@link-assistant/hive-mind 2.9.0 → 2.9.1

package/CHANGELOG.md

@@ -1,5 +1,11 @@
 # @link-assistant/hive-mind
 
+## 2.9.1
+
+### Patch Changes
+
+- 3e48cb3: Resume Codex sessions after transient high-demand failures and preserve uncommitted work before uploading failure logs.
+
 ## 2.9.0
 
 ### Minor Changes

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@link-assistant/hive-mind",
-  "version": "2.9.0",
+  "version": "2.9.1",
   "description": "AI-powered issue solver and hive mind for collaborative problem solving",
   "main": "src/hive.mjs",
   "type": "module",

package/src/solve.mjs

@@ -1117,6 +1117,17 @@ try {
       await log('');
     }
 
+    // Preserve work before remote diagnostics; issue #2101 ended during log upload.
+    try {
+      const { criticalErrorRecovery } = await import('./config.lib.mjs');
+      if (criticalErrorRecovery.autoCommitUncommittedChanges) {
+        const { commitUncommittedChangesOnCriticalError } = await import('./critical-error-commit.lib.mjs');
+        await commitUncommittedChangesOnCriticalError({ tempDir, branchName, $, log, reason: toolFailureMessage });
+      }
+    } catch (preserveError) {
+      await log(`  ⚠️  Could not auto-commit before failure exit: ${preserveError.message}`, { verbose: true });
+    }
+
     // Attach failure logs before exiting (Issues #1212, #1462: fall back to issue if no PR)
     const hasPR = global.createdPR && global.createdPR.number;
     const hasIssue = global.issueNumber;
@@ -1167,19 +1178,6 @@ try {
       }
     }
 
-    // Issue #1834 (PR #1835 feedback): "on all critical errors we auto commit uncommitted changes by
-    // default." A failed session exits here before the normal auto-commit chokepoint below, so commit
-    // + push any work first. On by default; disable via HIVE_MIND_AUTO_COMMIT_ON_CRITICAL_ERROR=false.
-    try {
-      const { criticalErrorRecovery } = await import('./config.lib.mjs');
-      if (criticalErrorRecovery.autoCommitUncommittedChanges) {
-        const { commitUncommittedChangesOnCriticalError } = await import('./critical-error-commit.lib.mjs');
-        await commitUncommittedChangesOnCriticalError({ tempDir, branchName, $, log, reason: toolFailureMessage });
-      }
-    } catch (preserveError) {
-      await log(`  ⚠️  Could not auto-commit before failure exit: ${preserveError.message}`, { verbose: true });
-    }
-
     await safeExit(1, toolFailureMessage);
   }
 

package/src/tool-retry.lib.mjs

@@ -47,7 +47,11 @@ export const classifyRetryableError = value => {
   // overloaded API anyway. Claude Code already exposes its own per-request fallback
   // via `--fallback-model` (wired in claude.lib.mjs), so we keep `--model` stable and
   // simply retry. Therefore isCapacity is false → retry with the same model.
-  if (lower.includes('overloaded') || lower.includes('overloaded_error')) {
+  // Issue #2101: Codex may exhaust its internal WebSocket/HTTPS attempts and
+  // surface only "We're currently experiencing high demand, which may cause
+  // temporary errors." The terminal message omits both the preceding 503 and
+  // the backend's concurrency_limit code, so it must be recognized directly.
+  if (lower.includes('overloaded') || lower.includes('overloaded_error') || lower.includes('currently experiencing high demand') || lower.includes('too many concurrent requests') || lower.includes('concurrency_limit')) {
     return { message, isRetryable: true, isCapacity: false, label: 'API overload' };
   }