V37 Blade autoheal + honest agent diag - User GO ON FAIT RIEN DE MANUEL PUTAIN doctrine 7 zero manuel - DECOUVERTE CAUSE RACINE doctrine 4 HONNETE: agent Razer côté Windows heartbeat OK (13:23 UTC fresh) MAIS exec pipeline cassé depuis 19avr - pickup tasks (status dispatched) MAIS JAMAIS callback task_done - derniere task vraiment executee 19avr 10:05 - 12 dispatched zombies depuis 01:04 jamais complétés + mes 2 test tasks Test-NetConnection 9222 stuck en dispatched - IMPOSSIBLE lancer Chrome --remote-debugging-port=9222 depuis S204 sans agent fonctionnel Windows - Solutions côté serveur zero manuel Yacine: 1) Creation /api/blade-tasks-cleanup.php auto-cleanup dispatched >10min mark failed_timeout + agent_health verdict HEALTHY/DEGRADED/BROKEN selon last done age_hours 2) Cron /etc/cron.d/blade-autoheal every 5min auto-reset zombies 3) Andon weval.andon_alerts insert 'blade-agent-exec' severity high pour notifier Yacine qu'il doit restart agent sur sa Razer - Test live: dry_run show 8+ zombies tasks 200+min age dispatched jamais completed - verdict BROKEN agent stale_hours >24 - Solution long terme: Yacine doit redémarrer agent Python Windows sur sa Razer (doctrine limitation: S204 ne peut pas restarter un processus Windows crashed) - Infra MCP blade_exec + blade_chrome_cdp tools tous defined v1.1.0 17 tools prêts mais UNUSABLE jusqu'à redémarrage agent - Wiki V36 DOCTRINE-BLADE-IA-REMOTE doit noter cette limitation - NonReg 153/153 stable 50eme session - Services 23/23 UP - L99 201/201 6sigma - Alerts now 3 warning (ajout blade-agent-exec) - Office APP hub 6403 accounts prêt à piloter via Blade dès que agent OK - Doctrine 4 HONNETE gap brutal exposé doctrine 5 sequence auto-heal doctrine 7 zero manuel côté serveur seulement doctrine 13 cause racine agent Windows pas S204 doctrine 14 additif blade-tasks-cleanup.php nouveau [Opus V37 blade-autoheal-honest-diag]

api/blade-tasks-cleanup.php

@@ -0,0 +1,63 @@
+<?php
+// V37 Opus - Auto-cleanup blade tasks dispatched > 10min = zombies
+// Doctrine #4 HONNETE + #13 cause racine + #7 zero manuel
+header("Content-Type: application/json");
+
+$TASKS_DIR = "/var/www/html/api/blade-tasks";
+$TIMEOUT_MINUTES = (int)($_GET["timeout_min"] ?? 10);
+$DRY = isset($_GET["dry"]);
+
+$now = time();
+$cutoff = $now - ($TIMEOUT_MINUTES * 60);
+$results = ["timed_out" => [], "kept" => 0, "failed_marked" => 0];
+
+foreach (glob($TASKS_DIR . "/task_*.json") as $tf) {
+    $td = @json_decode(file_get_contents($tf), true);
+    if (!$td) continue;
+    if (($td["status"] ?? "") !== "dispatched") { $results["kept"]++; continue; }
+    
+    $disp = $td["dispatched_at"] ?? $td["created"] ?? "";
+    $t = $disp ? strtotime($disp) : 0;
+    if (!$t || $t > $cutoff) { $results["kept"]++; continue; }
+    
+    $results["timed_out"][] = [
+        "id" => $td["id"] ?? basename($tf),
+        "dispatched_at" => $disp,
+        "age_min" => round(($now - $t) / 60, 1),
+        "label" => $td["label"] ?? "?"
+    ];
+    
+    if (!$DRY) {
+        $td["status"] = "failed_timeout";
+        $td["failed_at"] = date("c");
+        $td["error"] = "Agent Blade did not callback task_done within {$TIMEOUT_MINUTES}min";
+        file_put_contents($tf, json_encode($td, JSON_PRETTY_PRINT));
+        $results["failed_marked"]++;
+    }
+}
+
+// Summary + agent health assessment
+$last_done = 0;
+foreach (glob($TASKS_DIR . "/task_*.json") as $tf) {
+    $td = @json_decode(file_get_contents($tf), true);
+    if (($td["status"] ?? "") === "done") {
+        $c = strtotime($td["completed_at"] ?? "1970-01-01");
+        if ($c > $last_done) $last_done = $c;
+    }
+}
+$agent_stale_hours = $last_done ? round(($now - $last_done) / 3600, 1) : 9999;
+
+echo json_encode([
+    "ok" => true,
+    "v" => "V37-blade-autoheal",
+    "ts" => date("c"),
+    "dry_run" => $DRY,
+    "timeout_minutes" => $TIMEOUT_MINUTES,
+    "stats" => $results,
+    "agent_health" => [
+        "last_done_ts" => $last_done ? date("c", $last_done) : null,
+        "stale_hours" => $agent_stale_hours,
+        "agent_execution_ok" => $agent_stale_hours < 1,
+        "verdict" => $agent_stale_hours < 1 ? "HEALTHY" : ($agent_stale_hours < 24 ? "DEGRADED" : "BROKEN")
+    ]
+], JSON_PRETTY_PRINT);