@link-assistant/hive-mind 1.24.4 → 1.24.6

package/CHANGELOG.md

@@ -1,5 +1,31 @@
 # @link-assistant/hive-mind
 
+## 1.24.6
+
+### Patch Changes
+
+- Make `--auto-resume-on-limit-reset` enabled by default to improve user experience when hitting API rate limits. Previously defaulted to `false`, now defaults to `true` for both `solve` and `hive` commands. Users can explicitly disable with `--no-auto-resume-on-limit-reset` if needed.
+
+  Fix false positive error detection for step_finish with reason stop
+
+  When an agent encounters a timeout error during execution but successfully recovers and completes (indicated by `step_finish` with `reason: "stop"`), the error detection was incorrectly flagging this as a failure due to fallback pattern matching.
+
+  The `agentCompletedSuccessfully` flag was only being set for `session.idle` and `"exiting loop"` log messages (Issue #1276), but not for the more common `step_finish` event with `reason: "stop"`. This meant the fallback pattern matching would still trigger and detect error patterns in the full output, even when the agent had clearly completed successfully.
+
+  Fix: Add `step_finish` with `reason: "stop"` as a success marker in both stdout and stderr processing loops in `src/agent.lib.mjs`.
+
+## 1.24.5
+
+### Patch Changes
+
+- 17317bb: fix: prevent false positive error detection for JSON-structured stderr warnings (Issue #1337)
+
+  Claude Code SDK can emit structured JSON log messages to stderr with format `{"level":"warn","message":"..."}`. When these messages contained error-related keywords like "failed", the detection logic incorrectly flagged them as errors.
+
+  Added JSON parsing for stderr messages starting with `{`. If the parsed JSON has a `level` field that is not `"error"` or `"fatal"`, the message is treated as a warning (non-error), preserving existing emoji-prefix detection as a fallback.
+
+  Also enables `ANTHROPIC_LOG=debug` when running with `--verbose` flag, allowing users to see detailed API request information as suggested by the BashTool pre-flight warning.
+
 ## 1.24.4
 
 ### Patch Changes

package/README.md

@@ -268,12 +268,10 @@ See [docs/HELM.md](./docs/HELM.md) for detailed Helm configuration options.
        --attach-logs
        --verbose
        --no-tool-check
-       --auto-resume-on-limit-reset
      TELEGRAM_SOLVE_OVERRIDES:
        --attach-logs
        --verbose
        --no-tool-check
-       --auto-resume-on-limit-reset
      TELEGRAM_BOT_VERBOSE: true
    "
 
@@ -295,12 +293,10 @@ See [docs/HELM.md](./docs/HELM.md) for detailed Helm configuration options.
      --attach-logs
      --verbose
      --no-tool-check
-     --auto-resume-on-limit-reset
    )" --solve-overrides "(
      --attach-logs
      --verbose
      --no-tool-check
-     --auto-resume-on-limit-reset
    )" --verbose
 
    # Press CTRL + A + D for detach from screen

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@link-assistant/hive-mind",
-  "version": "1.24.4",
+  "version": "1.24.6",
   "description": "AI-powered issue solver and hive mind for collaborative problem solving",
   "main": "src/hive.mjs",
   "type": "module",

package/src/agent.lib.mjs

@@ -665,6 +665,13 @@ export const executeAgentCommand = async params => {
               if (data.type === 'session.idle' || (data.type === 'log' && data.message === 'exiting loop')) {
                 agentCompletedSuccessfully = true;
               }
+              // Issue #1296: Detect step_finish with reason "stop" as successful completion
+              // This is a clear marker of success - agent finished normally, not due to error or limit
+              // When this event appears, we should ignore any error events that appeared earlier in the stream
+              // (e.g., timeout errors that were recovered from via retry logic)
+              if (data.type === 'step_finish' && data.part?.reason === 'stop') {
+                agentCompletedSuccessfully = true;
+              }
             } catch {
               // Not JSON - log as plain text
               await log(line);
@@ -727,6 +734,11 @@ export const executeAgentCommand = async params => {
                 if (stderrData.type === 'session.idle' || (stderrData.type === 'log' && stderrData.message === 'exiting loop')) {
                   agentCompletedSuccessfully = true;
                 }
+                // Issue #1296: Detect step_finish with reason "stop" as successful completion (stderr)
+                // This is a clear marker of success - agent finished normally, not due to error or limit
+                if (stderrData.type === 'step_finish' && stderrData.part?.reason === 'stop') {
+                  agentCompletedSuccessfully = true;
+                }
               } catch {
                 // Not JSON - log as plain text
                 await log(stderrLine);

package/src/claude.lib.mjs

@@ -724,6 +724,48 @@ export const calculateSessionTokens = async (sessionId, tempDir) => {
     throw new Error(`Failed to read session file: ${readError.message}`);
   }
 };
+/**
+ * Determines whether a stderr message line should be treated as an error.
+ *
+ * Excludes:
+ * - Emoji-prefixed warnings (Issue #477): lines starting with ⚠️ or ⚠
+ * - JSON-structured log messages with non-error level (Issue #1337):
+ *   e.g. {"level":"warn","message":"...failed..."} — the word "failed" is in
+ *   the message text but the level is "warn", so it is NOT an error.
+ *   Only JSON lines with level "error" or "fatal" are treated as real errors.
+ *
+ * @param {string} message - A single trimmed stderr line
+ * @returns {boolean} true if the line should count as an error
+ */
+export const isStderrError = message => {
+  const trimmed = message.trim();
+  if (!trimmed) return false;
+
+  // Detection 1: Emoji-prefixed warnings (Issue #477)
+  let isWarning = trimmed.startsWith('⚠️') || trimmed.startsWith('⚠');
+
+  // Detection 2: JSON-structured log messages (Issue #1337)
+  if (!isWarning && trimmed.startsWith('{')) {
+    try {
+      const parsed = JSON.parse(trimmed);
+      if (parsed && typeof parsed.level === 'string') {
+        const level = parsed.level.toLowerCase();
+        // Only "error" and "fatal" levels are real errors.
+        if (level !== 'error' && level !== 'fatal') {
+          isWarning = true;
+        }
+      }
+    } catch {
+      // Not valid JSON — fall through to keyword matching
+    }
+  }
+
+  if (!isWarning && (trimmed.includes('Error:') || trimmed.includes('error') || trimmed.includes('failed') || trimmed.includes('not found'))) {
+    return true;
+  }
+  return false;
+};
+
 export const executeClaudeCommand = async params => {
   const {
     tempDir,
@@ -847,10 +889,17 @@ export const executeClaudeCommand = async params => {
       // Pass model for model-specific max output tokens (Issue #1221)
       // Pass thinkLevel and maxBudget for Opus 4.6 effort level conversion (Issue #1238)
       const claudeEnv = getClaudeEnv({ thinkingBudget: resolvedThinkingBudget, model: mappedModel, thinkLevel, maxBudget });
+      // Issue #1337: Enable ANTHROPIC_LOG=debug in --verbose mode to diagnose slow API requests.
+      // The BashTool pre-flight check suggests "Run with ANTHROPIC_LOG=debug to check for failed or slow API requests."
+      // When --verbose is enabled, we propagate ANTHROPIC_LOG=debug so users can see detailed API request info.
+      if (argv.verbose) {
+        claudeEnv.ANTHROPIC_LOG = 'debug';
+      }
       const modelMaxOutputTokens = getMaxOutputTokensForModel(mappedModel);
       if (argv.verbose) await log(`📊 CLAUDE_CODE_MAX_OUTPUT_TOKENS: ${modelMaxOutputTokens}`, { verbose: true });
       if (argv.verbose) await log(`📊 MCP_TIMEOUT: ${claudeCode.mcpTimeout}ms (server startup)`, { verbose: true });
       if (argv.verbose) await log(`📊 MCP_TOOL_TIMEOUT: ${claudeCode.mcpToolTimeout}ms (tool execution)`, { verbose: true });
+      if (argv.verbose) await log(`📊 ANTHROPIC_LOG: debug (verbose mode)`, { verbose: true });
       if (resolvedThinkingBudget !== undefined) await log(`📊 MAX_THINKING_TOKENS: ${resolvedThinkingBudget}`, { verbose: true });
       if (claudeEnv.CLAUDE_CODE_EFFORT_LEVEL) await log(`📊 CLAUDE_CODE_EFFORT_LEVEL: ${claudeEnv.CLAUDE_CODE_EFFORT_LEVEL}`, { verbose: true });
       if (!isNewVersion && thinkLevel) await log(`📊 Thinking level (via keywords): ${thinkLevel}`, { verbose: true });
@@ -1065,16 +1114,9 @@ export const executeClaudeCommand = async params => {
           // Log stderr immediately
           if (errorOutput) {
             await log(errorOutput, { stream: 'stderr' });
-            // Track stderr errors for failure detection
-            const trimmed = errorOutput.trim();
-            // Exclude warnings (messages starting with ⚠️) from being treated as errors
-            // Example: "⚠️  [BashTool] Pre-flight check is taking longer than expected. Run with ANTHROPIC_LOG=debug to check for failed or slow API requests."
-            // Even though this contains the word "failed", it's a warning, not an error
-            const isWarning = trimmed.startsWith('⚠️') || trimmed.startsWith('⚠');
-            // Issue #1165: Also detect "command not found" errors (e.g., "/bin/sh: 1: claude: not found")
-            // These indicate the Claude CLI is not installed or not in PATH
-            if (trimmed && !isWarning && (trimmed.includes('Error:') || trimmed.includes('error') || trimmed.includes('failed') || trimmed.includes('not found'))) {
-              stderrErrors.push(trimmed);
+            // Track stderr errors for failure detection using shared helper (Issue #477, #1165, #1337)
+            if (isStderrError(errorOutput)) {
+              stderrErrors.push(errorOutput.trim());
             }
           }
         } else if (chunk.type === 'exit') {

package/src/hive.config.lib.mjs

@@ -36,7 +36,7 @@ const HIVE_CUSTOM_SOLVE_OPTIONS = {
   'auto-resume-on-limit-reset': {
     type: 'boolean',
     description: 'Automatically resume when AI tool limit resets (calculates reset time and waits). Passed to solve command.',
-    default: false,
+    default: true,
   },
   'auto-cleanup': {
     type: 'boolean',

package/src/solve.config.lib.mjs

@@ -132,7 +132,7 @@ export const SOLVE_OPTION_DEFINITIONS = {
   'auto-resume-on-limit-reset': {
     type: 'boolean',
     description: 'Automatically resume when AI tool limit resets (maintains session context with --resume flag)',
-    default: false,
+    default: true,
   },
   'auto-restart-on-limit-reset': {
     type: 'boolean',