groove-dev 0.27.35 → 0.27.37

package/node_modules/@groove-dev/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@groove-dev/cli",
-  "version": "0.27.35",
+  "version": "0.27.37",
   "description": "GROOVE CLI — manage AI coding agents from your terminal",
   "license": "FSL-1.1-Apache-2.0",
   "type": "module",

package/node_modules/@groove-dev/daemon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@groove-dev/daemon",
-  "version": "0.27.35",
+  "version": "0.27.37",
   "description": "GROOVE daemon — agent orchestration engine",
   "license": "FSL-1.1-Apache-2.0",
   "type": "module",

package/node_modules/@groove-dev/daemon/src/api.js

@@ -2206,10 +2206,19 @@ Keep responses concise. Help them think, don't lecture them about the system the
       const entries = [];
 
       // Dirs first (sorted), then files (sorted)
-      const dirs = raw.filter((e) => e.isDirectory() && !IGNORED_NAMES.has(e.name) && !e.name.startsWith('.'))
-        .sort((a, b) => a.name.localeCompare(b.name));
-      const files = raw.filter((e) => e.isFile() && !e.name.startsWith('.'))
+      const isDir = (e) => {
+        if (e.isDirectory()) return true;
+        if (e.isSymbolicLink()) { try { return statSync(resolve(fullPath, e.name)).isDirectory(); } catch { return false; } }
+        return false;
+      };
+      const dirs = raw.filter((e) => isDir(e) && !IGNORED_NAMES.has(e.name) && !e.name.startsWith('.'))
         .sort((a, b) => a.name.localeCompare(b.name));
+      const files = raw.filter((e) => {
+        if (e.name.startsWith('.')) return false;
+        if (e.isFile()) return true;
+        if (e.isSymbolicLink()) { try { return statSync(resolve(fullPath, e.name)).isFile(); } catch { return false; } }
+        return false;
+      }).sort((a, b) => a.name.localeCompare(b.name));
 
       for (const d of dirs) {
         const childPath = relPath ? `${relPath}/${d.name}` : d.name;

package/node_modules/@groove-dev/daemon/src/process.js

@@ -16,6 +16,8 @@ const COMPACTION_DROP_THRESHOLD = 0.25; // 25% drop from peak = natural compacti
 const COMPACTION_MIN_PEAK = 0.15;       // Ignore compaction if peak was below 15%
 const MAX_BUFFER_SIZE = 1_048_576;      // 1MB — discard oldest half if exceeded
 const STREAM_THROTTLE_MS = 250;         // 4 broadcasts/sec per agent
+const STALL_CHECK_INTERVAL_MS = 60_000; // Poll for stuck agents every 60s
+const STALL_THRESHOLD_MS = 5 * 60_000;  // 5 min of silence on a live PID = stalled stream
 
 // Role-specific prompt prefixes — applied during spawn regardless of entry point
 // (SpawnPanel, chat continue, CLI, API) for consistency
@@ -297,7 +299,44 @@ export class ProcessManager {
     this.pendingMessages = new Map(); // agentId -> { message, timestamp }
     this._streamThrottle = new Map(); // agentId -> { timer, pending }
     this._rotatingAgents = new Set(); // agentIds currently being rotated (rotator wrote handoff)
+    this._stalledAgents = new Set(); // agentIds already flagged as stalled (avoids duplicate broadcasts)
 
+    this._stallWatchdog = setInterval(() => this._checkStalls(), STALL_CHECK_INTERVAL_MS);
+    if (this._stallWatchdog.unref) this._stallWatchdog.unref();
+  }
+
+  _checkStalls() {
+    const { registry } = this.daemon;
+    const now = Date.now();
+    for (const agentId of this.handles.keys()) {
+      const agent = registry.get(agentId);
+      if (!agent || agent.status !== 'running') continue;
+      const lastActivity = agent.lastActivity ? new Date(agent.lastActivity).getTime() : now;
+      const silentMs = now - lastActivity;
+      if (silentMs < STALL_THRESHOLD_MS) {
+        if (this._stalledAgents.has(agentId)) {
+          this._stalledAgents.delete(agentId);
+          registry.update(agentId, { stalled: false });
+        }
+        continue;
+      }
+      if (this._stalledAgents.has(agentId)) continue;
+      this._stalledAgents.add(agentId);
+      registry.update(agentId, { stalled: true, stalledSince: new Date(lastActivity).toISOString() });
+      if (this.daemon.timeline) {
+        this.daemon.timeline.recordEvent('stall', {
+          agentId, agentName: agent.name, role: agent.role, silentMs,
+        });
+      }
+      this.daemon.broadcast({
+        type: 'agent:stalled',
+        agentId,
+        agentName: agent.name,
+        silentMs,
+        lastActivity: agent.lastActivity,
+      });
+      console.warn(`[Groove] Agent ${agent.name} (${agentId}) silent for ${Math.round(silentMs / 1000)}s — possible stalled API stream`);
+    }
   }
 
   async spawn(config) {
@@ -746,6 +785,7 @@ For normal file edits within your scope, proceed without review.
       // Clean up per-agent maps to prevent unbounded growth in long sessions
       this.peakContextUsage.delete(agent.id);
       this.pendingMessages.delete(agent.id);
+      this._stalledAgents.delete(agent.id);
 
       // Release file-scope locks so they don't persist after agent death
       if (this.daemon.locks) this.daemon.locks.release(agent.id);
@@ -917,6 +957,12 @@ For normal file edits within your scope, proceed without review.
 
     const updates = { lastActivity: new Date().toISOString() };
 
+    // Clear stall flag — output means the stream is alive again
+    if (this._stalledAgents.has(agentId)) {
+      this._stalledAgents.delete(agentId);
+      updates.stalled = false;
+    }
+
     // Token tracking — feed subsystems with full breakdown
     if (output.tokensUsed !== undefined && output.tokensUsed > 0) {
       updates.tokensUsed = agent.tokensUsed + output.tokensUsed;
@@ -1398,6 +1444,7 @@ For normal file edits within your scope, proceed without review.
       logStream.write(`[${new Date().toISOString()}] Process exited: code=${code} signal=${signal}\n`);
       logStream.end();
       this.handles.delete(newAgent.id);
+      this._stalledAgents.delete(newAgent.id);
       const finalStatus = signal === 'SIGTERM' || signal === 'SIGKILL' ? 'killed' : code === 0 ? 'completed' : 'crashed';
       registry.update(newAgent.id, { status: finalStatus, pid: null });
       this.daemon.broadcast({ type: 'agent:exit', agentId: newAgent.id, code, signal, status: finalStatus });
@@ -1410,6 +1457,7 @@ For normal file edits within your scope, proceed without review.
       logStream.write(`[error] ${err.message}\n`);
       logStream.end();
       this.handles.delete(newAgent.id);
+      this._stalledAgents.delete(newAgent.id);
       registry.update(newAgent.id, { status: 'crashed', pid: null });
     });
 
@@ -1548,6 +1596,10 @@ For normal file edits within your scope, proceed without review.
   async killAll() {
     const ids = Array.from(this.handles.keys());
     await Promise.all(ids.map((id) => this.kill(id)));
+    if (this._stallWatchdog) {
+      clearInterval(this._stallWatchdog);
+      this._stallWatchdog = null;
+    }
   }
 
   isRunning(agentId) {

package/node_modules/@groove-dev/daemon/src/providers/claude-code.js

@@ -149,19 +149,22 @@ export class ClaudeCodeProvider extends Provider {
           const cacheReadTokens = usage?.cache_read_input_tokens || 0;
           const cacheCreationTokens = usage?.cache_creation_input_tokens || 0;
           const outputTokens = usage?.output_tokens || 0;
+          // tokensUsed = new processing tokens only (input + output). Cache reads are
+          // the same bytes re-read every turn and must NOT be accumulated — doing so
+          // inflated agent.tokensUsed ~50× and created the phantom "freeze at 1M".
+          // totalIn still drives contextUsage because cached bytes DO occupy context.
           const totalIn = inputTokens + cacheReadTokens + cacheCreationTokens;
           events.push({
             type: 'activity',
             subtype: 'assistant',
             data: data.message?.content || '',
-            tokensUsed: totalIn + outputTokens,
+            tokensUsed: inputTokens + outputTokens,
             inputTokens,
             outputTokens,
             cacheReadTokens,
             cacheCreationTokens,
             model: data.message?.model,
           });
-          // Compute context usage from assistant message usage
           if (totalIn > 0) {
             const modelId = data.message?.model || '';
             const modelMeta = ClaudeCodeProvider.models.find((m) => modelId.includes(m.id));
@@ -172,21 +175,12 @@ export class ClaudeCodeProvider extends Provider {
             });
           }
         } else if (data.type === 'result') {
-          // Result has cumulative usage for the full session
-          const usage = data.usage;
-          const inputTokens = usage?.input_tokens || 0;
-          const cacheReadTokens = usage?.cache_read_input_tokens || 0;
-          const cacheCreationTokens = usage?.cache_creation_input_tokens || 0;
-          const outputTokens = usage?.output_tokens || 0;
-          const totalIn = inputTokens + cacheReadTokens + cacheCreationTokens;
+          // Result carries cumulative session usage — per-turn counts were already
+          // accumulated from assistant events, so we do NOT emit tokensUsed here
+          // (that was the double-count). Only emit session-level metadata.
           events.push({
             type: 'result',
             data: data.result,
-            tokensUsed: totalIn + outputTokens,
-            inputTokens,
-            outputTokens,
-            cacheReadTokens,
-            cacheCreationTokens,
             cost: data.total_cost_usd,
             duration: data.duration_ms,
             turns: data.num_turns,

package/node_modules/@groove-dev/gui/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@groove-dev/gui",
-  "version": "0.27.35",
+  "version": "0.27.37",
   "description": "GROOVE GUI — visual agent control plane",
   "license": "FSL-1.1-Apache-2.0",
   "type": "module",

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "groove-dev",
-  "version": "0.27.35",
+  "version": "0.27.37",
   "description": "Open-source agent orchestration layer — the AI company OS. Local model agent engine (GGUF/Ollama/llama-server), HuggingFace model browser, MCP integrations (Slack, Gmail, Stripe, 15+), agent scheduling (cron), business roles (CMO, CFO, EA). GUI dashboard, multi-agent coordination, zero cold-start, infinite sessions. Works with Claude Code, Codex, Gemini CLI, Ollama, any local model.",
   "license": "FSL-1.1-Apache-2.0",
   "author": "Groove Dev <hello@groovedev.ai> (https://groovedev.ai)",

package/packages/cli/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@groove-dev/cli",
-  "version": "0.27.35",
+  "version": "0.27.37",
   "description": "GROOVE CLI — manage AI coding agents from your terminal",
   "license": "FSL-1.1-Apache-2.0",
   "type": "module",

package/packages/daemon/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@groove-dev/daemon",
-  "version": "0.27.35",
+  "version": "0.27.37",
   "description": "GROOVE daemon — agent orchestration engine",
   "license": "FSL-1.1-Apache-2.0",
   "type": "module",

package/packages/daemon/src/api.js

@@ -2206,10 +2206,19 @@ Keep responses concise. Help them think, don't lecture them about the system the
       const entries = [];
 
       // Dirs first (sorted), then files (sorted)
-      const dirs = raw.filter((e) => e.isDirectory() && !IGNORED_NAMES.has(e.name) && !e.name.startsWith('.'))
-        .sort((a, b) => a.name.localeCompare(b.name));
-      const files = raw.filter((e) => e.isFile() && !e.name.startsWith('.'))
+      const isDir = (e) => {
+        if (e.isDirectory()) return true;
+        if (e.isSymbolicLink()) { try { return statSync(resolve(fullPath, e.name)).isDirectory(); } catch { return false; } }
+        return false;
+      };
+      const dirs = raw.filter((e) => isDir(e) && !IGNORED_NAMES.has(e.name) && !e.name.startsWith('.'))
         .sort((a, b) => a.name.localeCompare(b.name));
+      const files = raw.filter((e) => {
+        if (e.name.startsWith('.')) return false;
+        if (e.isFile()) return true;
+        if (e.isSymbolicLink()) { try { return statSync(resolve(fullPath, e.name)).isFile(); } catch { return false; } }
+        return false;
+      }).sort((a, b) => a.name.localeCompare(b.name));
 
       for (const d of dirs) {
         const childPath = relPath ? `${relPath}/${d.name}` : d.name;

package/packages/daemon/src/process.js

@@ -16,6 +16,8 @@ const COMPACTION_DROP_THRESHOLD = 0.25; // 25% drop from peak = natural compacti
 const COMPACTION_MIN_PEAK = 0.15;       // Ignore compaction if peak was below 15%
 const MAX_BUFFER_SIZE = 1_048_576;      // 1MB — discard oldest half if exceeded
 const STREAM_THROTTLE_MS = 250;         // 4 broadcasts/sec per agent
+const STALL_CHECK_INTERVAL_MS = 60_000; // Poll for stuck agents every 60s
+const STALL_THRESHOLD_MS = 5 * 60_000;  // 5 min of silence on a live PID = stalled stream
 
 // Role-specific prompt prefixes — applied during spawn regardless of entry point
 // (SpawnPanel, chat continue, CLI, API) for consistency
@@ -297,7 +299,44 @@ export class ProcessManager {
     this.pendingMessages = new Map(); // agentId -> { message, timestamp }
     this._streamThrottle = new Map(); // agentId -> { timer, pending }
     this._rotatingAgents = new Set(); // agentIds currently being rotated (rotator wrote handoff)
+    this._stalledAgents = new Set(); // agentIds already flagged as stalled (avoids duplicate broadcasts)
 
+    this._stallWatchdog = setInterval(() => this._checkStalls(), STALL_CHECK_INTERVAL_MS);
+    if (this._stallWatchdog.unref) this._stallWatchdog.unref();
+  }
+
+  _checkStalls() {
+    const { registry } = this.daemon;
+    const now = Date.now();
+    for (const agentId of this.handles.keys()) {
+      const agent = registry.get(agentId);
+      if (!agent || agent.status !== 'running') continue;
+      const lastActivity = agent.lastActivity ? new Date(agent.lastActivity).getTime() : now;
+      const silentMs = now - lastActivity;
+      if (silentMs < STALL_THRESHOLD_MS) {
+        if (this._stalledAgents.has(agentId)) {
+          this._stalledAgents.delete(agentId);
+          registry.update(agentId, { stalled: false });
+        }
+        continue;
+      }
+      if (this._stalledAgents.has(agentId)) continue;
+      this._stalledAgents.add(agentId);
+      registry.update(agentId, { stalled: true, stalledSince: new Date(lastActivity).toISOString() });
+      if (this.daemon.timeline) {
+        this.daemon.timeline.recordEvent('stall', {
+          agentId, agentName: agent.name, role: agent.role, silentMs,
+        });
+      }
+      this.daemon.broadcast({
+        type: 'agent:stalled',
+        agentId,
+        agentName: agent.name,
+        silentMs,
+        lastActivity: agent.lastActivity,
+      });
+      console.warn(`[Groove] Agent ${agent.name} (${agentId}) silent for ${Math.round(silentMs / 1000)}s — possible stalled API stream`);
+    }
   }
 
   async spawn(config) {
@@ -746,6 +785,7 @@ For normal file edits within your scope, proceed without review.
       // Clean up per-agent maps to prevent unbounded growth in long sessions
       this.peakContextUsage.delete(agent.id);
       this.pendingMessages.delete(agent.id);
+      this._stalledAgents.delete(agent.id);
 
       // Release file-scope locks so they don't persist after agent death
       if (this.daemon.locks) this.daemon.locks.release(agent.id);
@@ -917,6 +957,12 @@ For normal file edits within your scope, proceed without review.
 
     const updates = { lastActivity: new Date().toISOString() };
 
+    // Clear stall flag — output means the stream is alive again
+    if (this._stalledAgents.has(agentId)) {
+      this._stalledAgents.delete(agentId);
+      updates.stalled = false;
+    }
+
     // Token tracking — feed subsystems with full breakdown
     if (output.tokensUsed !== undefined && output.tokensUsed > 0) {
       updates.tokensUsed = agent.tokensUsed + output.tokensUsed;
@@ -1398,6 +1444,7 @@ For normal file edits within your scope, proceed without review.
       logStream.write(`[${new Date().toISOString()}] Process exited: code=${code} signal=${signal}\n`);
       logStream.end();
       this.handles.delete(newAgent.id);
+      this._stalledAgents.delete(newAgent.id);
       const finalStatus = signal === 'SIGTERM' || signal === 'SIGKILL' ? 'killed' : code === 0 ? 'completed' : 'crashed';
       registry.update(newAgent.id, { status: finalStatus, pid: null });
       this.daemon.broadcast({ type: 'agent:exit', agentId: newAgent.id, code, signal, status: finalStatus });
@@ -1410,6 +1457,7 @@ For normal file edits within your scope, proceed without review.
       logStream.write(`[error] ${err.message}\n`);
       logStream.end();
       this.handles.delete(newAgent.id);
+      this._stalledAgents.delete(newAgent.id);
       registry.update(newAgent.id, { status: 'crashed', pid: null });
     });
 
@@ -1548,6 +1596,10 @@ For normal file edits within your scope, proceed without review.
   async killAll() {
     const ids = Array.from(this.handles.keys());
     await Promise.all(ids.map((id) => this.kill(id)));
+    if (this._stallWatchdog) {
+      clearInterval(this._stallWatchdog);
+      this._stallWatchdog = null;
+    }
   }
 
   isRunning(agentId) {

package/packages/daemon/src/providers/claude-code.js

@@ -149,19 +149,22 @@ export class ClaudeCodeProvider extends Provider {
           const cacheReadTokens = usage?.cache_read_input_tokens || 0;
           const cacheCreationTokens = usage?.cache_creation_input_tokens || 0;
           const outputTokens = usage?.output_tokens || 0;
+          // tokensUsed = new processing tokens only (input + output). Cache reads are
+          // the same bytes re-read every turn and must NOT be accumulated — doing so
+          // inflated agent.tokensUsed ~50× and created the phantom "freeze at 1M".
+          // totalIn still drives contextUsage because cached bytes DO occupy context.
           const totalIn = inputTokens + cacheReadTokens + cacheCreationTokens;
           events.push({
             type: 'activity',
             subtype: 'assistant',
             data: data.message?.content || '',
-            tokensUsed: totalIn + outputTokens,
+            tokensUsed: inputTokens + outputTokens,
             inputTokens,
             outputTokens,
             cacheReadTokens,
             cacheCreationTokens,
             model: data.message?.model,
           });
-          // Compute context usage from assistant message usage
           if (totalIn > 0) {
             const modelId = data.message?.model || '';
             const modelMeta = ClaudeCodeProvider.models.find((m) => modelId.includes(m.id));
@@ -172,21 +175,12 @@ export class ClaudeCodeProvider extends Provider {
             });
           }
         } else if (data.type === 'result') {
-          // Result has cumulative usage for the full session
-          const usage = data.usage;
-          const inputTokens = usage?.input_tokens || 0;
-          const cacheReadTokens = usage?.cache_read_input_tokens || 0;
-          const cacheCreationTokens = usage?.cache_creation_input_tokens || 0;
-          const outputTokens = usage?.output_tokens || 0;
-          const totalIn = inputTokens + cacheReadTokens + cacheCreationTokens;
+          // Result carries cumulative session usage — per-turn counts were already
+          // accumulated from assistant events, so we do NOT emit tokensUsed here
+          // (that was the double-count). Only emit session-level metadata.
           events.push({
             type: 'result',
             data: data.result,
-            tokensUsed: totalIn + outputTokens,
-            inputTokens,
-            outputTokens,
-            cacheReadTokens,
-            cacheCreationTokens,
             cost: data.total_cost_usd,
             duration: data.duration_ms,
             turns: data.num_turns,

package/packages/gui/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@groove-dev/gui",
-  "version": "0.27.35",
+  "version": "0.27.37",
   "description": "GROOVE GUI — visual agent control plane",
   "license": "FSL-1.1-Apache-2.0",
   "type": "module",