@melaya/runner 1.0.111 → 1.0.114

package/dist/assistantHost.py

@@ -72,6 +72,40 @@ def _memory_watermark(agent) -> int:
         return 0
 
 
+def _emit_usage(agent, turn_id: str) -> None:
+    """Live header meter (runner parity): the context-fullness GAUGE — memory
+    watermark vs budget — so the FE shows how full the window is before the next
+    compaction, exactly like the cloud path. Per-turn TOKEN totals are exported as
+    spans (Overview dashboard); the live in/out counter stays cloud-only for now."""
+    try:
+        mem = _agent_memory(agent)
+        if mem is None:
+            return
+        budget = int(getattr(mem, "max_tokens", 0) or 0)
+        # Real context fullness = current retained-content tokens (NOT the rehydration
+        # watermark, which reads ~0). Falls back to 0 on an old shared bundle without
+        # current_tokens() — the gauge just won't move until the bundle updates.
+        used = int(mem.current_tokens()) if hasattr(mem, "current_tokens") else 0
+        # Live in/out tokens for THIS turn: force-flush the pending trace spans, then
+        # drain the exporter's accumulator (the SAME tokens that land in agents.spans).
+        # The BatchSpanProcessor exports asynchronously, so flush first or the turn's
+        # spans may not have been counted yet. Best-effort; 0 on an old shared bundle.
+        in_tok = out_tok = 0
+        try:
+            from opentelemetry import trace as _ot
+            tp = _ot.get_tracer_provider()
+            if hasattr(tp, "force_flush"):
+                tp.force_flush()
+            from shared.runtime.tracing_exporter import drain_token_totals
+            in_tok, out_tok = drain_token_totals()
+        except Exception:
+            pass
+        if budget > 0 or in_tok or out_tok:
+            _emit(turn_id, "usage", turnInTok=int(in_tok), turnOutTok=int(out_tok), ctxUsed=used, ctxBudget=budget)
+    except Exception:
+        pass
+
+
 def _sync_ollama_memory_budget(agent) -> None:
     """P2-6 (persistent host): the Assistant builds ONE agent + BoundedMemory, so an
     OOM num_ctx downgrade that shrinks the ollama context cache would otherwise leave
@@ -392,6 +426,11 @@ def _build_agent():
         "to find the right content yourself. Keep going until the task is COMPLETE (e.g. "
         "all N comments posted) or you are genuinely blocked; only then report what you "
         "did and what (if anything) is blocked.\n"
+        "- NEVER fabricate a reason for stopping. Do NOT claim 'the connection dropped', "
+        "'I lost connection', 'the session ended' or invent ANY infra failure — you "
+        "cannot observe that and it is almost always false. If a phone action is slow or "
+        "errors, RETRY it; only if it truly won't recover after retries do you stop, and "
+        "then state the EXACT tool that failed + what the screen showed.\n"
         if phone_enabled else ""
     )
     connector_rule = (
@@ -761,6 +800,7 @@ def main() -> int:
         # the ollama context (no-op for every other provider / when unchanged).
         _sync_ollama_memory_budget(agent)
         _run_turn(agent, turn_id, message)
+        _emit_usage(agent, turn_id)
 
 
 if __name__ == "__main__":

package/package.json

@@ -1,42 +1,42 @@
-{
-  "name": "@melaya/runner",
-  "version": "1.0.111",
-  "description": "Run Melaya AI pipelines locally with your own LM Studio or Ollama models",
-  "license": "UNLICENSED",
-  "private": false,
-  "type": "module",
-  "bin": {
-    "melaya-runner": "dist/cli.js"
-  },
-  "main": "dist/index.js",
-  "files": [
-    "dist/**/*.js",
-    "dist/**/*.d.ts",
-    "dist/**/*.py",
-    "localRagIngest.py",
-    "localRagRetrieve.py",
-    "nltk_data/**",
-    "README.md"
-  ],
-  "scripts": {
-    "build": "tsc && node -e \"const fs=require('fs'); fs.copyFileSync('localRagIngest.py','dist/localRagIngest.py'); fs.copyFileSync('localRagRetrieve.py','dist/localRagRetrieve.py'); fs.copyFileSync('src/assistantHost.py','dist/assistantHost.py')\"",
-    "prepublishOnly": "npm run build"
-  },
-  "dependencies": {
-    "chalk": "^5.3.0",
-    "commander": "^12.0.0",
-    "ora": "^8.0.0",
-    "playwright": "^1.47.0",
-    "socket.io-client": "^4.8.0"
-  },
-  "devDependencies": {
-    "@types/node": "^20.0.0",
-    "typescript": "^5.5.0"
-  },
-  "engines": {
-    "node": ">=18"
-  },
-  "publishConfig": {
-    "access": "public"
-  }
-}
+{
+  "name": "@melaya/runner",
+  "version": "1.0.114",
+  "description": "Run Melaya AI pipelines locally with your own LM Studio or Ollama models",
+  "license": "UNLICENSED",
+  "private": false,
+  "type": "module",
+  "bin": {
+    "melaya-runner": "dist/cli.js"
+  },
+  "main": "dist/index.js",
+  "files": [
+    "dist/**/*.js",
+    "dist/**/*.d.ts",
+    "dist/**/*.py",
+    "localRagIngest.py",
+    "localRagRetrieve.py",
+    "nltk_data/**",
+    "README.md"
+  ],
+  "scripts": {
+    "build": "tsc && node -e \"const fs=require('fs'); fs.copyFileSync('localRagIngest.py','dist/localRagIngest.py'); fs.copyFileSync('localRagRetrieve.py','dist/localRagRetrieve.py'); fs.copyFileSync('src/assistantHost.py','dist/assistantHost.py')\"",
+    "prepublishOnly": "npm run build"
+  },
+  "dependencies": {
+    "chalk": "^5.3.0",
+    "commander": "^12.0.0",
+    "ora": "^8.0.0",
+    "playwright": "^1.47.0",
+    "socket.io-client": "^4.8.0"
+  },
+  "devDependencies": {
+    "@types/node": "^20.0.0",
+    "typescript": "^5.5.0"
+  },
+  "engines": {
+    "node": ">=18"
+  },
+  "publishConfig": {
+    "access": "public"
+  }
+}