@mediagato/modelreins-channel 4.5.0 → 4.5.1

package/modelreins-channel.mjs

@@ -1013,50 +1013,82 @@ async function heartbeat() {
  *   4. Claude reads it, executes the task, calls modelreins_complete when done
  */
 let _pollCount = 0;
+let _pollInProgress = false;
 
 async function pollJobs() {
+    // setInterval does not await us, so a slow tick (e.g. a stalled API
+    // call) can overlap with the next one. Guard against re-entry.
+    if (_pollInProgress) return;
     // One job at a time — don't accept new work while something is running.
     if (currentJobId) return;
+    _pollInProgress = true;
 
-    // Check killswitch every 12 polls (~60s at default POLL_MS).
-    _pollCount++;
-    if (_pollCount % 12 === 1) {
-        try {
+    // Track the optimistic local claim so we can roll it back if the
+    // claim PUT or the MCP notification fails before Claude sees the job.
+    let claimedJobId = null;
+
+    try {
+        // Check killswitch every 12 polls (~60s at default POLL_MS).
+        _pollCount++;
+        if (_pollCount % 12 === 1) {
             const ks = await api('GET', '/killswitch');
-            if (ks.data?.active) {
+            if (ks.status >= 200 && ks.status < 300 && ks.data?.active) {
                 const reason = ks.data.reason || 'killswitch active';
                 await api('PUT', '/presence', { instance: WORKER_NAME, status: 'offline', details: `Killswitch: ${reason}` });
                 process.exit(0);
             }
-        } catch (_) { /* network error — don't stop polling */ }
-    }
+        }
+
+        const res = await api('GET', `/jobs?status=pending&assigned_to=${encodeURIComponent(WORKER_NAME)}&limit=1`);
+        if (res.status < 200 || res.status >= 300) return;
+        const jobs = res.data?.jobs || (Array.isArray(res.data) ? res.data : []);
+
+        if (jobs.length === 0) return;
+
+        const job = jobs[0];
+        currentJobId = job.id;
+        claimedJobId = job.id;
+
+        // Claim the job immediately so no other worker picks it up. If the
+        // claim fails, roll back so a future poll can retry — without this,
+        // a transient 5xx leaves currentJobId stuck and the channel deadlocks.
+        const claim = await api('PUT', `/jobs/${job.id}`, { status: 'running' });
+        if (claim.status < 200 || claim.status >= 300) {
+            throw new Error(`claim PUT returned ${claim.status}: ${JSON.stringify(claim.data).slice(0, 120)}`);
+        }
 
-    const res = await api('GET', `/jobs?status=pending&assigned_to=${encodeURIComponent(WORKER_NAME)}&limit=1`);
-    const jobs = res.data?.jobs || (Array.isArray(res.data) ? res.data : []);
-
-    if (jobs.length === 0) return;
-
-    const job = jobs[0];
-    currentJobId = job.id;
-
-    // Claim the job immediately so no other worker picks it up.
-    await api('PUT', `/jobs/${job.id}`, { status: 'running' });
-
-    // Inject the job prompt into Claude's context via the channel notification.
-    // Claude Code intercepts this and presents it as a <channel> tag in the session.
-    await mcp.notification({
-        method: 'notifications/claude/channel',
-        params: {
-            content: job.prompt,
-            meta: {
-                job_id:           String(job.id),
-                priority:         String(job.priority ?? 0),
-                model_preference: job.model_preference || '',
-                working_dir:      job.working_dir || '',
-                source:           'modelreins-dashboard',
+        // Inject the job prompt into Claude's context via the channel notification.
+        // Claude Code intercepts this and presents it as a <channel> tag in the session.
+        await mcp.notification({
+            method: 'notifications/claude/channel',
+            params: {
+                content: job.prompt,
+                meta: {
+                    job_id:           String(job.id),
+                    priority:         String(job.priority ?? 0),
+                    model_preference: job.model_preference || '',
+                    working_dir:      job.working_dir || '',
+                    source:           'modelreins-dashboard',
+                },
             },
-        },
-    });
+        });
+
+        // Notification delivered — Claude now owns the job. modelreins_complete
+        // will clear currentJobId when Claude reports back.
+        claimedJobId = null;
+    } catch (err) {
+        // Roll back the optimistic local claim so the next poll can retry.
+        if (claimedJobId !== null && currentJobId === claimedJobId) {
+            currentJobId = null;
+            // Best-effort: return the job to pending on the server too.
+            // If this also fails, the server-side stale-running cleanup will
+            // eventually reclaim it; we don't want to retry forever here.
+            api('PUT', `/jobs/${claimedJobId}`, { status: 'pending' }).catch(() => {});
+        }
+        console.error(`[channel] pollJobs error: ${err?.message || err}`);
+    } finally {
+        _pollInProgress = false;
+    }
 }
 
 // Kick off the background loops.

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mediagato/modelreins-channel",
-  "version": "4.5.0",
+  "version": "4.5.1",
   "description": "ModelReins MCP Channel — turn any Claude Code session into a ModelReins worker",
   "type": "module",
   "main": "modelreins-channel.mjs",