@nexus-cortex/core 4.81.0 → 4.82.0

package/dist/training/thrashDetector.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"thrashDetector.d.ts","sourceRoot":"","sources":["../../src/training/thrashDetector.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAEH,MAAM,WAAW,YAAY;IAC3B,8DAA8D;IAC9D,aAAa,EAAE,MAAM,CAAC;IACtB,kEAAkE;IAClE,MAAM,EAAE,MAAM,CAAC;IACf,4FAA4F;IAC5F,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,MAAM,WAAW,WAAW;IAC1B,0EAA0E;IAC1E,SAAS,EAAE,OAAO,CAAC;IACnB,+CAA+C;IAC/C,QAAQ,EAAE,MAAM,CAAC;IACjB,8DAA8D;IAC9D,QAAQ,EAAE,MAAM,CAAC;CAClB;AAED,eAAO,MAAM,eAAe,EAAE,YAA2D,CAAC;AAO1F,wBAAgB,mBAAmB,CAAC,GAAG,GAAE,MAAM,CAAC,UAAwB,GAAG,YAAY,CAMtF;AAED;;;;;;;;;;GAUG;AACH,wBAAgB,kBAAkB,CAChC,QAAQ,EAAE,OAAO,EAAE,EACnB,SAAS,EAAE,MAAM,EACjB,GAAG,GAAE,YAAoC,GACxC,WAAW,CAUb"}

package/dist/training/thrashDetector.js

@@ -0,0 +1,45 @@
+/**
+ * thrashDetector — shape-agnostic STRUGGLE signal (MENTORSHIP_ASK_FOR_ADVICE_SPEC §2).
+ *
+ * The dominant bench failure mode is DIVERSE-exploration thrash: the model makes many
+ * DIFFERENT failing attempts (44–105 calls on hard tasks) that the loop/approach
+ * detectors miss because it is not a clean repeated loop. This fires on failure
+ * DENSITY over a recent window — regardless of shape — once past a turn floor (a few
+ * failing probes early are normal debugging; cf. MAX_CONSECUTIVE_ERRORS=6). Pure +
+ * env-configurable; the orchestrator feeds it the recent tool outcomes it already
+ * tracks (the decision-store success/fail signal). Sterile-bench-safe (hot tier, no key).
+ */
+export const THRASH_DEFAULTS = { failThreshold: 4, window: 6, minTurns: 5 };
+function posInt(v, d) {
+    const n = parseInt((v ?? '').trim(), 10);
+    return Number.isInteger(n) && n > 0 ? n : d;
+}
+export function resolveThrashConfig(env = process.env) {
+    return {
+        failThreshold: posInt(env.CORTEX_THRASH_FAILS, THRASH_DEFAULTS.failThreshold),
+        window: posInt(env.CORTEX_THRASH_WINDOW, THRASH_DEFAULTS.window),
+        minTurns: posInt(env.CORTEX_THRASH_MIN_TURNS, THRASH_DEFAULTS.minTurns),
+    };
+}
+/**
+ * Decide whether the agent is thrashing.
+ *
+ * @param outcomes   success booleans of tool calls, chronological (oldest→newest).
+ * @param turnCount  total tool calls so far (>= outcomes.length).
+ * @param cfg        thresholds (defaults from env).
+ *
+ * Fires only when: past the turn floor, a FULL window is available, the failure
+ * count meets the threshold, AND the most-recent outcome is a failure (so it never
+ * fires right after the model just made progress).
+ */
+export function resolveThrashState(outcomes, turnCount, cfg = resolveThrashConfig()) {
+    const win = outcomes.slice(-cfg.window);
+    const failures = win.reduce((n, ok) => n + (ok ? 0 : 1), 0);
+    const currentlyFailing = win.length > 0 && win[win.length - 1] === false;
+    const thrashing = turnCount >= cfg.minTurns &&
+        win.length >= cfg.window &&
+        failures >= cfg.failThreshold &&
+        currentlyFailing;
+    return { thrashing, failures, examined: win.length };
+}
+//# sourceMappingURL=thrashDetector.js.map

package/dist/training/thrashDetector.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"thrashDetector.js","sourceRoot":"","sources":["../../src/training/thrashDetector.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;GAUG;AAoBH,MAAM,CAAC,MAAM,eAAe,GAAiB,EAAE,aAAa,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,QAAQ,EAAE,CAAC,EAAE,CAAC;AAE1F,SAAS,MAAM,CAAC,CAAqB,EAAE,CAAS;IAC9C,MAAM,CAAC,GAAG,QAAQ,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,EAAE,EAAE,CAAC,CAAC;IACzC,OAAO,MAAM,CAAC,SAAS,CAAC,CAAC,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;AAC9C,CAAC;AAED,MAAM,UAAU,mBAAmB,CAAC,MAAyB,OAAO,CAAC,GAAG;IACtE,OAAO;QACL,aAAa,EAAE,MAAM,CAAC,GAAG,CAAC,mBAAmB,EAAE,eAAe,CAAC,aAAa,CAAC;QAC7E,MAAM,EAAE,MAAM,CAAC,GAAG,CAAC,oBAAoB,EAAE,eAAe,CAAC,MAAM,CAAC;QAChE,QAAQ,EAAE,MAAM,CAAC,GAAG,CAAC,uBAAuB,EAAE,eAAe,CAAC,QAAQ,CAAC;KACxE,CAAC;AACJ,CAAC;AAED;;;;;;;;;;GAUG;AACH,MAAM,UAAU,kBAAkB,CAChC,QAAmB,EACnB,SAAiB,EACjB,MAAoB,mBAAmB,EAAE;IAEzC,MAAM,GAAG,GAAG,QAAQ,CAAC,KAAK,CAAC,CAAC,GAAG,CAAC,MAAM,CAAC,CAAC;IACxC,MAAM,QAAQ,GAAG,GAAG,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,EAAE,EAAE,CAAC,CAAC,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC;IAC5D,MAAM,gBAAgB,GAAG,GAAG,CAAC,MAAM,GAAG,CAAC,IAAI,GAAG,CAAC,GAAG,CAAC,MAAM,GAAG,CAAC,CAAC,KAAK,KAAK,CAAC;IACzE,MAAM,SAAS,GACb,SAAS,IAAI,GAAG,CAAC,QAAQ;QACzB,GAAG,CAAC,MAAM,IAAI,GAAG,CAAC,MAAM;QACxB,QAAQ,IAAI,GAAG,CAAC,aAAa;QAC7B,gBAAgB,CAAC;IACnB,OAAO,EAAE,SAAS,EAAE,QAAQ,EAAE,QAAQ,EAAE,GAAG,CAAC,MAAM,EAAE,CAAC;AACvD,CAAC"}

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@nexus-cortex/core",
-  "version": "4.81.0",
+  "version": "4.82.0",
   "description": "Core orchestration library implementing Claude CLI architecture patterns",
   "type": "module",
   "main": "./dist/index.js",
@@ -47,8 +47,8 @@
     "@google/generative-ai": "^0.2.1",
     "@gradio/client": "^2.3.1",
     "@modelcontextprotocol/sdk": "^1.20.2",
-    "@nexus-cortex/executors": "4.81.0",
-    "@nexus-cortex/types": "4.81.0",
+    "@nexus-cortex/executors": "4.82.0",
+    "@nexus-cortex/types": "4.82.0",
     "@types/uuid": "^10.0.0",
     "ajv": "^8.12.0",
     "chokidar": "^3.6.0",