanges = diff ? parseDiff(diff) : undefined;
   try {
     const fromBlock = tryParseJsonBlock(output);
     if (fromBlock)
       return validateAndReturn(fromBlock, diffRanges);
-    const fromLast = tryParseLastJson(output);
-    if (fromLast)
-      return validateAndReturn(fromLast, diffRanges);
-    const fromFirst = tryParseFirstJson(output);
-    if (fromFirst)
-      return validateAndReturn(fromFirst, diffRanges);
+    const fromDirect = tryParseWholeOutput(output);
+    if (fromDirect)
+      return validateAndReturn(fromDirect, diffRanges);
+    if (output.length <= MAX_OUTPUT_SIZE_FOR_JSON_PROBE) {
+      const fromLast = tryParseLastJson(output);
+      if (fromLast)
+        return validateAndReturn(fromLast, diffRanges);
+      const fromFirst = tryParseFirstJson(output);
+      if (fromFirst)
+        return validateAndReturn(fromFirst, diffRanges);
+    }
     return {
       status: "error",
-      message: "No valid JSON object found in output"
+      message: output.length > MAX_OUTPUT_SIZE_FOR_JSON_PROBE ? `Output too large (${output.length} bytes) and no JSON found` : "No valid JSON object found in output"
     };
   } catch (error) {
     const err = error;
@@ -4626,12 +4646,26 @@ function tryParseJsonBlock(output) {
     return null;
   }
 }
+function tryParseWholeOutput(output) {
+  const trimmed = output.trim();
+  if (!(trimmed.startsWith("{") && trimmed.endsWith("}")))
+    return null;
+  try {
+    const json = JSON.parse(trimmed);
+    if (json.status)
+      return json;
+  } catch {}
+  return null;
+}
+var MAX_JSON_PROBE_ITERATIONS = 50;
 function tryParseLastJson(output) {
   const end = output.lastIndexOf("}");
   if (end === -1)
     return null;
   let start = output.lastIndexOf("{", end);
-  while (start !== -1) {
+  let iterations = 0;
+  while (start !== -1 && iterations < MAX_JSON_PROBE_ITERATIONS) {
+    iterations++;
     try {
       const json = JSON.parse(output.substring(start, end + 1));
       if (json.status)
@@ -6122,6 +6156,7 @@ ${chalk2.bold(SEPARATOR)}`);
     }
     const { overallStatus, statusColor } = computeOverallStatus(results, statusOverride);
     console.error(statusColor(`Status: ${overallStatus}`));
+    console.log(`Status: ${overallStatus}`);
     console.error(chalk2.bold(`${SEPARATOR}
 `));
   }
@@ -8993,6 +9028,17 @@ async function handleNoChanges(ctx, failuresMap) {
     log10.info(getStatusMessage2(status));
     return { status, message: getStatusMessage2(status), gatesRun: 0 };
   }
+  if (failuresMap && failuresMap.size > 0) {
+    let totalViolations = 0;
+    for (const adapterMap of failuresMap.values()) {
+      for (const violations of adapterMap.values()) {
+        totalViolations += violations.length;
+      }
+    }
+    const message = `No changes detected — ${totalViolations} violation(s) still outstanding.`;
+    log10.warn(message);
+    return { status: "failed", message, gatesRun: 0 };
+  }
   log10.info("No changes detected.");
   return {
     status: "no_changes",
@@ -9577,4 +9623,4 @@ if (process.argv.length < 3) {
 }
 program.parse(process.argv);
 
-//# debugId=CD4FBA492EA1ADA764756E2164756E21
+//# debugId=5FCEF2943643F3B864756E2164756E21