bosun 0.40.15 → 0.40.17

package/agent/agent-pool.mjs

@@ -40,6 +40,8 @@
  */
 
 import { resolve, dirname } from "node:path";
+import { existsSync, readFileSync } from "node:fs";
+import { homedir } from "node:os";
 import { fileURLToPath } from "node:url";
 import { loadConfig } from "../config/config.mjs";
 import { resolveRepoRoot, resolveAgentRepoRoot } from "../config/repo-root.mjs";
@@ -566,16 +568,28 @@ function hasSdkPrerequisites(name, runtimeEnv = process.env) {
   }
 
   if (name === "codex") {
-    // Codex needs an OpenAI API key (or Azure key, or profile-specific key)
+    // Codex needs an OpenAI API key (or Azure key, or profile-specific key),
+    // OR a valid ~/.codex/config.toml where an env_key reference is satisfied.
     const hasKey =
       runtimeEnv.OPENAI_API_KEY ||
       runtimeEnv.AZURE_OPENAI_API_KEY ||
       runtimeEnv.CODEX_MODEL_PROFILE_XL_API_KEY ||
       runtimeEnv.CODEX_MODEL_PROFILE_M_API_KEY;
-    if (!hasKey) {
-      return { ok: false, reason: "no API key (OPENAI_API_KEY / AZURE_OPENAI_API_KEY)" };
+    if (hasKey) return { ok: true, reason: null };
+    // Check ~/.codex/config.toml — Codex CLI SDK reads auth env_key refs from there
+    try {
+      const configToml = resolve(homedir(), ".codex", "config.toml");
+      if (existsSync(configToml)) {
+        const tomlText = readFileSync(configToml, "utf8");
+        // Extract all env_key = "VAR_NAME" entries and check if any are set
+        for (const match of tomlText.matchAll(/env_key\s*=\s*"([^"]+)"/g)) {
+          if (runtimeEnv[match[1]]) return { ok: true, reason: null };
+        }
+      }
+    } catch {
+      // best effort — fall through to failure
     }
-    return { ok: true, reason: null };
+    return { ok: false, reason: "no API key (OPENAI_API_KEY / AZURE_OPENAI_API_KEY) and no satisfied env_key in ~/.codex/config.toml" };
   }
   if (name === "copilot") {
     // Copilot auth can come from multiple sources (OAuth manager, gh auth,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "bosun",
-  "version": "0.40.15",
+  "version": "0.40.17",
   "description": "Bosun Autonomous Engineering — manages AI agent executors with failover, extremely powerful workflow builder, and a massive amount of included default workflow templates for autonomous engineering, creates PRs via Vibe-Kanban API, and sends Telegram notifications. Supports N executors with weighted distribution, multi-repo projects, and auto-setup.",
   "type": "module",
   "license": "Apache-2.0",

package/task/task-executor.mjs

@@ -822,7 +822,7 @@ const MAX_IDLE_CONTINUES = 5;
 const WATCHDOG_WARMUP_MS = 5 * 60_000; // 5 minutes warmup
 const SHARED_STATE_ENABLED = process.env.SHARED_STATE_ENABLED !== "false";
 const SHARED_STATE_STALE_THRESHOLD_MS =
-  Number(process.env.SHARED_STATE_STALE_THRESHOLD_MS) || 300_000;
+  Number(process.env.SHARED_STATE_STALE_THRESHOLD_MS) || 600_000;
 const NO_COMMIT_STATE_FILE = resolve(
   dirname(fileURLToPath(import.meta.url)),
   "..",
@@ -3900,9 +3900,12 @@ class TaskExecutor {
           const ownerId = sharedState?.ownerId || null;
           const heartbeat =
             sharedState?.ownerHeartbeat || sharedState?.heartbeat || null;
+          // Any non-stale owner (workflow run or executor instance) should
+          // block recovery re-dispatch. Removing the ownerId !== instanceId
+          // guard ensures workflow-owned tasks (wf-<uuid> owners) are also
+          // protected when action.claim_task IS used.
           if (
             ownerId &&
-            ownerId !== this._instanceId &&
             !isSharedHeartbeatStale(heartbeat, SHARED_STATE_STALE_THRESHOLD_MS)
           ) {
             skippedForActiveClaim++;
@@ -3977,6 +3980,43 @@ class TaskExecutor {
       }
       const isFreshEnough =
         ageMs === 0 || ageMs <= INPROGRESS_RECOVERY_MAX_AGE_MS;
+
+      // In workflow-owned mode, calling executeTask() fires task.assigned
+      // which launches a new workflow run. If one already exists (evidenced by
+      // an alive agent thread), that creates two competing runs → owner_mismatch.
+      // trigger.task_assigned workflows don't call action.claim_task, so ownerId
+      // is null and the shared-state guard above cannot protect against this.
+      // • Active thread → workflow is still managing it; leave it alone.
+      // • No active thread but fresh → agent died; reset to todo so
+      //   trigger.task_available re-dispatches cleanly, without double-dispatch.
+      if (this.workflowOwnsTaskLifecycle) {
+        if (hasThread) {
+          skippedForActiveClaim++;
+          continue;
+        }
+        if (isFreshEnough) {
+          try {
+            await transitionTaskStatus(id, "todo", {
+              source: "task-executor-recovery-workflow-owned",
+            });
+          } catch {
+            /* best effort */
+          }
+          try {
+            transitionInternalTaskStatus(
+              id,
+              "todo",
+              "task-executor-recovery-workflow-owned",
+            );
+          } catch {
+            /* best effort */
+          }
+          this._removeRuntimeSlot(id);
+          resetToTodo++;
+          continue;
+        }
+      }
+
       if (hasThread || isFreshEnough) {
         if (!internalTask) {
           try {

package/ui/components/chat-view.js.bak

@@ -0,0 +1 @@
+/* backup of original chat-view.js before MUI transformation */

package/ui/tabs/infra.js.bak

@@ -0,0 +1 @@
+/* backup - original infra.js before MUI transformation */