@botbuddy/cli 1.33.2 → 1.33.3

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@botbuddy/cli",
-  "version": "1.33.2",
+  "version": "1.33.3",
   "description": "BotBuddy — Swarm coordination CLI for multi-agent workflows",
   "type": "module",
   "bin": {

package/src/stack.mjs

@@ -713,6 +713,20 @@ async function registerLeaseWait(leaseId, timeoutSec, auth, signal, fetchImpl =
   return { waitSessionId: body.wait_session_id ?? null, cursorStart: body.cursor_start ?? null };
 }
 
+// BOT-1827: the abort reason is always one of these four explicit strings —
+// branch on the reason itself rather than inferring "anything but interrupted
+// is a timeout". "matched"/"failed" are only ever set AFTER the SSE loop has
+// already been left via `break` (never from inside the `for await`), so the
+// controller's own teardown can never race a `{ woke: true }` / `{ failed: true }`
+// return the way it did when `ac.abort()` was called before returning from
+// inside the loop.
+function resultForAbortReason(reason) {
+  if (reason === "interrupted") return { result: { interrupted: true }, status: "error" };
+  // "timeout", or any other/unset reason (e.g. an externally-triggered AbortError
+  // with no reason attached) — defensively treated the same as a real deadline.
+  return { result: { timeout: true }, status: "timeout" };
+}
+
 /**
  * Zero-poll wait for `lease:<leaseId>` to reach a state satisfying `isDone(state)`.
  * Registers a wait_session (visible on /waits), then blocks on the event-stream SSE.
@@ -751,11 +765,17 @@ export async function waitForLease(leaseId, isDone, isFailed, { timeoutSec, auth
     }
     return result;
   };
-  if (ac.signal.aborted) return finish(ac.signal.reason === "interrupted" ? { interrupted: true } : { timeout: true }, ac.signal.reason === "interrupted" ? "error" : "timeout");
+  if (ac.signal.aborted) {
+    const { result, status } = resultForAbortReason(ac.signal.reason);
+    return finish(result, status);
+  }
   try {
     ({ cursorStart, waitSessionId } = await registerLeaseWait(leaseId, Math.max(0, Math.ceil((absoluteDeadlineMs - Date.now()) / 1000)), auth, ac.signal, fetchImpl));
   } catch (err) {
-    if (ac.signal.aborted) return finish(ac.signal.reason === "interrupted" ? { interrupted: true } : { timeout: true }, ac.signal.reason === "interrupted" ? "error" : "timeout");
+    if (ac.signal.aborted) {
+      const { result, status } = resultForAbortReason(ac.signal.reason);
+      return finish(result, status);
+    }
     process.stderr.write(`${yellow("⚠")} stack: wait registration failed (${err?.message ?? err}); the lease will not appear on /waits — parking live-only.\n`);
   }
   const url = new URL(eventStreamBase());
@@ -767,15 +787,27 @@ export async function waitForLease(leaseId, isDone, isFailed, { timeoutSec, auth
   try {
     res = await fetchImpl(url, { headers: { ...relayAuthHeaders(auth), Accept: "text/event-stream", "Accept-Encoding": "identity" }, signal: ac.signal });
   } catch (err) {
-    if (ac.signal.aborted) return finish(ac.signal.reason === "interrupted" ? { interrupted: true } : { timeout: true }, ac.signal.reason === "interrupted" ? "error" : "timeout");
+    if (ac.signal.aborted) {
+      const { result, status } = resultForAbortReason(ac.signal.reason);
+      return finish(result, status);
+    }
     return finish({ error: `sse connect: ${err?.message ?? err}` }, "error");
   }
   if (res.status === 401 || res.status === 403) return finish({ auth: true }, "error");
   if (!res.ok || !res.body) return finish({ error: `sse responded ${res.status}` }, "error");
   const decoder = new TextDecoder();
   let buf = "";
+  // BOT-1827: record a match/failure locally and `break` out of BOTH loops
+  // without touching `ac`/`abort()` from inside the `for await`. Aborting the
+  // controller while it is still the signal driving the body iterator makes
+  // the implicit IteratorClose (triggered by `break`/`return`) reject with an
+  // AbortError, which used to land in the `catch` below and override the
+  // already-decided `{ woke: true }` / `{ failed: true }` result with a
+  // `{ timeout: true }` — finalizing the wait session twice. Deciding first,
+  // aborting only AFTER the loop has been left cleanly, removes the race.
+  let decided = null;
   try {
-    for await (const chunk of res.body) {
+    frameLoop: for await (const chunk of res.body) {
       buf += decoder.decode(chunk, { stream: true });
       const { frames, rest } = parseSseFrames(buf);
       buf = rest;
@@ -784,13 +816,26 @@ export async function waitForLease(leaseId, isDone, isFailed, { timeoutSec, auth
         try { sig = JSON.parse(f.data); } catch { continue; }
         if (sig.signal_type !== "stack_lease" || sig.subject_key !== `lease:${leaseId}`) continue;
         const st = sig.payload?.state ?? null;
-        if (isFailed(st)) { ac.abort(); return finish({ failed: true, state: st }, "error"); }
-        if (isDone(st)) { ac.abort(); return finish({ woke: true, state: st }, "matched"); }
+        if (isFailed(st)) { decided = { reason: "failed", result: { failed: true, state: st }, status: "error" }; break frameLoop; }
+        if (isDone(st)) { decided = { reason: "matched", result: { woke: true, state: st }, status: "matched" }; break frameLoop; }
       }
     }
   } catch (err) {
-    if (ac.signal.aborted) return finish(ac.signal.reason === "interrupted" ? { interrupted: true } : { timeout: true }, ac.signal.reason === "interrupted" ? "error" : "timeout");
-    return finish({ error: `sse stream: ${err?.message ?? err}` }, "error");
+    // A decided result always wins over a teardown error surfaced by leaving
+    // the loop (e.g. the body iterator's `return()` rejecting during
+    // cleanup) — fall through to the `decided` handling below instead of
+    // reporting the teardown failure.
+    if (!decided) {
+      if (ac.signal.aborted) {
+        const { result, status } = resultForAbortReason(ac.signal.reason);
+        return finish(result, status);
+      }
+      return finish({ error: `sse stream: ${err?.message ?? err}` }, "error");
+    }
+  }
+  if (decided) {
+    ac.abort(decided.reason);
+    return finish(decided.result, decided.status);
   }
   // A clean relay EOF is not a timeout: the server intentionally closes on
   // scope changes and proxies can recycle idle streams. Re-arm from a fresh
@@ -1868,8 +1913,16 @@ export async function runStackLifecycle(opts, childArgv, adapters = {}) {
     if (reaped?.failed) return cleanupResult = { ok: false, error: "the signed reap failed and the lease may still hold containers (fenced) — see stack hygiene" };
     // Belt-and-braces: the same race can still slip between the pre-check above
     // and the wait registering (or the wait's own live-signal miss), so recheck
-    // once more before falling back to a generic timeout/error message.
-    const postCheck = durableReapFailure(await api.get(leaseId));
+    // once more before falling back to a generic timeout/error message. Mirrors
+    // the provision wait's BOT-1822 recheck (which only ever needed to confirm
+    // failure), but the reap wait can miss EITHER terminal outcome the same
+    // way — BOT-1827 (lease 239af20d): the reap genuinely completed (server
+    // `reaped` live emit) while the zero-poll wait still reported `{ timeout }`,
+    // so a healthy stack was reported as a hard failure. One direct read tells
+    // a real timeout apart from a missed wake in either direction.
+    const recheck = await api.get(leaseId);
+    if (recheck?.ok && recheck.data?.success && recheck.data.state === "reaped") return cleanupResult = { ok: true, state: "reaped" };
+    const postCheck = durableReapFailure(recheck);
     if (postCheck) return cleanupResult = postCheck;
     return cleanupResult = { ok: false, error: reaped?.timeout ? `signed reap did not arrive within ${opts.reapTimeout}s` : (reaped?.error || "signed reap was not proven") };
     })();