@melaya/runner 1.0.106 → 1.0.109

package/dist/assistantHost.py

@@ -421,8 +421,11 @@ def _build_agent():
         # the task is finished. Give phone runs real headroom (they are HITL-gated,
         # so each write still needs the user's tap — cost stays bounded). Connector
         # WRITE fan-outs (1 list + N sends, each its own reasoning+HITL round) keep
-        # the 40 that fits them; read-only Q&A needs very few.
-        max_iters=100 if phone_enabled else (40 if connector_services else 8),
+        # the 40 that fits them; read-only Q&A needs very few. Phone raised + env-tunable
+        # for long-horizon tasks (100s of steps) now that per-step tokens are pruned +
+        # cached; each write is still HITL-gated so cost stays bounded.
+        max_iters=(int(os.environ.get("MEL_ASSISTANT_MAX_ITERS_PHONE", "300") or "300") if phone_enabled
+                   else (40 if connector_services else 8)),
         reliability=True,
         bounded_memory=True,
     )

package/dist/connection.js

@@ -1030,6 +1030,13 @@ export async function connect(opts) {
             activeAssistants.set(sid, session);
             console.log(chalk.hex("#7c6ff0")(`  ◆ Assistant session ${sid.slice(0, 10)}… (${payload.provider})`));
             proc.stdout?.on("data", (data) => {
+                // ANY host output means the turn is actively working — refresh the idle
+                // clock. Without this, lastActivity was set ONLY at turn start (assistant_turn),
+                // so a SINGLE long turn (a phone task is one turn with ~100 tool calls streaming
+                // for >15 min) was reaped mid-action by the idle sweep despite continuous
+                // activity ("idle-swept" while properly working). Now only a genuinely silent
+                // host (no output for ASSISTANT_IDLE_MS) is swept.
+                session.lastActivity = Date.now();
                 session.stdoutBuf += data.toString();
                 let idx;
                 while ((idx = session.stdoutBuf.indexOf("\n")) >= 0) {
@@ -1170,6 +1177,14 @@ export async function connect(opts) {
             activeAssistants.delete(String(payload.sessionId));
         }
     });
+    // Keepalive: the server is blocked waiting on the user's behalf (a HITL approval
+    // poll, a slow phone action) so the host emits no stdout, but the turn is ALIVE.
+    // Refresh the idle clock so the idle sweep doesn't reap a host mid-approval.
+    socket.on("runner:assistant_keepalive", (payload) => {
+        const s = activeAssistants.get(String(payload.sessionId || ""));
+        if (s)
+            s.lastActivity = Date.now();
+    });
     // Cooperative cancel: the user pressed STOP in chat. Write a cancel line to the
     // host's stdin — its stdin reader thread flips a flag the running turn's
     // hooks poll, so the agent loop aborts BETWEEN steps (right where the next

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@melaya/runner",
-  "version": "1.0.106",
+  "version": "1.0.109",
   "description": "Run Melaya AI pipelines locally with your own LM Studio or Ollama models",
   "license": "UNLICENSED",
   "private": false,