@caupulican/pi-adaptative 0.80.103 → 0.81.0

package/dist/core/research/model-fitness.js

@@ -287,6 +287,27 @@ export async function runModelFitnessProbe(options) {
     const tokensPerSecond = overallSpeed.evalMs > 0 ? Math.round((overallSpeed.tokens / overallSpeed.evalMs) * 1000) : undefined;
     return { trials, tokensPerSecond, research, worker, judge, search, toolCall, digest, totalCostUsd };
 }
+/**
+ * Pure verdict: true when the probe found ZERO successes on every LANE surface it actually graded
+ * AND the judge (if it ran) also failed. A lane/judge with total 0 (i.e. never run) carries no
+ * evidence and is excluded from the lane check — but at least one lane must actually have been
+ * graded for an all-failed verdict at all: `gradedLanes.every(...)` is vacuously true over an
+ * empty array, so a report where only the judge ran (every research/worker/search/toolCall/digest
+ * lane is ungraded) is excluded explicitly rather than misread as "all lanes failed" on zero lane
+ * evidence. An empty/degenerate report (nothing graded at all, lanes AND judge) is likewise never
+ * mistaken for a failed one. This is the gate adoption flows must consult before assigning a role —
+ * see `isProbeAllFailed` callers in interactive-mode.ts and agent-session.ts.
+ */
+export function isProbeAllFailed(report) {
+    const lanes = [report.research, report.worker, report.search, report.toolCall, report.digest];
+    const gradedLanes = lanes.filter((lane) => lane.total > 0);
+    const judgeGraded = report.judge.total > 0;
+    if (gradedLanes.length === 0 && !judgeGraded)
+        return false;
+    const lanesAllFailed = gradedLanes.length > 0 && gradedLanes.every((lane) => lane.succeeded === 0);
+    const judgeFailed = !judgeGraded || report.judge.parsed === 0;
+    return lanesAllFailed && judgeFailed;
+}
 /** Compact human-readable report for tool output / interactive display. Bounded, no raw dumps. */
 export function formatModelFitnessReport(model, report) {
     const speed = (tokensPerSecond) => tokensPerSecond !== undefined ? `, ~${tokensPerSecond} tok/s` : "";

package/dist/core/research/model-fitness.js.map

@@ -1 +1 @@
-{"version":3,"file":"model-fitness.js","sourceRoot":"","sources":["../../../src/core/research/model-fitness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AAEzE,OAAO,EAAE,6BAA6B,EAAE,MAAM,6BAA6B,CAAC;AAC5E,OAAO,EAAE,SAAS,EAAE,MAAM,gCAAgC,CAAC;AAC3D,OAAO,EAAE,aAAa,EAAE,MAAM,gCAAgC,CAAC;AAC/D,OAAO,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAgCnD,qFAAqF;AACrF,MAAM,CAAC,MAAM,6BAA6B,GAAkC;IAC3E,EAAE,MAAM,EAAE,gEAAgE,EAAE,QAAQ,EAAE,IAAI,EAAE;IAC5F,EAAE,MAAM,EAAE,uDAAuD,EAAE,QAAQ,EAAE,IAAI,EAAE;IACnF,EAAE,MAAM,EAAE,yCAAyC,EAAE,QAAQ,EAAE,IAAI,EAAE;IACrE,EAAE,MAAM,EAAE,4CAA4C,EAAE,QAAQ,EAAE,KAAK,EAAE;IACzE,EAAE,MAAM,EAAE,yCAAyC,EAAE,QAAQ,EAAE,KAAK,EAAE;IACtE,EAAE,MAAM,EAAE,yBAAyB,EAAE,QAAQ,EAAE,KAAK,EAAE;CACtD,CAAC;AAoDF,yFAAyF;AACzF,MAAM,CAAC,MAAM,0BAA0B,GAAG;IACzC,oFAAoF;IACpF,8EAA8E;IAC9E,0FAA0F;IAC1F,6CAA6C;CAC7C,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,CAAC,MAAM,6BAA6B,GAAG;IAC5C,kEAAkE;IAClE,gFAAgF;IAChF,yDAAyD;IACzD,qEAAqE;CACrE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,kBAAkB,GAAsB;IAC7C,iEAAiE;IACjE,0DAA0D;IAC1D,yDAAyD;CACzD,CAAC;AAEF,wFAAsF;AACtF,OAAO,EAAE,6BAA6B,IAAI,0BAA0B,EAAE,MAAM,6BAA6B,CAAC;AAE1G;;;;GAIG;AACH,MAAM,kBAAkB,GAAgD;IACvE;QACC,KAAK,EAAE,sBAAsB;QAC7B,KAAK,EAAE;YACN,0CAA0C;YAC1C,iFAAiF;YACjF,gEAAgE;YAChE,6CAA6C;SAC7C,CAAC,IAAI,CAAC,IAAI,CAAC;KACZ;IACD;QACC,KAAK,EAAE,aAAa;QACpB,KAAK,EAAE;YACN,mBAAmB;YACnB,uBAAuB;YACvB,iFAAiF;YACjF,aAAa;SACb,CAAC,IAAI,CAAC,IAAI,CAAC;KACZ;IACD;QACC,KAAK,EAAE,iBAAiB;QACxB,KAAK,EAAE;YACN,+BAA+B;YAC/B,2BAA2B;YAC3B,iCAAiC;YACjC,kCAAkC;SAClC,CAAC,IAAI,CAAC,IAAI,CAAC;KACZ;CACD,CAAC;AAEF,SAAS,WAAW,CAAC,IAAY,EAAE,KAAa,EAAW;IAC1D,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACvC,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,MAAM,MAAM,GAAI,MAA+B,CAAC,MAAM,CAAC;IACvD,IAAI,OAAO,MAAM,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;IAC9B,6FAA6F;IAC7F,OAAO,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,MAAM,IAAI,GAAG,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;AAAA,CAC9E;AAED,MAAM,qBAAqB,GAAsB;IAChD,yDAAyD;IACzD,iEAAiE;IACjE,0DAA0D;CAC1D,CAAC;AAEF,SAAS,eAAe,CAAC,IAAY,EAAW;IAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACvC,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,MAAM,OAAO,GAAI,MAAgC,CAAC,OAAO,CAAC;IAC1D,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,KAAK,CAAC;IACxF,OAAO,OAAO,CAAC,KAAK,CACnB,CAAC,KAAK,EAAE,EAAE,CACT,KAAK;QACL,OAAO,KAAK,KAAK,QAAQ;QACzB,OAAQ,KAA+B,CAAC,OAAO,KAAK,QAAQ;QAC3D,KAA6B,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CACzD,CAAC;AAAA,CACF;AAED,SAAS,aAAa,CAAC,IAAY,EAAW;IAC7C,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACvC,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,MAAM,MAAM,GAAG,MAAiD,CAAC;IACjE,IAAI,MAAM,CAAC,IAAI,KAAK,MAAM;QAAE,OAAO,KAAK,CAAC;IACzC,MAAM,IAAI,GAAG,MAAM,CAAC,SAAS,CAAC;IAC9B,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3E,MAAM,OAAO,GAAI,IAA8B,CAAC,OAAO,CAAC;IACxD,MAAM,IAAI,GAAI,IAA2B,CAAC,IAAI,CAAC;IAC/C,OAAO,CACN,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAC9G,CAAC;AAAA,CACF;AAED,SAAS,iBAAiB,CAAC,IAAY,EAAuB;IAC7D,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,MAAM,UAAU,GAAa,CAAC,OAAO,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,8BAA8B,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAC5D,IAAI,MAAM,EAAE,CAAC,CAAC,CAAC;QAAE,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACnD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACrC,IAAI,KAAK,IAAI,CAAC,IAAI,GAAG,GAAG,KAAK;QAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;IAC9E,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;YACrC,IAAI,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;gBAAE,OAAO,MAAM,CAAC;QACnF,CAAC;QAAC,MAAM,CAAC;YACR,qBAAqB;QACtB,CAAC;IACF,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,SAAS,eAAe,GAAuB;IAC9C,OAAO;QACN,EAAE,EAAE,qBAAqB;QACzB,YAAY,EAAE,CAAC,UAAU,EAAE,YAAY,EAAE,aAAa,CAAC;QACvD,eAAe,EAAE,CAAC;QAClB,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACnC,CAAC;AAAA,CACF;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,OAA4B,EAA+B;IACrG,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,MAAM,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;IAC9D,MAAM,cAAc,GAAG,OAAO,CAAC,cAAc,IAAI,OAAO,CAAC;IACzD,MAAM,YAAY,GAAG,OAAO,CAAC,YAAY,IAAI,6BAA6B,CAAC;IAC3E,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC;IACpC,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,4FAA4F;IAC5F,mFAAmF;IACnF,MAAM,YAAY,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC9C,IAAI,YAAY,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC5C,MAAM,QAAQ,GAAoB,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC;QACjD,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QAChD,MAAM,MAAM,GAAG,UAAU,CAAC,YAAY,IAAI,CAAC,CAAC;QAC5C,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,IAAI,CAAC,CAAC;QACtC,IAAI,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;QAC/B,CAAC;QACD,OAAO,UAAU,CAAC;IAAA,CAClB,CAAC;IACF,MAAM,gBAAgB,GAAG,GAAuB,EAAE,CAAC;QAClD,MAAM,KAAK,GACV,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACtG,YAAY,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;QACxC,OAAO,KAAK,CAAC;IAAA,CACb,CAAC;IAEF,MAAM,QAAQ,GAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC5F,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;QACtB,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC;YAChC,KAAK,EAAE,kCAAkC,CAAC,EAAE;YAC5C,OAAO,EAAE;gBACR,oDAAoD;gBACpD,oBAAoB;gBACpB,iEAAiE;gBACjE,oDAAoD;aACpD,CAAC,IAAI,CAAC,IAAI,CAAC;YACZ,QAAQ,EAAE,eAAe,EAAE;YAC3B,MAAM,EAAE,CAAC;YACT,UAAU,EAAE,CAAC;YACb,WAAW,EAAE,CAAC;YACd,cAAc;YACd,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ;SACR,CAAC,CAAC;QACH,QAAQ,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QACnC,YAAY,IAAI,MAAM,CAAC,OAAO,CAAC;QAC/B,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW;YAAE,QAAQ,CAAC,SAAS,EAAE,CAAC;QACxD,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC,CAAC;IACjE,CAAC;IACD,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC;IACvD,QAAQ,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE9C,MAAM,MAAM,GAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC1F,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;QACtB,MAAM,OAAO,GAAG,MAAM,SAAS,CAAC;YAC/B,OAAO,EAAE;gBACR,EAAE,EAAE,kBAAkB,CAAC,EAAE;gBACzB,YAAY,EACX,oKAAoK;gBACrK,KAAK,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,UAAU,EAAE,CAAC,EAAE,UAAU,EAAE,eAAe,EAAE,OAAO,EAAE,EAAE,EAAE;gBACpG,QAAQ,EAAE,EAAE,EAAE,EAAE,eAAe,CAAC,EAAE,EAAE,YAAY,EAAE,CAAC,YAAY,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE;gBACtF,eAAe,EAAE,CAAC;aAClB;YACD,MAAM,EAAE,CAAC;YACT,cAAc;YACd,aAAa,EAAE,WAAW,CAAC,EAAE;YAC7B,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ;SACR,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QACjC,YAAY,IAAI,OAAO,CAAC,OAAO,CAAC;QAChC,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,OAAO,CAAC,QAAQ;YAAE,MAAM,CAAC,SAAS,EAAE,CAAC;QAClF,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC;IACxE,CAAC;IACD,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC;IACnD,MAAM,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE5C,MAAM,KAAK,GAAsB;QAChC,MAAM,EAAE,CAAC;QACT,gBAAgB,EAAE,CAAC;QACnB,aAAa,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,MAAM;QACpE,YAAY,EAAE,CAAC;QACf,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,MAAM;QACpE,KAAK,EAAE,YAAY,CAAC,MAAM;QAC1B,QAAQ,EAAE,EAAE;QACZ,MAAM,EAAE,CAAC;KACT,CAAC;IACF,KAAK,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;QAClC,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;QACtB,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC;YAClC,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,QAAQ,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,EAAE,UAAU,EAAE,eAAe,EAAE,OAAO,EAAE,EAAE,EAAE;YACzG,cAAc;YACd,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ;SACR,CAAC,CAAC;QACH,KAAK,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QAChC,YAAY,IAAI,MAAM,CAAC,OAAO,CAAC;QAC/B,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC;QAClC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,KAAK,CAAC,MAAM,EAAE,CAAC;YACf,sFAAsF;YACtF,mEAAiE;YACjE,IAAI,KAAK,CAAC,QAAQ,IAAI,IAAI,KAAK,OAAO;gBAAE,KAAK,CAAC,gBAAgB,EAAE,CAAC;YACjE,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,IAAI,KAAK,OAAO;gBAAE,KAAK,CAAC,YAAY,EAAE,CAAC;QAC/D,CAAC;QACD,KAAK,CAAC,QAAQ,CAAC,IAAI,CAClB,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,QAAQ,IAAI,GAAG,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,cAAc,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CACxG,CAAC;IACH,CAAC;IACD,KAAK,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5F,KAAK,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE3C,MAAM,YAAY,GAAG,KAAK,EACzB,YAAoB,EACpB,KAAwB,EACxB,OAAqD,EACzB,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;QAC/F,KAAK,MAAM,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC;YACjD,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;YACtB,0FAAwF;YACxF,MAAM,OAAO,GAAG,MAAM,oBAAoB,CAAC;gBAC1C,cAAc;gBACd,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,QAAQ,CAAC,EAAE,YAAY,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;aACzE,CAAC,CAAC;YACH,IAAI,OAAO,CAAC,UAAU;gBAAE,YAAY,IAAI,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC;YACnE,IAAI,OAAO,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;gBAC5C,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC;YACpF,CAAC;iBAAM,CAAC;gBACP,MAAM,EAAE,GAAG,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;gBACvD,IAAI,EAAE;oBAAE,KAAK,CAAC,SAAS,EAAE,CAAC;gBAC1B,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC;YACvD,CAAC;YACD,KAAK,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QACjC,CAAC;QACD,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC9E,OAAO,KAAK,CAAC;IAAA,CACb,CAAC;IAEF,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,0BAA0B,EAAE,kBAAkB,EAAE,eAAe,CAAC,CAAC;IACnG,MAAM,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAC5C,MAAM,QAAQ,GAAG,MAAM,YAAY,CAAC,6BAA6B,EAAE,qBAAqB,EAAE,aAAa,CAAC,CAAC;IACzG,QAAQ,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAC9C,MAAM,MAAM,GAAG,MAAM,YAAY,CAChC,6BAA6B,EAC7B,kBAAkB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,EAC5C,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,EAAE,kBAAkB,CAAC,SAAS,CAAE,CAAC,KAAK,CAAC,CAC5E,CAAC;IACF,MAAM,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE5C,MAAM,eAAe,GACpB,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAEtG,OAAO,EAAE,MAAM,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC;AAAA,CACpG;AAED,kGAAkG;AAClG,MAAM,UAAU,wBAAwB,CAAC,KAAa,EAAE,MAA0B,EAAU;IAC3F,MAAM,KAAK,GAAG,CAAC,eAAmC,EAAE,EAAE,CACrD,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,eAAe,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;IACpE,MAAM,KAAK,GAAG;QACb,kBAAkB,KAAK,KAAK,MAAM,CAAC,MAAM,eAAe,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC,GAAG;QACxF,oBAAoB,MAAM,CAAC,QAAQ,CAAC,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,oBAAoB,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC,KAAK,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG;QACtM,oBAAoB,MAAM,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,6BAA6B,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG;QACrM,oBAAoB,MAAM,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,sBAAsB,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE;QACvJ,oBAAoB,MAAM,CAAC,QAAQ,CAAC,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,sBAAsB,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC,EAAE;QAC/J,oBAAoB,MAAM,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,mBAAmB,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE;QACpJ,2BAA2B,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,CAAC,KAAK,uBAAuB,MAAM,CAAC,KAAK,CAAC,gBAAgB,IAAI,MAAM,CAAC,KAAK,CAAC,aAAa,mBAAmB,MAAM,CAAC,KAAK,CAAC,YAAY,IAAI,MAAM,CAAC,KAAK,CAAC,YAAY,UAAU,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,eAAe,CAAC,EAAE;QACtS,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,OAAO,EAAE,CAAC;KAC3D,CAAC;IACF,IAAI,MAAM,CAAC,YAAY,GAAG,CAAC,EAAE,CAAC;QAC7B,KAAK,CAAC,IAAI,CAAC,kBAAkB,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACxB","sourcesContent":["import { runBoundedCompletion } from \"../autonomy/bounded-completion.ts\";\nimport type { CapabilityEnvelope } from \"../autonomy/contracts.ts\";\nimport { CURATION_DIGEST_SYSTEM_PROMPT } from \"../context/brain-curator.ts\";\nimport { runWorker } from \"../delegation/worker-runner.ts\";\nimport { runRouteJudge } from \"../model-router/route-judge.ts\";\nimport { runResearch } from \"./research-runner.ts\";\n\n/**\n * Model fitness probe: measures whether a candidate model can actually drive the harness's\n * subagent contracts — the research lane, the scout-worker lane, and the routing judge — by\n * running each real runner against the model and scoring parse/success rates plus judge\n * discrimination. Provider-free: the completion executor is injected, so this works against any\n * registered model (local Ollama models included) and against faux providers in tests.\n */\n\nexport interface FitnessCompletion {\n\ttext: string;\n\tcostUsd: number;\n\tstopReason: string;\n\t/** Output tokens generated (for tok/s). Optional: providers that don't report it are skipped. */\n\toutputTokens?: number;\n\t/** Pure generation time in ms (e.g. Ollama eval_duration). Falls back to wall-clock if absent. */\n\tevalMs?: number;\n}\n\nexport type FitnessComplete = (args: {\n\tsystemPrompt: string;\n\tuserPrompt: string;\n\tsignal?: AbortSignal;\n}) => Promise<FitnessCompletion>;\n\nexport interface JudgeFitnessPrompt {\n\tprompt: string;\n\t/** True when the prompt is planning-shaped and must never route cheap. */\n\tplanning: boolean;\n}\n\n/** Default judge probe set: three planning-shaped prompts, three trivial lookups. */\nexport const DEFAULT_JUDGE_FITNESS_PROMPTS: readonly JudgeFitnessPrompt[] = [\n\t{ prompt: \"how should we plan the migration of the session storage layer?\", planning: true },\n\t{ prompt: \"design an approach for splitting the settings manager\", planning: true },\n\t{ prompt: \"draft a roadmap for the autonomy rework\", planning: true },\n\t{ prompt: \"what does the resolvePath function return?\", planning: false },\n\t{ prompt: \"list the files in the delegation module\", planning: false },\n\t{ prompt: \"why is this test flaky?\", planning: false },\n];\n\nexport interface ModelFitnessOptions {\n\tcomplete: FitnessComplete;\n\t/** Trials per lane surface. Default 3. */\n\ttrials?: number;\n\t/** Wall-clock budget per call in ms. Default 120000. */\n\tmaxWallClockMs?: number;\n\tjudgePrompts?: readonly JudgeFitnessPrompt[];\n\tsignal?: AbortSignal;\n\t/** Injected clock for latency measurement (test seam). Defaults to Date.now. */\n\tnow?: () => number;\n}\n\nexport interface LaneFitnessScore {\n\tsucceeded: number;\n\ttotal: number;\n\toutcomes: string[];\n\tmeanMs: number;\n\t/** Mean output tokens/second across the surface's calls; undefined when not reported. */\n\ttokensPerSecond?: number;\n}\n\nexport interface JudgeFitnessScore {\n\tparsed: number;\n\tplanningElevated: number;\n\tplanningTotal: number;\n\ttrivialCheap: number;\n\ttrivialTotal: number;\n\ttotal: number;\n\toutcomes: string[];\n\tmeanMs: number;\n\t/** Mean output tokens/second across the judge calls; undefined when not reported. */\n\ttokensPerSecond?: number;\n}\n\nexport interface ModelFitnessReport {\n\ttrials: number;\n\t/** Aggregate output tokens/second across ALL probe calls (the headline speed number). */\n\ttokensPerSecond?: number;\n\tresearch: LaneFitnessScore;\n\tworker: LaneFitnessScore;\n\tjudge: JudgeFitnessScore;\n\t/** Heavy-lifter surface: can the model formulate a structured search plan? */\n\tsearch: LaneFitnessScore;\n\t/** Heavy-lifter surface: can the model emit a well-formed tool call against a schema? */\n\ttoolCall: LaneFitnessScore;\n\t/** Curator surface: can the model digest a context chunk to strict JSON WITHOUT losing key facts? */\n\tdigest: LaneFitnessScore;\n\ttotalCostUsd: number;\n}\n\n/** Static prompts for the heavy-lifter surfaces (stable for provider prompt caching). */\nexport const SEARCH_PROBE_SYSTEM_PROMPT = [\n\t\"You plan code searches for a coding agent. You never answer the question yourself.\",\n\t\"Given a question about a codebase, respond with STRICT JSON only - no prose:\",\n\t'{\"queries\":[{\"pattern\":\"<regex or literal to grep>\",\"glob\":\"<file glob like **/*.ts>\"}]}',\n\t\"Return 1 to 4 queries, most specific first.\",\n].join(\"\\n\");\n\nexport const TOOL_CALL_PROBE_SYSTEM_PROMPT = [\n\t\"You operate tools for a coding agent. You have exactly one tool:\",\n\t\"grep(pattern: string, path: string) - search files under a path for a pattern.\",\n\t\"Respond to every task with STRICT JSON only - no prose:\",\n\t'{\"tool\":\"grep\",\"arguments\":{\"pattern\":\"<pattern>\",\"path\":\"<path>\"}}',\n].join(\"\\n\");\n\nconst SEARCH_PROBE_TASKS: readonly string[] = [\n\t\"Where is the retry/backoff logic for HTTP requests implemented?\",\n\t\"Which files define the settings for background research?\",\n\t\"Find where session entries of type custom are appended.\",\n];\n\n// The probe measures the REAL curation contract — same prompt the BrainCurator ships.\nexport { CURATION_DIGEST_SYSTEM_PROMPT as DIGEST_PROBE_SYSTEM_PROMPT } from \"../context/brain-curator.ts\";\n\n/**\n * Digest probe chunks each carry a NONCE identifier that cannot be guessed from the\n * instructions: acceptance requires the digest to RETAIN the nonce verbatim, so the score\n * measures extraction fidelity, not narration (a model cannot pass by paraphrasing).\n */\nconst DIGEST_PROBE_TASKS: readonly { chunk: string; nonce: string }[] = [\n\t{\n\t\tnonce: \"retryWithJitter_zx41\",\n\t\tchunk: [\n\t\t\t\"grep results for 'retry' under src/http:\",\n\t\t\t\"src/http/client.ts:88: export function retryWithJitter_zx41(fn, attempts = 3) {\",\n\t\t\t\"src/http/client.ts:112:   // exponential backoff capped at 30s\",\n\t\t\t\"src/http/pool.ts:41:   client.retry = false\",\n\t\t].join(\"\\n\"),\n\t},\n\t{\n\t\tnonce: \"ERR_QM_7734\",\n\t\tchunk: [\n\t\t\t\"$ npm run migrate\",\n\t\t\t\"migrating 14 files...\",\n\t\t\t\"error ERR_QM_7734: column 'owner_id' missing on table sessions (migration 0009)\",\n\t\t\t\"exit code 1\",\n\t\t].join(\"\\n\"),\n\t},\n\t{\n\t\tnonce: \"v3.9.2-hotfix.1\",\n\t\tchunk: [\n\t\t\t\"read package.json (34 lines):\",\n\t\t\t'  \"name\": \"acme-billing\",',\n\t\t\t'  \"version\": \"v3.9.2-hotfix.1\",',\n\t\t\t'  \"engines\": { \"node\": \">=22\" },',\n\t\t].join(\"\\n\"),\n\t},\n];\n\nfunction parseDigest(text: string, nonce: string): boolean {\n\tconst parsed = extractJsonObject(text);\n\tif (!parsed) return false;\n\tconst digest = (parsed as { digest?: unknown }).digest;\n\tif (typeof digest !== \"string\") return false;\n\tconst trimmed = digest.trim();\n\t// Bounded and faithful: short enough to be a stub annotation, still carrying the nonce fact.\n\treturn trimmed.length > 0 && trimmed.length <= 240 && trimmed.includes(nonce);\n}\n\nconst TOOL_CALL_PROBE_TASKS: readonly string[] = [\n\t\"Find usages of the function resolveCliModel under src/.\",\n\t\"Search for the string 'budget_exhausted' in the core directory.\",\n\t\"Locate where LaneTracker is instantiated under src/core.\",\n];\n\nfunction parseSearchPlan(text: string): boolean {\n\tconst parsed = extractJsonObject(text);\n\tif (!parsed) return false;\n\tconst queries = (parsed as { queries?: unknown }).queries;\n\tif (!Array.isArray(queries) || queries.length === 0 || queries.length > 8) return false;\n\treturn queries.every(\n\t\t(query) =>\n\t\t\tquery &&\n\t\t\ttypeof query === \"object\" &&\n\t\t\ttypeof (query as { pattern?: unknown }).pattern === \"string\" &&\n\t\t\t(query as { pattern: string }).pattern.trim().length > 0,\n\t);\n}\n\nfunction parseToolCall(text: string): boolean {\n\tconst parsed = extractJsonObject(text);\n\tif (!parsed) return false;\n\tconst record = parsed as { tool?: unknown; arguments?: unknown };\n\tif (record.tool !== \"grep\") return false;\n\tconst args = record.arguments;\n\tif (!args || typeof args !== \"object\" || Array.isArray(args)) return false;\n\tconst pattern = (args as { pattern?: unknown }).pattern;\n\tconst path = (args as { path?: unknown }).path;\n\treturn (\n\t\ttypeof pattern === \"string\" && pattern.trim().length > 0 && typeof path === \"string\" && path.trim().length > 0\n\t);\n}\n\nfunction extractJsonObject(text: string): unknown | undefined {\n\tconst trimmed = text.trim();\n\tconst candidates: string[] = [trimmed];\n\tconst fenced = /```(?:json)?\\s*([\\s\\S]*?)```/.exec(trimmed);\n\tif (fenced?.[1]) candidates.push(fenced[1].trim());\n\tconst start = trimmed.indexOf(\"{\");\n\tconst end = trimmed.lastIndexOf(\"}\");\n\tif (start >= 0 && end > start) candidates.push(trimmed.slice(start, end + 1));\n\tfor (const candidate of candidates) {\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(candidate);\n\t\t\tif (parsed && typeof parsed === \"object\" && !Array.isArray(parsed)) return parsed;\n\t\t} catch {\n\t\t\t// try next candidate\n\t\t}\n\t}\n\treturn undefined;\n}\n\nfunction fitnessEnvelope(): CapabilityEnvelope {\n\treturn {\n\t\tid: \"model-fitness-probe\",\n\t\tcapabilities: [\"research\", \"read_files\", \"memory_read\"],\n\t\tmaxEstimatedUsd: 1,\n\t\tcreatedAt: new Date().toISOString(),\n\t};\n}\n\nexport async function runModelFitnessProbe(options: ModelFitnessOptions): Promise<ModelFitnessReport> {\n\tconst trials = Math.max(1, Math.min(options.trials ?? 3, 20));\n\tconst maxWallClockMs = options.maxWallClockMs ?? 120_000;\n\tconst judgePrompts = options.judgePrompts ?? DEFAULT_JUDGE_FITNESS_PROMPTS;\n\tconst now = options.now ?? Date.now;\n\tlet totalCostUsd = 0;\n\n\t// Token-speed instrumentation: the lane runners' own contracts carry text/cost only, so the\n\t// completer is wrapped once here and generation stats are accumulated per surface.\n\tconst overallSpeed = { tokens: 0, evalMs: 0 };\n\tlet surfaceSpeed = { tokens: 0, evalMs: 0 };\n\tconst complete: FitnessComplete = async (args) => {\n\t\tconst completion = await options.complete(args);\n\t\tconst tokens = completion.outputTokens ?? 0;\n\t\tconst evalMs = completion.evalMs ?? 0;\n\t\tif (tokens > 0 && evalMs > 0) {\n\t\t\tsurfaceSpeed.tokens += tokens;\n\t\t\tsurfaceSpeed.evalMs += evalMs;\n\t\t\toverallSpeed.tokens += tokens;\n\t\t\toverallSpeed.evalMs += evalMs;\n\t\t}\n\t\treturn completion;\n\t};\n\tconst takeSurfaceSpeed = (): number | undefined => {\n\t\tconst speed =\n\t\t\tsurfaceSpeed.evalMs > 0 ? Math.round((surfaceSpeed.tokens / surfaceSpeed.evalMs) * 1000) : undefined;\n\t\tsurfaceSpeed = { tokens: 0, evalMs: 0 };\n\t\treturn speed;\n\t};\n\n\tconst research: LaneFitnessScore = { succeeded: 0, total: trials, outcomes: [], meanMs: 0 };\n\tfor (let i = 0; i < trials; i++) {\n\t\tconst started = now();\n\t\tconst result = await runResearch({\n\t\t\tquery: `fitness:probe requirements:req-${i}`,\n\t\t\tcontext: [\n\t\t\t\t\"Goal: add a retry helper to the HTTP client module\",\n\t\t\t\t\"Open requirements:\",\n\t\t\t\t\"- Find what retry/backoff conventions the codebase already uses\",\n\t\t\t\t\"- Identify which call sites would adopt the helper\",\n\t\t\t].join(\"\\n\"),\n\t\t\tenvelope: fitnessEnvelope(),\n\t\t\tmaxUsd: 1,\n\t\t\tmaxSources: 8,\n\t\t\tmaxFindings: 5,\n\t\t\tmaxWallClockMs,\n\t\t\tsignal: options.signal,\n\t\t\tcomplete,\n\t\t});\n\t\tresearch.meanMs += now() - started;\n\t\ttotalCostUsd += result.costUsd;\n\t\tif (result.status === \"succeeded\") research.succeeded++;\n\t\tresearch.outcomes.push(`${result.status}/${result.reasonCode}`);\n\t}\n\tresearch.meanMs = Math.round(research.meanMs / trials);\n\tresearch.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst worker: LaneFitnessScore = { succeeded: 0, total: trials, outcomes: [], meanMs: 0 };\n\tfor (let i = 0; i < trials; i++) {\n\t\tconst started = now();\n\t\tconst outcome = await runWorker({\n\t\t\trequest: {\n\t\t\t\tid: `fitness-worker-${i}`,\n\t\t\t\tinstructions:\n\t\t\t\t\t\"Summarize in two sentences what a capability envelope is: a declared set of allowed tools, paths, and capability names that bounds what a delegated worker may do.\",\n\t\t\t\troute: { tier: \"cheap\", risk: \"read-only\", confidence: 1, reasonCode: \"fitness_probe\", reasons: [] },\n\t\t\t\tenvelope: { id: `fitness-env-${i}`, capabilities: [\"read_files\"], maxEstimatedUsd: 1 },\n\t\t\t\tmaxEstimatedUsd: 1,\n\t\t\t},\n\t\t\tmaxUsd: 1,\n\t\t\tmaxWallClockMs,\n\t\t\tusageReportId: `fitness:${i}`,\n\t\t\tsignal: options.signal,\n\t\t\tcomplete,\n\t\t});\n\t\tworker.meanMs += now() - started;\n\t\ttotalCostUsd += outcome.costUsd;\n\t\tif (outcome.result.status === \"completed\" && outcome.accepted) worker.succeeded++;\n\t\tworker.outcomes.push(`${outcome.result.status}/${outcome.reasonCode}`);\n\t}\n\tworker.meanMs = Math.round(worker.meanMs / trials);\n\tworker.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst judge: JudgeFitnessScore = {\n\t\tparsed: 0,\n\t\tplanningElevated: 0,\n\t\tplanningTotal: judgePrompts.filter((entry) => entry.planning).length,\n\t\ttrivialCheap: 0,\n\t\ttrivialTotal: judgePrompts.filter((entry) => !entry.planning).length,\n\t\ttotal: judgePrompts.length,\n\t\toutcomes: [],\n\t\tmeanMs: 0,\n\t};\n\tfor (const entry of judgePrompts) {\n\t\tconst started = now();\n\t\tconst result = await runRouteJudge({\n\t\t\tprompt: entry.prompt,\n\t\t\tbaseline: { tier: \"cheap\", risk: \"read-only\", confidence: 0.5, reasonCode: \"fitness_probe\", reasons: [] },\n\t\t\tmaxWallClockMs,\n\t\t\tsignal: options.signal,\n\t\t\tcomplete,\n\t\t});\n\t\tjudge.meanMs += now() - started;\n\t\ttotalCostUsd += result.costUsd;\n\t\tconst tier = result.decision.tier;\n\t\tif (result.verdict) {\n\t\t\tjudge.parsed++;\n\t\t\t// A useful judge must both keep planning off the cheap tier AND actually send trivial\n\t\t\t// prompts there — all-medium verdicts are safe but save nothing.\n\t\t\tif (entry.planning && tier !== \"cheap\") judge.planningElevated++;\n\t\t\tif (!entry.planning && tier === \"cheap\") judge.trivialCheap++;\n\t\t}\n\t\tjudge.outcomes.push(\n\t\t\t`\"${entry.prompt.slice(0, 40)}\" -> ${tier}${result.fallbackReason ? ` (${result.fallbackReason})` : \"\"}`,\n\t\t);\n\t}\n\tjudge.meanMs = judgePrompts.length > 0 ? Math.round(judge.meanMs / judgePrompts.length) : 0;\n\tjudge.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst probeSurface = async (\n\t\tsystemPrompt: string,\n\t\ttasks: readonly string[],\n\t\taccepts: (text: string, taskIndex: number) => boolean,\n\t): Promise<LaneFitnessScore> => {\n\t\tconst score: LaneFitnessScore = { succeeded: 0, total: tasks.length, outcomes: [], meanMs: 0 };\n\t\tfor (const [taskIndex, task] of tasks.entries()) {\n\t\t\tconst started = now();\n\t\t\t// Same wall-clock envelope as the lane surfaces — a hung model must not hang the probe.\n\t\t\tconst bounded = await runBoundedCompletion({\n\t\t\t\tmaxWallClockMs,\n\t\t\t\tsignal: options.signal,\n\t\t\t\texecute: (signal) => complete({ systemPrompt, userPrompt: task, signal }),\n\t\t\t});\n\t\t\tif (bounded.completion) totalCostUsd += bounded.completion.costUsd;\n\t\t\tif (bounded.failure || !bounded.completion) {\n\t\t\t\tscore.outcomes.push(bounded.failure ? bounded.failure.status : \"completion_error\");\n\t\t\t} else {\n\t\t\t\tconst ok = accepts(bounded.completion.text, taskIndex);\n\t\t\t\tif (ok) score.succeeded++;\n\t\t\t\tscore.outcomes.push(ok ? \"ok\" : \"unparseable_output\");\n\t\t\t}\n\t\t\tscore.meanMs += now() - started;\n\t\t}\n\t\tscore.meanMs = tasks.length > 0 ? Math.round(score.meanMs / tasks.length) : 0;\n\t\treturn score;\n\t};\n\n\tconst search = await probeSurface(SEARCH_PROBE_SYSTEM_PROMPT, SEARCH_PROBE_TASKS, parseSearchPlan);\n\tsearch.tokensPerSecond = takeSurfaceSpeed();\n\tconst toolCall = await probeSurface(TOOL_CALL_PROBE_SYSTEM_PROMPT, TOOL_CALL_PROBE_TASKS, parseToolCall);\n\ttoolCall.tokensPerSecond = takeSurfaceSpeed();\n\tconst digest = await probeSurface(\n\t\tCURATION_DIGEST_SYSTEM_PROMPT,\n\t\tDIGEST_PROBE_TASKS.map((task) => task.chunk),\n\t\t(text, taskIndex) => parseDigest(text, DIGEST_PROBE_TASKS[taskIndex]!.nonce),\n\t);\n\tdigest.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst tokensPerSecond =\n\t\toverallSpeed.evalMs > 0 ? Math.round((overallSpeed.tokens / overallSpeed.evalMs) * 1000) : undefined;\n\n\treturn { trials, tokensPerSecond, research, worker, judge, search, toolCall, digest, totalCostUsd };\n}\n\n/** Compact human-readable report for tool output / interactive display. Bounded, no raw dumps. */\nexport function formatModelFitnessReport(model: string, report: ModelFitnessReport): string {\n\tconst speed = (tokensPerSecond: number | undefined) =>\n\t\ttokensPerSecond !== undefined ? `, ~${tokensPerSecond} tok/s` : \"\";\n\tconst lines = [\n\t\t`Model fitness: ${model} (${report.trials} trials/lane${speed(report.tokensPerSecond)})`,\n\t\t`- research lane: ${report.research.succeeded}/${report.research.total} succeeded, mean ${report.research.meanMs}ms${speed(report.research.tokensPerSecond)} [${report.research.outcomes.join(\", \")}]`,\n\t\t`- worker lane:   ${report.worker.succeeded}/${report.worker.total} completed+accepted, mean ${report.worker.meanMs}ms${speed(report.worker.tokensPerSecond)} [${report.worker.outcomes.join(\", \")}]`,\n\t\t`- search plans:  ${report.search.succeeded}/${report.search.total} well-formed, mean ${report.search.meanMs}ms${speed(report.search.tokensPerSecond)}`,\n\t\t`- tool calls:    ${report.toolCall.succeeded}/${report.toolCall.total} well-formed, mean ${report.toolCall.meanMs}ms${speed(report.toolCall.tokensPerSecond)}`,\n\t\t`- digests:       ${report.digest.succeeded}/${report.digest.total} faithful, mean ${report.digest.meanMs}ms${speed(report.digest.tokensPerSecond)}`,\n\t\t`- route judge:   parsed ${report.judge.parsed}/${report.judge.total}, planning-elevated ${report.judge.planningElevated}/${report.judge.planningTotal}, trivial-cheap ${report.judge.trivialCheap}/${report.judge.trivialTotal}, mean ${report.judge.meanMs}ms${speed(report.judge.tokensPerSecond)}`,\n\t\t...report.judge.outcomes.map((outcome) => `    ${outcome}`),\n\t];\n\tif (report.totalCostUsd > 0) {\n\t\tlines.push(`- probe cost: $${report.totalCostUsd.toFixed(4)}`);\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}
+{"version":3,"file":"model-fitness.js","sourceRoot":"","sources":["../../../src/core/research/model-fitness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,oBAAoB,EAAE,MAAM,mCAAmC,CAAC;AAEzE,OAAO,EAAE,6BAA6B,EAAE,MAAM,6BAA6B,CAAC;AAC5E,OAAO,EAAE,SAAS,EAAE,MAAM,gCAAgC,CAAC;AAC3D,OAAO,EAAE,aAAa,EAAE,MAAM,gCAAgC,CAAC;AAC/D,OAAO,EAAE,WAAW,EAAE,MAAM,sBAAsB,CAAC;AAgCnD,qFAAqF;AACrF,MAAM,CAAC,MAAM,6BAA6B,GAAkC;IAC3E,EAAE,MAAM,EAAE,gEAAgE,EAAE,QAAQ,EAAE,IAAI,EAAE;IAC5F,EAAE,MAAM,EAAE,uDAAuD,EAAE,QAAQ,EAAE,IAAI,EAAE;IACnF,EAAE,MAAM,EAAE,yCAAyC,EAAE,QAAQ,EAAE,IAAI,EAAE;IACrE,EAAE,MAAM,EAAE,4CAA4C,EAAE,QAAQ,EAAE,KAAK,EAAE;IACzE,EAAE,MAAM,EAAE,yCAAyC,EAAE,QAAQ,EAAE,KAAK,EAAE;IACtE,EAAE,MAAM,EAAE,yBAAyB,EAAE,QAAQ,EAAE,KAAK,EAAE;CACtD,CAAC;AAoDF,yFAAyF;AACzF,MAAM,CAAC,MAAM,0BAA0B,GAAG;IACzC,oFAAoF;IACpF,8EAA8E;IAC9E,0FAA0F;IAC1F,6CAA6C;CAC7C,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,CAAC,MAAM,6BAA6B,GAAG;IAC5C,kEAAkE;IAClE,gFAAgF;IAChF,yDAAyD;IACzD,qEAAqE;CACrE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAEb,MAAM,kBAAkB,GAAsB;IAC7C,iEAAiE;IACjE,0DAA0D;IAC1D,yDAAyD;CACzD,CAAC;AAEF,wFAAsF;AACtF,OAAO,EAAE,6BAA6B,IAAI,0BAA0B,EAAE,MAAM,6BAA6B,CAAC;AAE1G;;;;GAIG;AACH,MAAM,kBAAkB,GAAgD;IACvE;QACC,KAAK,EAAE,sBAAsB;QAC7B,KAAK,EAAE;YACN,0CAA0C;YAC1C,iFAAiF;YACjF,gEAAgE;YAChE,6CAA6C;SAC7C,CAAC,IAAI,CAAC,IAAI,CAAC;KACZ;IACD;QACC,KAAK,EAAE,aAAa;QACpB,KAAK,EAAE;YACN,mBAAmB;YACnB,uBAAuB;YACvB,iFAAiF;YACjF,aAAa;SACb,CAAC,IAAI,CAAC,IAAI,CAAC;KACZ;IACD;QACC,KAAK,EAAE,iBAAiB;QACxB,KAAK,EAAE;YACN,+BAA+B;YAC/B,2BAA2B;YAC3B,iCAAiC;YACjC,kCAAkC;SAClC,CAAC,IAAI,CAAC,IAAI,CAAC;KACZ;CACD,CAAC;AAEF,SAAS,WAAW,CAAC,IAAY,EAAE,KAAa,EAAW;IAC1D,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACvC,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,MAAM,MAAM,GAAI,MAA+B,CAAC,MAAM,CAAC;IACvD,IAAI,OAAO,MAAM,KAAK,QAAQ;QAAE,OAAO,KAAK,CAAC;IAC7C,MAAM,OAAO,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;IAC9B,6FAA6F;IAC7F,OAAO,OAAO,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,CAAC,MAAM,IAAI,GAAG,IAAI,OAAO,CAAC,QAAQ,CAAC,KAAK,CAAC,CAAC;AAAA,CAC9E;AAED,MAAM,qBAAqB,GAAsB;IAChD,yDAAyD;IACzD,iEAAiE;IACjE,0DAA0D;CAC1D,CAAC;AAEF,SAAS,eAAe,CAAC,IAAY,EAAW;IAC/C,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACvC,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,MAAM,OAAO,GAAI,MAAgC,CAAC,OAAO,CAAC;IAC1D,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,OAAO,CAAC,IAAI,OAAO,CAAC,MAAM,KAAK,CAAC,IAAI,OAAO,CAAC,MAAM,GAAG,CAAC;QAAE,OAAO,KAAK,CAAC;IACxF,OAAO,OAAO,CAAC,KAAK,CACnB,CAAC,KAAK,EAAE,EAAE,CACT,KAAK;QACL,OAAO,KAAK,KAAK,QAAQ;QACzB,OAAQ,KAA+B,CAAC,OAAO,KAAK,QAAQ;QAC3D,KAA6B,CAAC,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CACzD,CAAC;AAAA,CACF;AAED,SAAS,aAAa,CAAC,IAAY,EAAW;IAC7C,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC;IACvC,IAAI,CAAC,MAAM;QAAE,OAAO,KAAK,CAAC;IAC1B,MAAM,MAAM,GAAG,MAAiD,CAAC;IACjE,IAAI,MAAM,CAAC,IAAI,KAAK,MAAM;QAAE,OAAO,KAAK,CAAC;IACzC,MAAM,IAAI,GAAG,MAAM,CAAC,SAAS,CAAC;IAC9B,IAAI,CAAC,IAAI,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,IAAI,CAAC;QAAE,OAAO,KAAK,CAAC;IAC3E,MAAM,OAAO,GAAI,IAA8B,CAAC,OAAO,CAAC;IACxD,MAAM,IAAI,GAAI,IAA2B,CAAC,IAAI,CAAC;IAC/C,OAAO,CACN,OAAO,OAAO,KAAK,QAAQ,IAAI,OAAO,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,IAAI,OAAO,IAAI,KAAK,QAAQ,IAAI,IAAI,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CAC9G,CAAC;AAAA,CACF;AAED,SAAS,iBAAiB,CAAC,IAAY,EAAuB;IAC7D,MAAM,OAAO,GAAG,IAAI,CAAC,IAAI,EAAE,CAAC;IAC5B,MAAM,UAAU,GAAa,CAAC,OAAO,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,8BAA8B,CAAC,IAAI,CAAC,OAAO,CAAC,CAAC;IAC5D,IAAI,MAAM,EAAE,CAAC,CAAC,CAAC;QAAE,UAAU,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,IAAI,EAAE,CAAC,CAAC;IACnD,MAAM,KAAK,GAAG,OAAO,CAAC,OAAO,CAAC,GAAG,CAAC,CAAC;IACnC,MAAM,GAAG,GAAG,OAAO,CAAC,WAAW,CAAC,GAAG,CAAC,CAAC;IACrC,IAAI,KAAK,IAAI,CAAC,IAAI,GAAG,GAAG,KAAK;QAAE,UAAU,CAAC,IAAI,CAAC,OAAO,CAAC,KAAK,CAAC,KAAK,EAAE,GAAG,GAAG,CAAC,CAAC,CAAC,CAAC;IAC9E,KAAK,MAAM,SAAS,IAAI,UAAU,EAAE,CAAC;QACpC,IAAI,CAAC;YACJ,MAAM,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,SAAS,CAAC,CAAC;YACrC,IAAI,MAAM,IAAI,OAAO,MAAM,KAAK,QAAQ,IAAI,CAAC,KAAK,CAAC,OAAO,CAAC,MAAM,CAAC;gBAAE,OAAO,MAAM,CAAC;QACnF,CAAC;QAAC,MAAM,CAAC;YACR,qBAAqB;QACtB,CAAC;IACF,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB;AAED,SAAS,eAAe,GAAuB;IAC9C,OAAO;QACN,EAAE,EAAE,qBAAqB;QACzB,YAAY,EAAE,CAAC,UAAU,EAAE,YAAY,EAAE,aAAa,CAAC;QACvD,eAAe,EAAE,CAAC;QAClB,SAAS,EAAE,IAAI,IAAI,EAAE,CAAC,WAAW,EAAE;KACnC,CAAC;AAAA,CACF;AAED,MAAM,CAAC,KAAK,UAAU,oBAAoB,CAAC,OAA4B,EAA+B;IACrG,MAAM,MAAM,GAAG,IAAI,CAAC,GAAG,CAAC,CAAC,EAAE,IAAI,CAAC,GAAG,CAAC,OAAO,CAAC,MAAM,IAAI,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC;IAC9D,MAAM,cAAc,GAAG,OAAO,CAAC,cAAc,IAAI,OAAO,CAAC;IACzD,MAAM,YAAY,GAAG,OAAO,CAAC,YAAY,IAAI,6BAA6B,CAAC;IAC3E,MAAM,GAAG,GAAG,OAAO,CAAC,GAAG,IAAI,IAAI,CAAC,GAAG,CAAC;IACpC,IAAI,YAAY,GAAG,CAAC,CAAC;IAErB,4FAA4F;IAC5F,mFAAmF;IACnF,MAAM,YAAY,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC9C,IAAI,YAAY,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC5C,MAAM,QAAQ,GAAoB,KAAK,EAAE,IAAI,EAAE,EAAE,CAAC;QACjD,MAAM,UAAU,GAAG,MAAM,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC;QAChD,MAAM,MAAM,GAAG,UAAU,CAAC,YAAY,IAAI,CAAC,CAAC;QAC5C,MAAM,MAAM,GAAG,UAAU,CAAC,MAAM,IAAI,CAAC,CAAC;QACtC,IAAI,MAAM,GAAG,CAAC,IAAI,MAAM,GAAG,CAAC,EAAE,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;YAC9B,YAAY,CAAC,MAAM,IAAI,MAAM,CAAC;QAC/B,CAAC;QACD,OAAO,UAAU,CAAC;IAAA,CAClB,CAAC;IACF,MAAM,gBAAgB,GAAG,GAAuB,EAAE,CAAC;QAClD,MAAM,KAAK,GACV,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;QACtG,YAAY,GAAG,EAAE,MAAM,EAAE,CAAC,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;QACxC,OAAO,KAAK,CAAC;IAAA,CACb,CAAC;IAEF,MAAM,QAAQ,GAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC5F,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;QACtB,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC;YAChC,KAAK,EAAE,kCAAkC,CAAC,EAAE;YAC5C,OAAO,EAAE;gBACR,oDAAoD;gBACpD,oBAAoB;gBACpB,iEAAiE;gBACjE,oDAAoD;aACpD,CAAC,IAAI,CAAC,IAAI,CAAC;YACZ,QAAQ,EAAE,eAAe,EAAE;YAC3B,MAAM,EAAE,CAAC;YACT,UAAU,EAAE,CAAC;YACb,WAAW,EAAE,CAAC;YACd,cAAc;YACd,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ;SACR,CAAC,CAAC;QACH,QAAQ,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QACnC,YAAY,IAAI,MAAM,CAAC,OAAO,CAAC;QAC/B,IAAI,MAAM,CAAC,MAAM,KAAK,WAAW;YAAE,QAAQ,CAAC,SAAS,EAAE,CAAC;QACxD,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,MAAM,IAAI,MAAM,CAAC,UAAU,EAAE,CAAC,CAAC;IACjE,CAAC;IACD,QAAQ,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,QAAQ,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC;IACvD,QAAQ,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE9C,MAAM,MAAM,GAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;IAC1F,KAAK,IAAI,CAAC,GAAG,CAAC,EAAE,CAAC,GAAG,MAAM,EAAE,CAAC,EAAE,EAAE,CAAC;QACjC,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;QACtB,MAAM,OAAO,GAAG,MAAM,SAAS,CAAC;YAC/B,OAAO,EAAE;gBACR,EAAE,EAAE,kBAAkB,CAAC,EAAE;gBACzB,YAAY,EACX,oKAAoK;gBACrK,KAAK,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,UAAU,EAAE,CAAC,EAAE,UAAU,EAAE,eAAe,EAAE,OAAO,EAAE,EAAE,EAAE;gBACpG,QAAQ,EAAE,EAAE,EAAE,EAAE,eAAe,CAAC,EAAE,EAAE,YAAY,EAAE,CAAC,YAAY,CAAC,EAAE,eAAe,EAAE,CAAC,EAAE;gBACtF,eAAe,EAAE,CAAC;aAClB;YACD,MAAM,EAAE,CAAC;YACT,cAAc;YACd,aAAa,EAAE,WAAW,CAAC,EAAE;YAC7B,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ;SACR,CAAC,CAAC;QACH,MAAM,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QACjC,YAAY,IAAI,OAAO,CAAC,OAAO,CAAC;QAChC,IAAI,OAAO,CAAC,MAAM,CAAC,MAAM,KAAK,WAAW,IAAI,OAAO,CAAC,QAAQ;YAAE,MAAM,CAAC,SAAS,EAAE,CAAC;QAClF,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,OAAO,CAAC,MAAM,CAAC,MAAM,IAAI,OAAO,CAAC,UAAU,EAAE,CAAC,CAAC;IACxE,CAAC;IACD,MAAM,CAAC,MAAM,GAAG,IAAI,CAAC,KAAK,CAAC,MAAM,CAAC,MAAM,GAAG,MAAM,CAAC,CAAC;IACnD,MAAM,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE5C,MAAM,KAAK,GAAsB;QAChC,MAAM,EAAE,CAAC;QACT,gBAAgB,EAAE,CAAC;QACnB,aAAa,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,MAAM;QACpE,YAAY,EAAE,CAAC;QACf,YAAY,EAAE,YAAY,CAAC,MAAM,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,QAAQ,CAAC,CAAC,MAAM;QACpE,KAAK,EAAE,YAAY,CAAC,MAAM;QAC1B,QAAQ,EAAE,EAAE;QACZ,MAAM,EAAE,CAAC;KACT,CAAC;IACF,KAAK,MAAM,KAAK,IAAI,YAAY,EAAE,CAAC;QAClC,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;QACtB,MAAM,MAAM,GAAG,MAAM,aAAa,CAAC;YAClC,MAAM,EAAE,KAAK,CAAC,MAAM;YACpB,QAAQ,EAAE,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,WAAW,EAAE,UAAU,EAAE,GAAG,EAAE,UAAU,EAAE,eAAe,EAAE,OAAO,EAAE,EAAE,EAAE;YACzG,cAAc;YACd,MAAM,EAAE,OAAO,CAAC,MAAM;YACtB,QAAQ;SACR,CAAC,CAAC;QACH,KAAK,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QAChC,YAAY,IAAI,MAAM,CAAC,OAAO,CAAC;QAC/B,MAAM,IAAI,GAAG,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC;QAClC,IAAI,MAAM,CAAC,OAAO,EAAE,CAAC;YACpB,KAAK,CAAC,MAAM,EAAE,CAAC;YACf,sFAAsF;YACtF,mEAAiE;YACjE,IAAI,KAAK,CAAC,QAAQ,IAAI,IAAI,KAAK,OAAO;gBAAE,KAAK,CAAC,gBAAgB,EAAE,CAAC;YACjE,IAAI,CAAC,KAAK,CAAC,QAAQ,IAAI,IAAI,KAAK,OAAO;gBAAE,KAAK,CAAC,YAAY,EAAE,CAAC;QAC/D,CAAC;QACD,KAAK,CAAC,QAAQ,CAAC,IAAI,CAClB,IAAI,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC,EAAE,EAAE,CAAC,QAAQ,IAAI,GAAG,MAAM,CAAC,cAAc,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,cAAc,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CACxG,CAAC;IACH,CAAC;IACD,KAAK,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;IAC5F,KAAK,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE3C,MAAM,YAAY,GAAG,KAAK,EACzB,YAAoB,EACpB,KAAwB,EACxB,OAAqD,EACzB,EAAE,CAAC;QAC/B,MAAM,KAAK,GAAqB,EAAE,SAAS,EAAE,CAAC,EAAE,KAAK,EAAE,KAAK,CAAC,MAAM,EAAE,QAAQ,EAAE,EAAE,EAAE,MAAM,EAAE,CAAC,EAAE,CAAC;QAC/F,KAAK,MAAM,CAAC,SAAS,EAAE,IAAI,CAAC,IAAI,KAAK,CAAC,OAAO,EAAE,EAAE,CAAC;YACjD,MAAM,OAAO,GAAG,GAAG,EAAE,CAAC;YACtB,0FAAwF;YACxF,MAAM,OAAO,GAAG,MAAM,oBAAoB,CAAC;gBAC1C,cAAc;gBACd,MAAM,EAAE,OAAO,CAAC,MAAM;gBACtB,OAAO,EAAE,CAAC,MAAM,EAAE,EAAE,CAAC,QAAQ,CAAC,EAAE,YAAY,EAAE,UAAU,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC;aACzE,CAAC,CAAC;YACH,IAAI,OAAO,CAAC,UAAU;gBAAE,YAAY,IAAI,OAAO,CAAC,UAAU,CAAC,OAAO,CAAC;YACnE,IAAI,OAAO,CAAC,OAAO,IAAI,CAAC,OAAO,CAAC,UAAU,EAAE,CAAC;gBAC5C,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC,CAAC,OAAO,CAAC,OAAO,CAAC,MAAM,CAAC,CAAC,CAAC,kBAAkB,CAAC,CAAC;YACpF,CAAC;iBAAM,CAAC;gBACP,MAAM,EAAE,GAAG,OAAO,CAAC,OAAO,CAAC,UAAU,CAAC,IAAI,EAAE,SAAS,CAAC,CAAC;gBACvD,IAAI,EAAE;oBAAE,KAAK,CAAC,SAAS,EAAE,CAAC;gBAC1B,KAAK,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,oBAAoB,CAAC,CAAC;YACvD,CAAC;YACD,KAAK,CAAC,MAAM,IAAI,GAAG,EAAE,GAAG,OAAO,CAAC;QACjC,CAAC;QACD,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,KAAK,CAAC,MAAM,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,CAAC,CAAC,CAAC;QAC9E,OAAO,KAAK,CAAC;IAAA,CACb,CAAC;IAEF,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC,0BAA0B,EAAE,kBAAkB,EAAE,eAAe,CAAC,CAAC;IACnG,MAAM,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAC5C,MAAM,QAAQ,GAAG,MAAM,YAAY,CAAC,6BAA6B,EAAE,qBAAqB,EAAE,aAAa,CAAC,CAAC;IACzG,QAAQ,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAC9C,MAAM,MAAM,GAAG,MAAM,YAAY,CAChC,6BAA6B,EAC7B,kBAAkB,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,CAAC,EAC5C,CAAC,IAAI,EAAE,SAAS,EAAE,EAAE,CAAC,WAAW,CAAC,IAAI,EAAE,kBAAkB,CAAC,SAAS,CAAE,CAAC,KAAK,CAAC,CAC5E,CAAC;IACF,MAAM,CAAC,eAAe,GAAG,gBAAgB,EAAE,CAAC;IAE5C,MAAM,eAAe,GACpB,YAAY,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,YAAY,CAAC,MAAM,GAAG,YAAY,CAAC,MAAM,CAAC,GAAG,IAAI,CAAC,CAAC,CAAC,CAAC,SAAS,CAAC;IAEtG,OAAO,EAAE,MAAM,EAAE,eAAe,EAAE,QAAQ,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,EAAE,QAAQ,EAAE,MAAM,EAAE,YAAY,EAAE,CAAC;AAAA,CACpG;AAED;;;;;;;;;;GAUG;AACH,MAAM,UAAU,gBAAgB,CAAC,MAA0B,EAAW;IACrE,MAAM,KAAK,GAAG,CAAC,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,QAAQ,EAAE,MAAM,CAAC,MAAM,CAAC,CAAC;IAC9F,MAAM,WAAW,GAAG,KAAK,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;IAC3D,MAAM,WAAW,GAAG,MAAM,CAAC,KAAK,CAAC,KAAK,GAAG,CAAC,CAAC;IAC3C,IAAI,WAAW,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,WAAW;QAAE,OAAO,KAAK,CAAC;IAC3D,MAAM,cAAc,GAAG,WAAW,CAAC,MAAM,GAAG,CAAC,IAAI,WAAW,CAAC,KAAK,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,SAAS,KAAK,CAAC,CAAC,CAAC;IACnG,MAAM,WAAW,GAAG,CAAC,WAAW,IAAI,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,CAAC;IAC9D,OAAO,cAAc,IAAI,WAAW,CAAC;AAAA,CACrC;AAED,kGAAkG;AAClG,MAAM,UAAU,wBAAwB,CAAC,KAAa,EAAE,MAA0B,EAAU;IAC3F,MAAM,KAAK,GAAG,CAAC,eAAmC,EAAE,EAAE,CACrD,eAAe,KAAK,SAAS,CAAC,CAAC,CAAC,MAAM,eAAe,QAAQ,CAAC,CAAC,CAAC,EAAE,CAAC;IACpE,MAAM,KAAK,GAAG;QACb,kBAAkB,KAAK,KAAK,MAAM,CAAC,MAAM,eAAe,KAAK,CAAC,MAAM,CAAC,eAAe,CAAC,GAAG;QACxF,oBAAoB,MAAM,CAAC,QAAQ,CAAC,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,oBAAoB,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC,KAAK,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG;QACtM,oBAAoB,MAAM,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,6BAA6B,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,KAAK,MAAM,CAAC,MAAM,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,GAAG;QACrM,oBAAoB,MAAM,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,sBAAsB,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE;QACvJ,oBAAoB,MAAM,CAAC,QAAQ,CAAC,SAAS,IAAI,MAAM,CAAC,QAAQ,CAAC,KAAK,sBAAsB,MAAM,CAAC,QAAQ,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,QAAQ,CAAC,eAAe,CAAC,EAAE;QAC/J,oBAAoB,MAAM,CAAC,MAAM,CAAC,SAAS,IAAI,MAAM,CAAC,MAAM,CAAC,KAAK,mBAAmB,MAAM,CAAC,MAAM,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,MAAM,CAAC,eAAe,CAAC,EAAE;QACpJ,2BAA2B,MAAM,CAAC,KAAK,CAAC,MAAM,IAAI,MAAM,CAAC,KAAK,CAAC,KAAK,uBAAuB,MAAM,CAAC,KAAK,CAAC,gBAAgB,IAAI,MAAM,CAAC,KAAK,CAAC,aAAa,mBAAmB,MAAM,CAAC,KAAK,CAAC,YAAY,IAAI,MAAM,CAAC,KAAK,CAAC,YAAY,UAAU,MAAM,CAAC,KAAK,CAAC,MAAM,KAAK,KAAK,CAAC,MAAM,CAAC,KAAK,CAAC,eAAe,CAAC,EAAE;QACtS,GAAG,MAAM,CAAC,KAAK,CAAC,QAAQ,CAAC,GAAG,CAAC,CAAC,OAAO,EAAE,EAAE,CAAC,OAAO,OAAO,EAAE,CAAC;KAC3D,CAAC;IACF,IAAI,MAAM,CAAC,YAAY,GAAG,CAAC,EAAE,CAAC;QAC7B,KAAK,CAAC,IAAI,CAAC,kBAAkB,MAAM,CAAC,YAAY,CAAC,OAAO,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAAA,CACxB","sourcesContent":["import { runBoundedCompletion } from \"../autonomy/bounded-completion.ts\";\nimport type { CapabilityEnvelope } from \"../autonomy/contracts.ts\";\nimport { CURATION_DIGEST_SYSTEM_PROMPT } from \"../context/brain-curator.ts\";\nimport { runWorker } from \"../delegation/worker-runner.ts\";\nimport { runRouteJudge } from \"../model-router/route-judge.ts\";\nimport { runResearch } from \"./research-runner.ts\";\n\n/**\n * Model fitness probe: measures whether a candidate model can actually drive the harness's\n * subagent contracts — the research lane, the scout-worker lane, and the routing judge — by\n * running each real runner against the model and scoring parse/success rates plus judge\n * discrimination. Provider-free: the completion executor is injected, so this works against any\n * registered model (local Ollama models included) and against faux providers in tests.\n */\n\nexport interface FitnessCompletion {\n\ttext: string;\n\tcostUsd: number;\n\tstopReason: string;\n\t/** Output tokens generated (for tok/s). Optional: providers that don't report it are skipped. */\n\toutputTokens?: number;\n\t/** Pure generation time in ms (e.g. Ollama eval_duration). Falls back to wall-clock if absent. */\n\tevalMs?: number;\n}\n\nexport type FitnessComplete = (args: {\n\tsystemPrompt: string;\n\tuserPrompt: string;\n\tsignal?: AbortSignal;\n}) => Promise<FitnessCompletion>;\n\nexport interface JudgeFitnessPrompt {\n\tprompt: string;\n\t/** True when the prompt is planning-shaped and must never route cheap. */\n\tplanning: boolean;\n}\n\n/** Default judge probe set: three planning-shaped prompts, three trivial lookups. */\nexport const DEFAULT_JUDGE_FITNESS_PROMPTS: readonly JudgeFitnessPrompt[] = [\n\t{ prompt: \"how should we plan the migration of the session storage layer?\", planning: true },\n\t{ prompt: \"design an approach for splitting the settings manager\", planning: true },\n\t{ prompt: \"draft a roadmap for the autonomy rework\", planning: true },\n\t{ prompt: \"what does the resolvePath function return?\", planning: false },\n\t{ prompt: \"list the files in the delegation module\", planning: false },\n\t{ prompt: \"why is this test flaky?\", planning: false },\n];\n\nexport interface ModelFitnessOptions {\n\tcomplete: FitnessComplete;\n\t/** Trials per lane surface. Default 3. */\n\ttrials?: number;\n\t/** Wall-clock budget per call in ms. Default 120000. */\n\tmaxWallClockMs?: number;\n\tjudgePrompts?: readonly JudgeFitnessPrompt[];\n\tsignal?: AbortSignal;\n\t/** Injected clock for latency measurement (test seam). Defaults to Date.now. */\n\tnow?: () => number;\n}\n\nexport interface LaneFitnessScore {\n\tsucceeded: number;\n\ttotal: number;\n\toutcomes: string[];\n\tmeanMs: number;\n\t/** Mean output tokens/second across the surface's calls; undefined when not reported. */\n\ttokensPerSecond?: number;\n}\n\nexport interface JudgeFitnessScore {\n\tparsed: number;\n\tplanningElevated: number;\n\tplanningTotal: number;\n\ttrivialCheap: number;\n\ttrivialTotal: number;\n\ttotal: number;\n\toutcomes: string[];\n\tmeanMs: number;\n\t/** Mean output tokens/second across the judge calls; undefined when not reported. */\n\ttokensPerSecond?: number;\n}\n\nexport interface ModelFitnessReport {\n\ttrials: number;\n\t/** Aggregate output tokens/second across ALL probe calls (the headline speed number). */\n\ttokensPerSecond?: number;\n\tresearch: LaneFitnessScore;\n\tworker: LaneFitnessScore;\n\tjudge: JudgeFitnessScore;\n\t/** Heavy-lifter surface: can the model formulate a structured search plan? */\n\tsearch: LaneFitnessScore;\n\t/** Heavy-lifter surface: can the model emit a well-formed tool call against a schema? */\n\ttoolCall: LaneFitnessScore;\n\t/** Curator surface: can the model digest a context chunk to strict JSON WITHOUT losing key facts? */\n\tdigest: LaneFitnessScore;\n\ttotalCostUsd: number;\n}\n\n/** Static prompts for the heavy-lifter surfaces (stable for provider prompt caching). */\nexport const SEARCH_PROBE_SYSTEM_PROMPT = [\n\t\"You plan code searches for a coding agent. You never answer the question yourself.\",\n\t\"Given a question about a codebase, respond with STRICT JSON only - no prose:\",\n\t'{\"queries\":[{\"pattern\":\"<regex or literal to grep>\",\"glob\":\"<file glob like **/*.ts>\"}]}',\n\t\"Return 1 to 4 queries, most specific first.\",\n].join(\"\\n\");\n\nexport const TOOL_CALL_PROBE_SYSTEM_PROMPT = [\n\t\"You operate tools for a coding agent. You have exactly one tool:\",\n\t\"grep(pattern: string, path: string) - search files under a path for a pattern.\",\n\t\"Respond to every task with STRICT JSON only - no prose:\",\n\t'{\"tool\":\"grep\",\"arguments\":{\"pattern\":\"<pattern>\",\"path\":\"<path>\"}}',\n].join(\"\\n\");\n\nconst SEARCH_PROBE_TASKS: readonly string[] = [\n\t\"Where is the retry/backoff logic for HTTP requests implemented?\",\n\t\"Which files define the settings for background research?\",\n\t\"Find where session entries of type custom are appended.\",\n];\n\n// The probe measures the REAL curation contract — same prompt the BrainCurator ships.\nexport { CURATION_DIGEST_SYSTEM_PROMPT as DIGEST_PROBE_SYSTEM_PROMPT } from \"../context/brain-curator.ts\";\n\n/**\n * Digest probe chunks each carry a NONCE identifier that cannot be guessed from the\n * instructions: acceptance requires the digest to RETAIN the nonce verbatim, so the score\n * measures extraction fidelity, not narration (a model cannot pass by paraphrasing).\n */\nconst DIGEST_PROBE_TASKS: readonly { chunk: string; nonce: string }[] = [\n\t{\n\t\tnonce: \"retryWithJitter_zx41\",\n\t\tchunk: [\n\t\t\t\"grep results for 'retry' under src/http:\",\n\t\t\t\"src/http/client.ts:88: export function retryWithJitter_zx41(fn, attempts = 3) {\",\n\t\t\t\"src/http/client.ts:112:   // exponential backoff capped at 30s\",\n\t\t\t\"src/http/pool.ts:41:   client.retry = false\",\n\t\t].join(\"\\n\"),\n\t},\n\t{\n\t\tnonce: \"ERR_QM_7734\",\n\t\tchunk: [\n\t\t\t\"$ npm run migrate\",\n\t\t\t\"migrating 14 files...\",\n\t\t\t\"error ERR_QM_7734: column 'owner_id' missing on table sessions (migration 0009)\",\n\t\t\t\"exit code 1\",\n\t\t].join(\"\\n\"),\n\t},\n\t{\n\t\tnonce: \"v3.9.2-hotfix.1\",\n\t\tchunk: [\n\t\t\t\"read package.json (34 lines):\",\n\t\t\t'  \"name\": \"acme-billing\",',\n\t\t\t'  \"version\": \"v3.9.2-hotfix.1\",',\n\t\t\t'  \"engines\": { \"node\": \">=22\" },',\n\t\t].join(\"\\n\"),\n\t},\n];\n\nfunction parseDigest(text: string, nonce: string): boolean {\n\tconst parsed = extractJsonObject(text);\n\tif (!parsed) return false;\n\tconst digest = (parsed as { digest?: unknown }).digest;\n\tif (typeof digest !== \"string\") return false;\n\tconst trimmed = digest.trim();\n\t// Bounded and faithful: short enough to be a stub annotation, still carrying the nonce fact.\n\treturn trimmed.length > 0 && trimmed.length <= 240 && trimmed.includes(nonce);\n}\n\nconst TOOL_CALL_PROBE_TASKS: readonly string[] = [\n\t\"Find usages of the function resolveCliModel under src/.\",\n\t\"Search for the string 'budget_exhausted' in the core directory.\",\n\t\"Locate where LaneTracker is instantiated under src/core.\",\n];\n\nfunction parseSearchPlan(text: string): boolean {\n\tconst parsed = extractJsonObject(text);\n\tif (!parsed) return false;\n\tconst queries = (parsed as { queries?: unknown }).queries;\n\tif (!Array.isArray(queries) || queries.length === 0 || queries.length > 8) return false;\n\treturn queries.every(\n\t\t(query) =>\n\t\t\tquery &&\n\t\t\ttypeof query === \"object\" &&\n\t\t\ttypeof (query as { pattern?: unknown }).pattern === \"string\" &&\n\t\t\t(query as { pattern: string }).pattern.trim().length > 0,\n\t);\n}\n\nfunction parseToolCall(text: string): boolean {\n\tconst parsed = extractJsonObject(text);\n\tif (!parsed) return false;\n\tconst record = parsed as { tool?: unknown; arguments?: unknown };\n\tif (record.tool !== \"grep\") return false;\n\tconst args = record.arguments;\n\tif (!args || typeof args !== \"object\" || Array.isArray(args)) return false;\n\tconst pattern = (args as { pattern?: unknown }).pattern;\n\tconst path = (args as { path?: unknown }).path;\n\treturn (\n\t\ttypeof pattern === \"string\" && pattern.trim().length > 0 && typeof path === \"string\" && path.trim().length > 0\n\t);\n}\n\nfunction extractJsonObject(text: string): unknown | undefined {\n\tconst trimmed = text.trim();\n\tconst candidates: string[] = [trimmed];\n\tconst fenced = /```(?:json)?\\s*([\\s\\S]*?)```/.exec(trimmed);\n\tif (fenced?.[1]) candidates.push(fenced[1].trim());\n\tconst start = trimmed.indexOf(\"{\");\n\tconst end = trimmed.lastIndexOf(\"}\");\n\tif (start >= 0 && end > start) candidates.push(trimmed.slice(start, end + 1));\n\tfor (const candidate of candidates) {\n\t\ttry {\n\t\t\tconst parsed = JSON.parse(candidate);\n\t\t\tif (parsed && typeof parsed === \"object\" && !Array.isArray(parsed)) return parsed;\n\t\t} catch {\n\t\t\t// try next candidate\n\t\t}\n\t}\n\treturn undefined;\n}\n\nfunction fitnessEnvelope(): CapabilityEnvelope {\n\treturn {\n\t\tid: \"model-fitness-probe\",\n\t\tcapabilities: [\"research\", \"read_files\", \"memory_read\"],\n\t\tmaxEstimatedUsd: 1,\n\t\tcreatedAt: new Date().toISOString(),\n\t};\n}\n\nexport async function runModelFitnessProbe(options: ModelFitnessOptions): Promise<ModelFitnessReport> {\n\tconst trials = Math.max(1, Math.min(options.trials ?? 3, 20));\n\tconst maxWallClockMs = options.maxWallClockMs ?? 120_000;\n\tconst judgePrompts = options.judgePrompts ?? DEFAULT_JUDGE_FITNESS_PROMPTS;\n\tconst now = options.now ?? Date.now;\n\tlet totalCostUsd = 0;\n\n\t// Token-speed instrumentation: the lane runners' own contracts carry text/cost only, so the\n\t// completer is wrapped once here and generation stats are accumulated per surface.\n\tconst overallSpeed = { tokens: 0, evalMs: 0 };\n\tlet surfaceSpeed = { tokens: 0, evalMs: 0 };\n\tconst complete: FitnessComplete = async (args) => {\n\t\tconst completion = await options.complete(args);\n\t\tconst tokens = completion.outputTokens ?? 0;\n\t\tconst evalMs = completion.evalMs ?? 0;\n\t\tif (tokens > 0 && evalMs > 0) {\n\t\t\tsurfaceSpeed.tokens += tokens;\n\t\t\tsurfaceSpeed.evalMs += evalMs;\n\t\t\toverallSpeed.tokens += tokens;\n\t\t\toverallSpeed.evalMs += evalMs;\n\t\t}\n\t\treturn completion;\n\t};\n\tconst takeSurfaceSpeed = (): number | undefined => {\n\t\tconst speed =\n\t\t\tsurfaceSpeed.evalMs > 0 ? Math.round((surfaceSpeed.tokens / surfaceSpeed.evalMs) * 1000) : undefined;\n\t\tsurfaceSpeed = { tokens: 0, evalMs: 0 };\n\t\treturn speed;\n\t};\n\n\tconst research: LaneFitnessScore = { succeeded: 0, total: trials, outcomes: [], meanMs: 0 };\n\tfor (let i = 0; i < trials; i++) {\n\t\tconst started = now();\n\t\tconst result = await runResearch({\n\t\t\tquery: `fitness:probe requirements:req-${i}`,\n\t\t\tcontext: [\n\t\t\t\t\"Goal: add a retry helper to the HTTP client module\",\n\t\t\t\t\"Open requirements:\",\n\t\t\t\t\"- Find what retry/backoff conventions the codebase already uses\",\n\t\t\t\t\"- Identify which call sites would adopt the helper\",\n\t\t\t].join(\"\\n\"),\n\t\t\tenvelope: fitnessEnvelope(),\n\t\t\tmaxUsd: 1,\n\t\t\tmaxSources: 8,\n\t\t\tmaxFindings: 5,\n\t\t\tmaxWallClockMs,\n\t\t\tsignal: options.signal,\n\t\t\tcomplete,\n\t\t});\n\t\tresearch.meanMs += now() - started;\n\t\ttotalCostUsd += result.costUsd;\n\t\tif (result.status === \"succeeded\") research.succeeded++;\n\t\tresearch.outcomes.push(`${result.status}/${result.reasonCode}`);\n\t}\n\tresearch.meanMs = Math.round(research.meanMs / trials);\n\tresearch.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst worker: LaneFitnessScore = { succeeded: 0, total: trials, outcomes: [], meanMs: 0 };\n\tfor (let i = 0; i < trials; i++) {\n\t\tconst started = now();\n\t\tconst outcome = await runWorker({\n\t\t\trequest: {\n\t\t\t\tid: `fitness-worker-${i}`,\n\t\t\t\tinstructions:\n\t\t\t\t\t\"Summarize in two sentences what a capability envelope is: a declared set of allowed tools, paths, and capability names that bounds what a delegated worker may do.\",\n\t\t\t\troute: { tier: \"cheap\", risk: \"read-only\", confidence: 1, reasonCode: \"fitness_probe\", reasons: [] },\n\t\t\t\tenvelope: { id: `fitness-env-${i}`, capabilities: [\"read_files\"], maxEstimatedUsd: 1 },\n\t\t\t\tmaxEstimatedUsd: 1,\n\t\t\t},\n\t\t\tmaxUsd: 1,\n\t\t\tmaxWallClockMs,\n\t\t\tusageReportId: `fitness:${i}`,\n\t\t\tsignal: options.signal,\n\t\t\tcomplete,\n\t\t});\n\t\tworker.meanMs += now() - started;\n\t\ttotalCostUsd += outcome.costUsd;\n\t\tif (outcome.result.status === \"completed\" && outcome.accepted) worker.succeeded++;\n\t\tworker.outcomes.push(`${outcome.result.status}/${outcome.reasonCode}`);\n\t}\n\tworker.meanMs = Math.round(worker.meanMs / trials);\n\tworker.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst judge: JudgeFitnessScore = {\n\t\tparsed: 0,\n\t\tplanningElevated: 0,\n\t\tplanningTotal: judgePrompts.filter((entry) => entry.planning).length,\n\t\ttrivialCheap: 0,\n\t\ttrivialTotal: judgePrompts.filter((entry) => !entry.planning).length,\n\t\ttotal: judgePrompts.length,\n\t\toutcomes: [],\n\t\tmeanMs: 0,\n\t};\n\tfor (const entry of judgePrompts) {\n\t\tconst started = now();\n\t\tconst result = await runRouteJudge({\n\t\t\tprompt: entry.prompt,\n\t\t\tbaseline: { tier: \"cheap\", risk: \"read-only\", confidence: 0.5, reasonCode: \"fitness_probe\", reasons: [] },\n\t\t\tmaxWallClockMs,\n\t\t\tsignal: options.signal,\n\t\t\tcomplete,\n\t\t});\n\t\tjudge.meanMs += now() - started;\n\t\ttotalCostUsd += result.costUsd;\n\t\tconst tier = result.decision.tier;\n\t\tif (result.verdict) {\n\t\t\tjudge.parsed++;\n\t\t\t// A useful judge must both keep planning off the cheap tier AND actually send trivial\n\t\t\t// prompts there — all-medium verdicts are safe but save nothing.\n\t\t\tif (entry.planning && tier !== \"cheap\") judge.planningElevated++;\n\t\t\tif (!entry.planning && tier === \"cheap\") judge.trivialCheap++;\n\t\t}\n\t\tjudge.outcomes.push(\n\t\t\t`\"${entry.prompt.slice(0, 40)}\" -> ${tier}${result.fallbackReason ? ` (${result.fallbackReason})` : \"\"}`,\n\t\t);\n\t}\n\tjudge.meanMs = judgePrompts.length > 0 ? Math.round(judge.meanMs / judgePrompts.length) : 0;\n\tjudge.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst probeSurface = async (\n\t\tsystemPrompt: string,\n\t\ttasks: readonly string[],\n\t\taccepts: (text: string, taskIndex: number) => boolean,\n\t): Promise<LaneFitnessScore> => {\n\t\tconst score: LaneFitnessScore = { succeeded: 0, total: tasks.length, outcomes: [], meanMs: 0 };\n\t\tfor (const [taskIndex, task] of tasks.entries()) {\n\t\t\tconst started = now();\n\t\t\t// Same wall-clock envelope as the lane surfaces — a hung model must not hang the probe.\n\t\t\tconst bounded = await runBoundedCompletion({\n\t\t\t\tmaxWallClockMs,\n\t\t\t\tsignal: options.signal,\n\t\t\t\texecute: (signal) => complete({ systemPrompt, userPrompt: task, signal }),\n\t\t\t});\n\t\t\tif (bounded.completion) totalCostUsd += bounded.completion.costUsd;\n\t\t\tif (bounded.failure || !bounded.completion) {\n\t\t\t\tscore.outcomes.push(bounded.failure ? bounded.failure.status : \"completion_error\");\n\t\t\t} else {\n\t\t\t\tconst ok = accepts(bounded.completion.text, taskIndex);\n\t\t\t\tif (ok) score.succeeded++;\n\t\t\t\tscore.outcomes.push(ok ? \"ok\" : \"unparseable_output\");\n\t\t\t}\n\t\t\tscore.meanMs += now() - started;\n\t\t}\n\t\tscore.meanMs = tasks.length > 0 ? Math.round(score.meanMs / tasks.length) : 0;\n\t\treturn score;\n\t};\n\n\tconst search = await probeSurface(SEARCH_PROBE_SYSTEM_PROMPT, SEARCH_PROBE_TASKS, parseSearchPlan);\n\tsearch.tokensPerSecond = takeSurfaceSpeed();\n\tconst toolCall = await probeSurface(TOOL_CALL_PROBE_SYSTEM_PROMPT, TOOL_CALL_PROBE_TASKS, parseToolCall);\n\ttoolCall.tokensPerSecond = takeSurfaceSpeed();\n\tconst digest = await probeSurface(\n\t\tCURATION_DIGEST_SYSTEM_PROMPT,\n\t\tDIGEST_PROBE_TASKS.map((task) => task.chunk),\n\t\t(text, taskIndex) => parseDigest(text, DIGEST_PROBE_TASKS[taskIndex]!.nonce),\n\t);\n\tdigest.tokensPerSecond = takeSurfaceSpeed();\n\n\tconst tokensPerSecond =\n\t\toverallSpeed.evalMs > 0 ? Math.round((overallSpeed.tokens / overallSpeed.evalMs) * 1000) : undefined;\n\n\treturn { trials, tokensPerSecond, research, worker, judge, search, toolCall, digest, totalCostUsd };\n}\n\n/**\n * Pure verdict: true when the probe found ZERO successes on every LANE surface it actually graded\n * AND the judge (if it ran) also failed. A lane/judge with total 0 (i.e. never run) carries no\n * evidence and is excluded from the lane check — but at least one lane must actually have been\n * graded for an all-failed verdict at all: `gradedLanes.every(...)` is vacuously true over an\n * empty array, so a report where only the judge ran (every research/worker/search/toolCall/digest\n * lane is ungraded) is excluded explicitly rather than misread as \"all lanes failed\" on zero lane\n * evidence. An empty/degenerate report (nothing graded at all, lanes AND judge) is likewise never\n * mistaken for a failed one. This is the gate adoption flows must consult before assigning a role —\n * see `isProbeAllFailed` callers in interactive-mode.ts and agent-session.ts.\n */\nexport function isProbeAllFailed(report: ModelFitnessReport): boolean {\n\tconst lanes = [report.research, report.worker, report.search, report.toolCall, report.digest];\n\tconst gradedLanes = lanes.filter((lane) => lane.total > 0);\n\tconst judgeGraded = report.judge.total > 0;\n\tif (gradedLanes.length === 0 && !judgeGraded) return false;\n\tconst lanesAllFailed = gradedLanes.length > 0 && gradedLanes.every((lane) => lane.succeeded === 0);\n\tconst judgeFailed = !judgeGraded || report.judge.parsed === 0;\n\treturn lanesAllFailed && judgeFailed;\n}\n\n/** Compact human-readable report for tool output / interactive display. Bounded, no raw dumps. */\nexport function formatModelFitnessReport(model: string, report: ModelFitnessReport): string {\n\tconst speed = (tokensPerSecond: number | undefined) =>\n\t\ttokensPerSecond !== undefined ? `, ~${tokensPerSecond} tok/s` : \"\";\n\tconst lines = [\n\t\t`Model fitness: ${model} (${report.trials} trials/lane${speed(report.tokensPerSecond)})`,\n\t\t`- research lane: ${report.research.succeeded}/${report.research.total} succeeded, mean ${report.research.meanMs}ms${speed(report.research.tokensPerSecond)} [${report.research.outcomes.join(\", \")}]`,\n\t\t`- worker lane:   ${report.worker.succeeded}/${report.worker.total} completed+accepted, mean ${report.worker.meanMs}ms${speed(report.worker.tokensPerSecond)} [${report.worker.outcomes.join(\", \")}]`,\n\t\t`- search plans:  ${report.search.succeeded}/${report.search.total} well-formed, mean ${report.search.meanMs}ms${speed(report.search.tokensPerSecond)}`,\n\t\t`- tool calls:    ${report.toolCall.succeeded}/${report.toolCall.total} well-formed, mean ${report.toolCall.meanMs}ms${speed(report.toolCall.tokensPerSecond)}`,\n\t\t`- digests:       ${report.digest.succeeded}/${report.digest.total} faithful, mean ${report.digest.meanMs}ms${speed(report.digest.tokensPerSecond)}`,\n\t\t`- route judge:   parsed ${report.judge.parsed}/${report.judge.total}, planning-elevated ${report.judge.planningElevated}/${report.judge.planningTotal}, trivial-cheap ${report.judge.trivialCheap}/${report.judge.trivialTotal}, mean ${report.judge.meanMs}ms${speed(report.judge.tokensPerSecond)}`,\n\t\t...report.judge.outcomes.map((outcome) => `    ${outcome}`),\n\t];\n\tif (report.totalCostUsd > 0) {\n\t\tlines.push(`- probe cost: $${report.totalCostUsd.toFixed(4)}`);\n\t}\n\treturn lines.join(\"\\n\");\n}\n"]}

package/dist/core/research/session-evidence-bundle.d.ts

@@ -1,5 +1,5 @@
+import type { SessionEntry, SessionManager } from "@caupulican/pi-agent-core/node";
 import type { EvidenceBundle } from "../autonomy/contracts.ts";
-import type { SessionEntry, SessionManager } from "../session-manager.ts";
 export declare const EVIDENCE_BUNDLE_CUSTOM_TYPE = "evidence_bundle";
 export interface EvidenceBundleSnapshotPayload {
     version: 1;

package/dist/core/research/session-evidence-bundle.d.ts.map

@@ -1 +1 @@
-{"version":3,"file":"session-evidence-bundle.d.ts","sourceRoot":"","sources":["../../../src/core/research/session-evidence-bundle.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAC/D,OAAO,KAAK,EAAE,YAAY,EAAE,cAAc,EAAE,MAAM,uBAAuB,CAAC;AAG1E,eAAO,MAAM,2BAA2B,oBAAoB,CAAC;AAE7D,MAAM,WAAW,6BAA6B;IAC7C,OAAO,EAAE,CAAC,CAAC;IACX,MAAM,EAAE,cAAc,CAAC;CACvB;AAED,wBAAgB,4BAA4B,CAC3C,cAAc,EAAE,IAAI,CAAC,cAAc,EAAE,mBAAmB,CAAC,EACzD,MAAM,EAAE,cAAc,GACpB,MAAM,CAMR;AAQD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,SAAS,YAAY,EAAE,GAAG,cAAc,EAAE,CAmB7F;AAED,wBAAgB,+BAA+B,CAAC,OAAO,EAAE,SAAS,YAAY,EAAE,GAAG,cAAc,GAAG,SAAS,CAe5G","sourcesContent":["import type { EvidenceBundle } from \"../autonomy/contracts.ts\";\nimport type { SessionEntry, SessionManager } from \"../session-manager.ts\";\nimport { cloneEvidenceBundleForStorage, isEvidenceBundle } from \"./evidence-bundle.ts\";\n\nexport const EVIDENCE_BUNDLE_CUSTOM_TYPE = \"evidence_bundle\";\n\nexport interface EvidenceBundleSnapshotPayload {\n\tversion: 1;\n\tbundle: EvidenceBundle;\n}\n\nexport function appendEvidenceBundleSnapshot(\n\tsessionManager: Pick<SessionManager, \"appendCustomEntry\">,\n\tbundle: EvidenceBundle,\n): string {\n\tconst payload: EvidenceBundleSnapshotPayload = {\n\t\tversion: 1,\n\t\tbundle: cloneEvidenceBundleForStorage(bundle),\n\t};\n\treturn sessionManager.appendCustomEntry(EVIDENCE_BUNDLE_CUSTOM_TYPE, payload);\n}\n\nfunction isPlainRecord(value: unknown): value is Record<string, unknown> {\n\tif (!value || typeof value !== \"object\" || Array.isArray(value)) return false;\n\tconst prototype = Object.getPrototypeOf(value);\n\treturn prototype === Object.prototype || prototype === null;\n}\n\nexport function getEvidenceBundleSnapshots(entries: readonly SessionEntry[]): EvidenceBundle[] {\n\tconst bundles: EvidenceBundle[] = [];\n\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst payload = entry.data;\n\t\tif (!isPlainRecord(payload)) continue;\n\t\tif (payload.version !== 1) continue;\n\t\tif (!(\"bundle\" in payload)) continue;\n\t\tconst bundle = payload.bundle;\n\t\tif (isEvidenceBundle(bundle)) {\n\t\t\tbundles.push(cloneEvidenceBundleForStorage(bundle));\n\t\t}\n\t}\n\n\treturn bundles;\n}\n\nexport function getLatestEvidenceBundleSnapshot(entries: readonly SessionEntry[]): EvidenceBundle | undefined {\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type === \"custom\" && entry.customType === EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tconst payload = entry.data;\n\t\t\tif (!isPlainRecord(payload)) continue;\n\t\t\tif (payload.version !== 1) continue;\n\t\t\tif (!(\"bundle\" in payload)) continue;\n\t\t\tconst bundle = payload.bundle;\n\t\t\tif (isEvidenceBundle(bundle)) {\n\t\t\t\treturn cloneEvidenceBundleForStorage(bundle);\n\t\t\t}\n\t\t}\n\t}\n\treturn undefined;\n}\n"]}
+{"version":3,"file":"session-evidence-bundle.d.ts","sourceRoot":"","sources":["../../../src/core/research/session-evidence-bundle.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,cAAc,EAAE,MAAM,gCAAgC,CAAC;AACnF,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,0BAA0B,CAAC;AAG/D,eAAO,MAAM,2BAA2B,oBAAoB,CAAC;AAE7D,MAAM,WAAW,6BAA6B;IAC7C,OAAO,EAAE,CAAC,CAAC;IACX,MAAM,EAAE,cAAc,CAAC;CACvB;AAED,wBAAgB,4BAA4B,CAC3C,cAAc,EAAE,IAAI,CAAC,cAAc,EAAE,mBAAmB,CAAC,EACzD,MAAM,EAAE,cAAc,GACpB,MAAM,CAMR;AAQD,wBAAgB,0BAA0B,CAAC,OAAO,EAAE,SAAS,YAAY,EAAE,GAAG,cAAc,EAAE,CAmB7F;AAED,wBAAgB,+BAA+B,CAAC,OAAO,EAAE,SAAS,YAAY,EAAE,GAAG,cAAc,GAAG,SAAS,CAe5G","sourcesContent":["import type { SessionEntry, SessionManager } from \"@caupulican/pi-agent-core/node\";\nimport type { EvidenceBundle } from \"../autonomy/contracts.ts\";\nimport { cloneEvidenceBundleForStorage, isEvidenceBundle } from \"./evidence-bundle.ts\";\n\nexport const EVIDENCE_BUNDLE_CUSTOM_TYPE = \"evidence_bundle\";\n\nexport interface EvidenceBundleSnapshotPayload {\n\tversion: 1;\n\tbundle: EvidenceBundle;\n}\n\nexport function appendEvidenceBundleSnapshot(\n\tsessionManager: Pick<SessionManager, \"appendCustomEntry\">,\n\tbundle: EvidenceBundle,\n): string {\n\tconst payload: EvidenceBundleSnapshotPayload = {\n\t\tversion: 1,\n\t\tbundle: cloneEvidenceBundleForStorage(bundle),\n\t};\n\treturn sessionManager.appendCustomEntry(EVIDENCE_BUNDLE_CUSTOM_TYPE, payload);\n}\n\nfunction isPlainRecord(value: unknown): value is Record<string, unknown> {\n\tif (!value || typeof value !== \"object\" || Array.isArray(value)) return false;\n\tconst prototype = Object.getPrototypeOf(value);\n\treturn prototype === Object.prototype || prototype === null;\n}\n\nexport function getEvidenceBundleSnapshots(entries: readonly SessionEntry[]): EvidenceBundle[] {\n\tconst bundles: EvidenceBundle[] = [];\n\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst payload = entry.data;\n\t\tif (!isPlainRecord(payload)) continue;\n\t\tif (payload.version !== 1) continue;\n\t\tif (!(\"bundle\" in payload)) continue;\n\t\tconst bundle = payload.bundle;\n\t\tif (isEvidenceBundle(bundle)) {\n\t\t\tbundles.push(cloneEvidenceBundleForStorage(bundle));\n\t\t}\n\t}\n\n\treturn bundles;\n}\n\nexport function getLatestEvidenceBundleSnapshot(entries: readonly SessionEntry[]): EvidenceBundle | undefined {\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type === \"custom\" && entry.customType === EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tconst payload = entry.data;\n\t\t\tif (!isPlainRecord(payload)) continue;\n\t\t\tif (payload.version !== 1) continue;\n\t\t\tif (!(\"bundle\" in payload)) continue;\n\t\t\tconst bundle = payload.bundle;\n\t\t\tif (isEvidenceBundle(bundle)) {\n\t\t\t\treturn cloneEvidenceBundleForStorage(bundle);\n\t\t\t}\n\t\t}\n\t}\n\treturn undefined;\n}\n"]}

package/dist/core/research/session-evidence-bundle.js.map

@@ -1 +1 @@
-{"version":3,"file":"session-evidence-bundle.js","sourceRoot":"","sources":["../../../src/core/research/session-evidence-bundle.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,6BAA6B,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAEvF,MAAM,CAAC,MAAM,2BAA2B,GAAG,iBAAiB,CAAC;AAO7D,MAAM,UAAU,4BAA4B,CAC3C,cAAyD,EACzD,MAAsB,EACb;IACT,MAAM,OAAO,GAAkC;QAC9C,OAAO,EAAE,CAAC;QACV,MAAM,EAAE,6BAA6B,CAAC,MAAM,CAAC;KAC7C,CAAC;IACF,OAAO,cAAc,CAAC,iBAAiB,CAAC,2BAA2B,EAAE,OAAO,CAAC,CAAC;AAAA,CAC9E;AAED,SAAS,aAAa,CAAC,KAAc,EAAoC;IACxE,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC9E,MAAM,SAAS,GAAG,MAAM,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;IAC/C,OAAO,SAAS,KAAK,MAAM,CAAC,SAAS,IAAI,SAAS,KAAK,IAAI,CAAC;AAAA,CAC5D;AAED,MAAM,UAAU,0BAA0B,CAAC,OAAgC,EAAoB;IAC9F,MAAM,OAAO,GAAqB,EAAE,CAAC;IAErC,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,UAAU,KAAK,2BAA2B,EAAE,CAAC;YACjF,SAAS;QACV,CAAC;QAED,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC;QAC3B,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC;YAAE,SAAS;QACtC,IAAI,OAAO,CAAC,OAAO,KAAK,CAAC;YAAE,SAAS;QACpC,IAAI,CAAC,CAAC,QAAQ,IAAI,OAAO,CAAC;YAAE,SAAS;QACrC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;QAC9B,IAAI,gBAAgB,CAAC,MAAM,CAAC,EAAE,CAAC;YAC9B,OAAO,CAAC,IAAI,CAAC,6BAA6B,CAAC,MAAM,CAAC,CAAC,CAAC;QACrD,CAAC;IACF,CAAC;IAED,OAAO,OAAO,CAAC;AAAA,CACf;AAED,MAAM,UAAU,+BAA+B,CAAC,OAAgC,EAA8B;IAC7G,KAAK,IAAI,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9C,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QACzB,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,UAAU,KAAK,2BAA2B,EAAE,CAAC;YACjF,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC;YAC3B,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC;gBAAE,SAAS;YACtC,IAAI,OAAO,CAAC,OAAO,KAAK,CAAC;gBAAE,SAAS;YACpC,IAAI,CAAC,CAAC,QAAQ,IAAI,OAAO,CAAC;gBAAE,SAAS;YACrC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;YAC9B,IAAI,gBAAgB,CAAC,MAAM,CAAC,EAAE,CAAC;gBAC9B,OAAO,6BAA6B,CAAC,MAAM,CAAC,CAAC;YAC9C,CAAC;QACF,CAAC;IACF,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB","sourcesContent":["import type { EvidenceBundle } from \"../autonomy/contracts.ts\";\nimport type { SessionEntry, SessionManager } from \"../session-manager.ts\";\nimport { cloneEvidenceBundleForStorage, isEvidenceBundle } from \"./evidence-bundle.ts\";\n\nexport const EVIDENCE_BUNDLE_CUSTOM_TYPE = \"evidence_bundle\";\n\nexport interface EvidenceBundleSnapshotPayload {\n\tversion: 1;\n\tbundle: EvidenceBundle;\n}\n\nexport function appendEvidenceBundleSnapshot(\n\tsessionManager: Pick<SessionManager, \"appendCustomEntry\">,\n\tbundle: EvidenceBundle,\n): string {\n\tconst payload: EvidenceBundleSnapshotPayload = {\n\t\tversion: 1,\n\t\tbundle: cloneEvidenceBundleForStorage(bundle),\n\t};\n\treturn sessionManager.appendCustomEntry(EVIDENCE_BUNDLE_CUSTOM_TYPE, payload);\n}\n\nfunction isPlainRecord(value: unknown): value is Record<string, unknown> {\n\tif (!value || typeof value !== \"object\" || Array.isArray(value)) return false;\n\tconst prototype = Object.getPrototypeOf(value);\n\treturn prototype === Object.prototype || prototype === null;\n}\n\nexport function getEvidenceBundleSnapshots(entries: readonly SessionEntry[]): EvidenceBundle[] {\n\tconst bundles: EvidenceBundle[] = [];\n\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst payload = entry.data;\n\t\tif (!isPlainRecord(payload)) continue;\n\t\tif (payload.version !== 1) continue;\n\t\tif (!(\"bundle\" in payload)) continue;\n\t\tconst bundle = payload.bundle;\n\t\tif (isEvidenceBundle(bundle)) {\n\t\t\tbundles.push(cloneEvidenceBundleForStorage(bundle));\n\t\t}\n\t}\n\n\treturn bundles;\n}\n\nexport function getLatestEvidenceBundleSnapshot(entries: readonly SessionEntry[]): EvidenceBundle | undefined {\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type === \"custom\" && entry.customType === EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tconst payload = entry.data;\n\t\t\tif (!isPlainRecord(payload)) continue;\n\t\t\tif (payload.version !== 1) continue;\n\t\t\tif (!(\"bundle\" in payload)) continue;\n\t\t\tconst bundle = payload.bundle;\n\t\t\tif (isEvidenceBundle(bundle)) {\n\t\t\t\treturn cloneEvidenceBundleForStorage(bundle);\n\t\t\t}\n\t\t}\n\t}\n\treturn undefined;\n}\n"]}
+{"version":3,"file":"session-evidence-bundle.js","sourceRoot":"","sources":["../../../src/core/research/session-evidence-bundle.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,6BAA6B,EAAE,gBAAgB,EAAE,MAAM,sBAAsB,CAAC;AAEvF,MAAM,CAAC,MAAM,2BAA2B,GAAG,iBAAiB,CAAC;AAO7D,MAAM,UAAU,4BAA4B,CAC3C,cAAyD,EACzD,MAAsB,EACb;IACT,MAAM,OAAO,GAAkC;QAC9C,OAAO,EAAE,CAAC;QACV,MAAM,EAAE,6BAA6B,CAAC,MAAM,CAAC;KAC7C,CAAC;IACF,OAAO,cAAc,CAAC,iBAAiB,CAAC,2BAA2B,EAAE,OAAO,CAAC,CAAC;AAAA,CAC9E;AAED,SAAS,aAAa,CAAC,KAAc,EAAoC;IACxE,IAAI,CAAC,KAAK,IAAI,OAAO,KAAK,KAAK,QAAQ,IAAI,KAAK,CAAC,OAAO,CAAC,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC9E,MAAM,SAAS,GAAG,MAAM,CAAC,cAAc,CAAC,KAAK,CAAC,CAAC;IAC/C,OAAO,SAAS,KAAK,MAAM,CAAC,SAAS,IAAI,SAAS,KAAK,IAAI,CAAC;AAAA,CAC5D;AAED,MAAM,UAAU,0BAA0B,CAAC,OAAgC,EAAoB;IAC9F,MAAM,OAAO,GAAqB,EAAE,CAAC;IAErC,KAAK,MAAM,KAAK,IAAI,OAAO,EAAE,CAAC;QAC7B,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,UAAU,KAAK,2BAA2B,EAAE,CAAC;YACjF,SAAS;QACV,CAAC;QAED,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC;QAC3B,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC;YAAE,SAAS;QACtC,IAAI,OAAO,CAAC,OAAO,KAAK,CAAC;YAAE,SAAS;QACpC,IAAI,CAAC,CAAC,QAAQ,IAAI,OAAO,CAAC;YAAE,SAAS;QACrC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;QAC9B,IAAI,gBAAgB,CAAC,MAAM,CAAC,EAAE,CAAC;YAC9B,OAAO,CAAC,IAAI,CAAC,6BAA6B,CAAC,MAAM,CAAC,CAAC,CAAC;QACrD,CAAC;IACF,CAAC;IAED,OAAO,OAAO,CAAC;AAAA,CACf;AAED,MAAM,UAAU,+BAA+B,CAAC,OAAgC,EAA8B;IAC7G,KAAK,IAAI,CAAC,GAAG,OAAO,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC,IAAI,CAAC,EAAE,CAAC,EAAE,EAAE,CAAC;QAC9C,MAAM,KAAK,GAAG,OAAO,CAAC,CAAC,CAAC,CAAC;QACzB,IAAI,KAAK,CAAC,IAAI,KAAK,QAAQ,IAAI,KAAK,CAAC,UAAU,KAAK,2BAA2B,EAAE,CAAC;YACjF,MAAM,OAAO,GAAG,KAAK,CAAC,IAAI,CAAC;YAC3B,IAAI,CAAC,aAAa,CAAC,OAAO,CAAC;gBAAE,SAAS;YACtC,IAAI,OAAO,CAAC,OAAO,KAAK,CAAC;gBAAE,SAAS;YACpC,IAAI,CAAC,CAAC,QAAQ,IAAI,OAAO,CAAC;gBAAE,SAAS;YACrC,MAAM,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;YAC9B,IAAI,gBAAgB,CAAC,MAAM,CAAC,EAAE,CAAC;gBAC9B,OAAO,6BAA6B,CAAC,MAAM,CAAC,CAAC;YAC9C,CAAC;QACF,CAAC;IACF,CAAC;IACD,OAAO,SAAS,CAAC;AAAA,CACjB","sourcesContent":["import type { SessionEntry, SessionManager } from \"@caupulican/pi-agent-core/node\";\nimport type { EvidenceBundle } from \"../autonomy/contracts.ts\";\nimport { cloneEvidenceBundleForStorage, isEvidenceBundle } from \"./evidence-bundle.ts\";\n\nexport const EVIDENCE_BUNDLE_CUSTOM_TYPE = \"evidence_bundle\";\n\nexport interface EvidenceBundleSnapshotPayload {\n\tversion: 1;\n\tbundle: EvidenceBundle;\n}\n\nexport function appendEvidenceBundleSnapshot(\n\tsessionManager: Pick<SessionManager, \"appendCustomEntry\">,\n\tbundle: EvidenceBundle,\n): string {\n\tconst payload: EvidenceBundleSnapshotPayload = {\n\t\tversion: 1,\n\t\tbundle: cloneEvidenceBundleForStorage(bundle),\n\t};\n\treturn sessionManager.appendCustomEntry(EVIDENCE_BUNDLE_CUSTOM_TYPE, payload);\n}\n\nfunction isPlainRecord(value: unknown): value is Record<string, unknown> {\n\tif (!value || typeof value !== \"object\" || Array.isArray(value)) return false;\n\tconst prototype = Object.getPrototypeOf(value);\n\treturn prototype === Object.prototype || prototype === null;\n}\n\nexport function getEvidenceBundleSnapshots(entries: readonly SessionEntry[]): EvidenceBundle[] {\n\tconst bundles: EvidenceBundle[] = [];\n\n\tfor (const entry of entries) {\n\t\tif (entry.type !== \"custom\" || entry.customType !== EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tcontinue;\n\t\t}\n\n\t\tconst payload = entry.data;\n\t\tif (!isPlainRecord(payload)) continue;\n\t\tif (payload.version !== 1) continue;\n\t\tif (!(\"bundle\" in payload)) continue;\n\t\tconst bundle = payload.bundle;\n\t\tif (isEvidenceBundle(bundle)) {\n\t\t\tbundles.push(cloneEvidenceBundleForStorage(bundle));\n\t\t}\n\t}\n\n\treturn bundles;\n}\n\nexport function getLatestEvidenceBundleSnapshot(entries: readonly SessionEntry[]): EvidenceBundle | undefined {\n\tfor (let i = entries.length - 1; i >= 0; i--) {\n\t\tconst entry = entries[i];\n\t\tif (entry.type === \"custom\" && entry.customType === EVIDENCE_BUNDLE_CUSTOM_TYPE) {\n\t\t\tconst payload = entry.data;\n\t\t\tif (!isPlainRecord(payload)) continue;\n\t\t\tif (payload.version !== 1) continue;\n\t\t\tif (!(\"bundle\" in payload)) continue;\n\t\t\tconst bundle = payload.bundle;\n\t\t\tif (isEvidenceBundle(bundle)) {\n\t\t\t\treturn cloneEvidenceBundleForStorage(bundle);\n\t\t\t}\n\t\t}\n\t}\n\treturn undefined;\n}\n"]}

package/dist/core/runtime-builder.d.ts

@@ -0,0 +1,222 @@
+/**
+ * Runtime build & reload: the session's tool-registry assembly and the self-modification-safe
+ * extension reload path. Owns building the base tool definitions, wrapping them into the live tool
+ * registry (`_refreshToolRegistry`), constructing the {@link ExtensionRunner} for a rebuilt runtime
+ * (`_buildRuntime`), and the repo's self-modification safety crown jewel: the preflight-guarded
+ * `reload()` with its snapshot / doctor / commit-or-rollback sequence, plus the single-extension
+ * live load/unload/reconcile operations that rebuild the runtime in place.
+ *
+ * Extracted verbatim from agent-session.ts (god-file decomposition). Owns ONLY the tool-registry
+ * state — the base tool definitions, the wrapped tool registry, the definition registry, and the
+ * per-tool prompt snippet/guideline maps. Everything else the build & reload touch — the live
+ * {@link ExtensionRunner}, the agent's tool/system-prompt state, the base system prompt, the
+ * resource loader, the session/settings managers, the model registry, the profile tool filter and
+ * its warning sinks, the requested-active-tool-names request, the memory subsystem, and the many
+ * host callbacks the rebuilt tools/runtime are wired to — is reached through narrow deps accessors
+ * that read and write the SAME storage the host owns today.
+ *
+ * Snapshot-ownership boundary (deliberate, load-bearing for reload safety): the reload snapshot
+ * spans state owned by OTHER collaborators — the extension runner and its ref, `agent.state.tools`
+ * / `agent.state.systemPrompt`, and `_baseSystemPrompt`. Those are captured and restored through
+ * {@link RuntimeBuilderDeps} get/set accessors so save/restore mutate exactly the host/agent fields
+ * they mutated before extraction; only the five tool-registry maps are captured by direct field
+ * reference here (this builder owns them). The `_extensionRunnerRef.current` update folds into
+ * {@link RuntimeBuilderDeps.setExtensionRunner} (the same field-then-ref pattern the host used at
+ * every assignment site), so no observer runs between the two writes.
+ *
+ * Host-binding boundary (deliberate): `_bindExtensionCore` — which exposes the SESSION's own public
+ * surface (sendMessage, setModel, compact, abort, reload, …) to the extension runner — stays
+ * host-side and is invoked from {@link buildRuntime} via {@link RuntimeBuilderDeps.bindExtensionCore};
+ * it is host identity, not build logic, and moving it would only re-export ~30 host methods through
+ * deps to hand them straight back. The host keeps one-line delegations for the public reload /
+ * load / unload / reconcile API and for `getAllTools` / `getToolDefinition`, and a private
+ * `_refreshToolRegistry` delegation for its internal callers.
+ */
+import type { Agent, AgentContext, AgentMessage, AgentTool } from "@caupulican/pi-agent-core";
+import type { SessionManager } from "@caupulican/pi-agent-core/node";
+import type { Api, Model, Usage } from "@caupulican/pi-ai";
+import type { IsolatedCompletionOptions, IsolatedCompletionResult, WorkerDelegationRunOutcome } from "./agent-session.ts";
+import type { ArtifactStore } from "./context/context-artifacts.ts";
+import type { MemoryPromptInclusionReport, MemoryRetrievalDiagnostics } from "./context/memory-diagnostics.ts";
+import type { ContextGcReport } from "./context-gc.ts";
+import { type ContextUsage, type Extension, type ExtensionCommandContextActions, type ExtensionErrorListener, ExtensionRunner, type ExtensionUIContext, type ShutdownHandler, type ToolDefinition, type ToolInfo } from "./extensions/index.ts";
+import type { GoalState } from "./goals/goal-state.ts";
+import type { MemoryManager } from "./memory/memory-manager.ts";
+import type { ModelRegistry } from "./model-registry.ts";
+import type { ModelFitnessReport } from "./research/model-fitness.ts";
+import type { ResourceLoader } from "./resource-loader.ts";
+import { type ResourceProfileFilterSettings, type SettingsManager } from "./settings-manager.ts";
+export interface RuntimeBuilderDeps {
+    /** Live agent — the snapshot/doctor read and restore its `state.tools` / `state.systemPrompt`. */
+    getAgent(): Agent;
+    /** Workspace root, passed to the tool-definition factory and the extension runner. */
+    getCwd(): string;
+    /** Session log, passed to the extension runner. */
+    getSessionManager(): SessionManager;
+    /** Tool/shell/toolkit/resource settings + reload target (settingsManager.reload()). */
+    getSettingsManager(): SettingsManager;
+    /** Model registry, passed to the extension runner and profile model re-resolution. */
+    getModelRegistry(): ModelRegistry;
+    /** Extension/skill/prompt/theme discovery + the reload/commit/rollback generation swap. */
+    getResourceLoader(): ResourceLoader;
+    /** Live extension runner (host-owned; pervasive). */
+    getExtensionRunner(): ExtensionRunner;
+    /** Store the extension runner AND update the Agent's mutable `_extensionRunnerRef.current` (both together). */
+    setExtensionRunner(runner: ExtensionRunner): void;
+    /** Base (extension-free) system prompt; captured/restored by the reload snapshot. */
+    getBaseSystemPrompt(): string;
+    setBaseSystemPrompt(prompt: string): void;
+    /** SDK-provided plain tools, synthesized into the registry alongside built-ins. */
+    getCustomTools(): ToolDefinition[];
+    /** Optional plain-tool override; when set, the built-in factory + core diagnostics are skipped. */
+    getBaseToolsOverride(): Record<string, AgentTool> | undefined;
+    /** Pre-filter tool REQUEST (never the capability/profile-filtered active set) preserved across a rebuild. */
+    getRequestedActiveToolNames(): string[] | undefined;
+    setRequestedActiveToolNames(names: string[] | undefined): void;
+    /** Resource-profile tool allow/block filter + the raw allow/exclude sets applied during registry build. */
+    getToolProfileFilter(): Required<ResourceProfileFilterSettings> | undefined;
+    setToolProfileFilter(filter: Required<ResourceProfileFilterSettings> | undefined): void;
+    getAllowedToolNames(): Set<string> | undefined;
+    getExcludedToolNames(): Set<string> | undefined;
+    /** Re-derive the profile tool filter from freshly reloaded settings (reload only). */
+    deriveToolProfileFilter(): Required<ResourceProfileFilterSettings>;
+    /** True when a tool/command name survives the active profile's allow/block + user allow/exclude. */
+    isToolOrCommandAllowedByProfile(name: string): boolean;
+    /** Filter the loaded extensions through the active resource profile (records inert/denied warnings host-side). */
+    filterExtensionsForRuntime(extensions: Extension[]): Extension[];
+    /** Sink for the G13 unbound-profile-tool-grant warnings surfaced in /context. */
+    setUnboundToolGrantWarnings(warnings: string[]): void;
+    /** Currently-active tool names (reads agent.state.tools; the pre-filter fallback for a rebuild). */
+    getActiveToolNames(): string[];
+    /** Apply the recomputed active set (capability filter + companion auto-activation live here). */
+    setActiveToolsByName(toolNames: string[]): void;
+    /** Normalize a tool's prompt snippet / guidelines through the system-prompt builder. */
+    normalizePromptSnippet(text: string | undefined): string | undefined;
+    normalizePromptGuidelines(guidelines: string[] | undefined): string[];
+    /** Wire the session's own public surface into a freshly-built runner (host identity; stays host-side). */
+    bindExtensionCore(runner: ExtensionRunner): void;
+    /** Re-apply UI context / mode / command-context / error subscription to a runner. */
+    applyExtensionBindings(runner: ExtensionRunner): void;
+    /** Re-run resource discovery for the active extensions (reload only). */
+    extendResourcesFromExtensions(reason: "startup" | "reload"): Promise<void>;
+    /** Re-apply the active profile's model/thinking from reloaded settings (reload only). */
+    reapplyActiveProfileModelSettings(): Promise<void>;
+    /** Notify extensions-changed listeners after a single-extension live op. */
+    notifyExtensionsChanged(): void;
+    /** Session-scoped tool-output artifact store for grep/find/artifact_retrieve (gated on the profile). */
+    getToolArtifactStore(): ArtifactStore;
+    /** Live memory manager — its provider tools join the registry. */
+    getMemoryManager(): MemoryManager;
+    /** Memory retrieval + prompt-inclusion diagnostics for the core diagnostics tool. */
+    getMemoryAuditDiagnostics(): {
+        retrieval: MemoryRetrievalDiagnostics;
+        promptInclusion: MemoryPromptInclusionReport;
+    };
+    /** Drop extension-contributed pending memory providers before a reload re-registers them. */
+    clearPendingMemoryProviders(): void;
+    /** (Re)derive the memory subsystem from reloaded settings/providers. */
+    initializeMemory(): Promise<void>;
+    /** Goal-tool state accessors. */
+    getGoalStateSnapshot(): GoalState | undefined;
+    saveGoalStateSnapshot(state: GoalState): string;
+    /** Context-gc report for the core diagnostics tool. */
+    getContextGcReport(messages: AgentMessage[]): ContextGcReport;
+    /** Worker-delegation runner for the delegate tool. */
+    runWorkerDelegationOnce(request: {
+        instructions: string;
+        systemPrompt?: string;
+    }): Promise<WorkerDelegationRunOutcome>;
+    /** Model-fitness probe for the model_fitness tool. */
+    runModelFitness(args: {
+        model: string;
+        trials?: number;
+    }): Promise<{
+        started: true;
+        model: string;
+        report: ModelFitnessReport;
+    } | {
+        started: false;
+        skipReason: string;
+    }>;
+    /** Fitness-gated reflex-brain model resolver (run_toolkit_script interpretation). */
+    resolveCurationModelIfFit(): Model<Api> | undefined;
+    /** One-shot, tool-less LLM call — the reflex-brain interpreter rides this. */
+    runIsolatedCompletion(opts: IsolatedCompletionOptions): Promise<IsolatedCompletionResult>;
+    /** Roll reflex-brain spend into spawned-usage accounting. */
+    addSpawnedUsage(usage: Usage, opts?: {
+        label?: string;
+        sourceSessionId?: string;
+        reportId?: string;
+    }): string | undefined;
+    /** Post-rebuild doctor helpers (validate the rebuilt runtime can render a context). */
+    createAgentContextSnapshot(): AgentContext;
+    getContextUsage(): ContextUsage | undefined;
+    /** Reload/live-op preflight refusal guards (identical refusal behavior; host-checked). */
+    isStreaming(): boolean;
+    isCompacting(): boolean;
+    /** Extension bindings present — gate for re-emitting session_start on reload. */
+    getExtensionUIContext(): ExtensionUIContext | undefined;
+    getExtensionCommandContextActions(): ExtensionCommandContextActions | undefined;
+    getExtensionShutdownHandler(): ShutdownHandler | undefined;
+    getExtensionErrorListener(): ExtensionErrorListener | undefined;
+}
+/**
+ * Owns the tool-registry build and the self-modification-safe extension reload extracted from
+ * {@link AgentSession}. See the module header for the snapshot-ownership and host-binding boundaries.
+ */
+export declare class RuntimeBuilder {
+    private _toolRegistry;
+    private _toolDefinitions;
+    private _toolPromptSnippets;
+    private _toolPromptGuidelines;
+    private _baseToolDefinitions;
+    private readonly deps;
+    constructor(deps: RuntimeBuilderDeps);
+    /** Whether a tool name is present in the live wrapped registry. */
+    hasTool(name: string): boolean;
+    /** The live wrapped tool for a name, if registered (activation lookup). */
+    getRegisteredTool(name: string): AgentTool | undefined;
+    /** A registered tool's normalized prompt snippet, if any. */
+    getToolPromptSnippet(name: string): string | undefined;
+    /** A registered tool's normalized prompt guidelines, if any. */
+    getToolPromptGuidelines(name: string): string[] | undefined;
+    /**
+     * Get all configured tools with name, description, parameter schema, prompt guidelines, and source metadata.
+     */
+    getAllTools(): ToolInfo[];
+    getToolDefinition(name: string): ToolDefinition | undefined;
+    refreshToolRegistry(options?: {
+        activeToolNames?: string[];
+        includeAllExtensionTools?: boolean;
+    }): void;
+    private _createReloadRuntimeSnapshot;
+    private _restoreReloadRuntimeSnapshot;
+    private _doctorReloadRuntime;
+    buildRuntime(options: {
+        activeToolNames?: string[];
+        flagValues?: Map<string, boolean | string>;
+        includeAllExtensionTools?: boolean;
+    }): void;
+    reload(): Promise<void>;
+    /**
+     * Unload a single extension without full reload.
+     * Runs the extension's session_shutdown lifecycle, unregisters its providers,
+     * disposes its event subscriptions, and rebuilds the runtime.
+     * Falls back to full reload on error.
+     */
+    unloadExtensionLive(extensionPath: string): Promise<void>;
+    /**
+     * Load a single extension without full reload.
+     * Loads the extension with fresh import, rebuilds the runtime,
+     * and runs the extension's session_start lifecycle.
+     * Falls back to full reload on error.
+     */
+    loadExtensionLive(extensionPath: string): Promise<void>;
+    /**
+     * Reconcile loaded extensions with the active profile.
+     * Loads extensions that should be enabled but aren't, and unloads extensions that shouldn't be.
+     * Falls back to full reload if any individual load/unload fails.
+     */
+    reconcileLoadedExtensions(): Promise<void>;
+}
+//# sourceMappingURL=runtime-builder.d.ts.map

package/dist/core/runtime-builder.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"runtime-builder.d.ts","sourceRoot":"","sources":["../../src/core/runtime-builder.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;GAiCG;AAEH,OAAO,KAAK,EAAE,KAAK,EAAE,YAAY,EAAE,YAAY,EAAE,SAAS,EAAE,MAAM,2BAA2B,CAAC;AAC9F,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,gCAAgC,CAAC;AACrE,OAAO,KAAK,EAAE,GAAG,EAAE,KAAK,EAAE,KAAK,EAAE,MAAM,mBAAmB,CAAC;AAE3D,OAAO,KAAK,EACX,yBAAyB,EACzB,wBAAwB,EACxB,0BAA0B,EAC1B,MAAM,oBAAoB,CAAC;AAC5B,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,gCAAgC,CAAC;AACpE,OAAO,KAAK,EAAE,2BAA2B,EAAE,0BAA0B,EAAE,MAAM,iCAAiC,CAAC;AAC/G,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAEvD,OAAO,EACN,KAAK,YAAY,EACjB,KAAK,SAAS,EACd,KAAK,8BAA8B,EACnC,KAAK,sBAAsB,EAC3B,eAAe,EACf,KAAK,kBAAkB,EACvB,KAAK,eAAe,EACpB,KAAK,cAAc,EACnB,KAAK,QAAQ,EAEb,MAAM,uBAAuB,CAAC;AAG/B,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,uBAAuB,CAAC;AACvD,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,4BAA4B,CAAC;AAChE,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACzD,OAAO,KAAK,EAAE,kBAAkB,EAAE,MAAM,6BAA6B,CAAC;AACtE,OAAO,KAAK,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AAC3D,OAAO,EAEN,KAAK,6BAA6B,EAClC,KAAK,eAAe,EACpB,MAAM,uBAAuB,CAAC;AAgC/B,MAAM,WAAW,kBAAkB;IAClC,oGAAkG;IAClG,QAAQ,IAAI,KAAK,CAAC;IAClB,sFAAsF;IACtF,MAAM,IAAI,MAAM,CAAC;IACjB,mDAAmD;IACnD,iBAAiB,IAAI,cAAc,CAAC;IACpC,uFAAuF;IACvF,kBAAkB,IAAI,eAAe,CAAC;IACtC,sFAAsF;IACtF,gBAAgB,IAAI,aAAa,CAAC;IAClC,2FAA2F;IAC3F,iBAAiB,IAAI,cAAc,CAAC;IAEpC,qDAAqD;IACrD,kBAAkB,IAAI,eAAe,CAAC;IACtC,+GAA+G;IAC/G,kBAAkB,CAAC,MAAM,EAAE,eAAe,GAAG,IAAI,CAAC;IAClD,qFAAqF;IACrF,mBAAmB,IAAI,MAAM,CAAC;IAC9B,mBAAmB,CAAC,MAAM,EAAE,MAAM,GAAG,IAAI,CAAC;IAE1C,mFAAmF;IACnF,cAAc,IAAI,cAAc,EAAE,CAAC;IACnC,mGAAmG;IACnG,oBAAoB,IAAI,MAAM,CAAC,MAAM,EAAE,SAAS,CAAC,GAAG,SAAS,CAAC;IAC9D,6GAA6G;IAC7G,2BAA2B,IAAI,MAAM,EAAE,GAAG,SAAS,CAAC;IACpD,2BAA2B,CAAC,KAAK,EAAE,MAAM,EAAE,GAAG,SAAS,GAAG,IAAI,CAAC;IAE/D,2GAA2G;IAC3G,oBAAoB,IAAI,QAAQ,CAAC,6BAA6B,CAAC,GAAG,SAAS,CAAC;IAC5E,oBAAoB,CAAC,MAAM,EAAE,QAAQ,CAAC,6BAA6B,CAAC,GAAG,SAAS,GAAG,IAAI,CAAC;IACxF,mBAAmB,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,SAAS,CAAC;IAC/C,oBAAoB,IAAI,GAAG,CAAC,MAAM,CAAC,GAAG,SAAS,CAAC;IAChD,sFAAsF;IACtF,uBAAuB,IAAI,QAAQ,CAAC,6BAA6B,CAAC,CAAC;IACnE,oGAAoG;IACpG,+BAA+B,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAAC;IACvD,kHAAkH;IAClH,0BAA0B,CAAC,UAAU,EAAE,SAAS,EAAE,GAAG,SAAS,EAAE,CAAC;IACjE,iFAAiF;IACjF,2BAA2B,CAAC,QAAQ,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;IAEtD,oGAAoG;IACpG,kBAAkB,IAAI,MAAM,EAAE,CAAC;IAC/B,iGAAiG;IACjG,oBAAoB,CAAC,SAAS,EAAE,MAAM,EAAE,GAAG,IAAI,CAAC;IAChD,wFAAwF;IACxF,sBAAsB,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS,GAAG,MAAM,GAAG,SAAS,CAAC;IACrE,yBAAyB,CAAC,UAAU,EAAE,MAAM,EAAE,GAAG,SAAS,GAAG,MAAM,EAAE,CAAC;IAEtE,0GAA0G;IAC1G,iBAAiB,CAAC,MAAM,EAAE,eAAe,GAAG,IAAI,CAAC;IACjD,qFAAqF;IACrF,sBAAsB,CAAC,MAAM,EAAE,eAAe,GAAG,IAAI,CAAC;IACtD,yEAAyE;IACzE,6BAA6B,CAAC,MAAM,EAAE,SAAS,GAAG,QAAQ,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;IAC3E,yFAAyF;IACzF,iCAAiC,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;IACnD,4EAA4E;IAC5E,uBAAuB,IAAI,IAAI,CAAC;IAEhC,wGAAwG;IACxG,oBAAoB,IAAI,aAAa,CAAC;IAEtC,oEAAkE;IAClE,gBAAgB,IAAI,aAAa,CAAC;IAClC,qFAAqF;IACrF,yBAAyB,IAAI;QAAE,SAAS,EAAE,0BAA0B,CAAC;QAAC,eAAe,EAAE,2BAA2B,CAAA;KAAE,CAAC;IACrH,6FAA6F;IAC7F,2BAA2B,IAAI,IAAI,CAAC;IACpC,wEAAwE;IACxE,gBAAgB,IAAI,OAAO,CAAC,IAAI,CAAC,CAAC;IAElC,iCAAiC;IACjC,oBAAoB,IAAI,SAAS,GAAG,SAAS,CAAC;IAC9C,qBAAqB,CAAC,KAAK,EAAE,SAAS,GAAG,MAAM,CAAC;IAChD,uDAAuD;IACvD,kBAAkB,CAAC,QAAQ,EAAE,YAAY,EAAE,GAAG,eAAe,CAAC;IAC9D,sDAAsD;IACtD,uBAAuB,CAAC,OAAO,EAAE;QAChC,YAAY,EAAE,MAAM,CAAC;QACrB,YAAY,CAAC,EAAE,MAAM,CAAC;KACtB,GAAG,OAAO,CAAC,0BAA0B,CAAC,CAAC;IACxC,sDAAsD;IACtD,eAAe,CAAC,IAAI,EAAE;QACrB,KAAK,EAAE,MAAM,CAAC;QACd,MAAM,CAAC,EAAE,MAAM,CAAC;KAChB,GAAG,OAAO,CAAC;QAAE,OAAO,EAAE,IAAI,CAAC;QAAC,KAAK,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,kBAAkB,CAAA;KAAE,GAAG;QAAE,OAAO,EAAE,KAAK,CAAC;QAAC,UAAU,EAAE,MAAM,CAAA;KAAE,CAAC,CAAC;IACnH,qFAAqF;IACrF,yBAAyB,IAAI,KAAK,CAAC,GAAG,CAAC,GAAG,SAAS,CAAC;IACpD,gFAA8E;IAC9E,qBAAqB,CAAC,IAAI,EAAE,yBAAyB,GAAG,OAAO,CAAC,wBAAwB,CAAC,CAAC;IAC1F,6DAA6D;IAC7D,eAAe,CACd,KAAK,EAAE,KAAK,EACZ,IAAI,CAAC,EAAE;QAAE,KAAK,CAAC,EAAE,MAAM,CAAC;QAAC,eAAe,CAAC,EAAE,MAAM,CAAC;QAAC,QAAQ,CAAC,EAAE,MAAM,CAAA;KAAE,GACpE,MAAM,GAAG,SAAS,CAAC;IAEtB,uFAAuF;IACvF,0BAA0B,IAAI,YAAY,CAAC;IAC3C,eAAe,IAAI,YAAY,GAAG,SAAS,CAAC;IAE5C,0FAA0F;IAC1F,WAAW,IAAI,OAAO,CAAC;IACvB,YAAY,IAAI,OAAO,CAAC;IAExB,mFAAiF;IACjF,qBAAqB,IAAI,kBAAkB,GAAG,SAAS,CAAC;IACxD,iCAAiC,IAAI,8BAA8B,GAAG,SAAS,CAAC;IAChF,2BAA2B,IAAI,eAAe,GAAG,SAAS,CAAC;IAC3D,yBAAyB,IAAI,sBAAsB,GAAG,SAAS,CAAC;CAChE;AAED;;;GAGG;AACH,qBAAa,cAAc;IAE1B,OAAO,CAAC,aAAa,CAAqC;IAC1D,OAAO,CAAC,gBAAgB,CAA+C;IACvE,OAAO,CAAC,mBAAmB,CAAkC;IAC7D,OAAO,CAAC,qBAAqB,CAAoC;IACjE,OAAO,CAAC,oBAAoB,CAA0C;IAEtE,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAqB;IAE1C,YAAY,IAAI,EAAE,kBAAkB,EAEnC;IAED,mEAAmE;IACnE,OAAO,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,CAE7B;IAED,2EAA2E;IAC3E,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,SAAS,GAAG,SAAS,CAErD;IAED,6DAA6D;IAC7D,oBAAoB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,SAAS,CAErD;IAED,gEAAgE;IAChE,uBAAuB,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,EAAE,GAAG,SAAS,CAE1D;IAED;;OAEG;IACH,WAAW,IAAI,QAAQ,EAAE,CAQxB;IAED,iBAAiB,CAAC,IAAI,EAAE,MAAM,GAAG,cAAc,GAAG,SAAS,CAE1D;IAED,mBAAmB,CAAC,OAAO,CAAC,EAAE;QAAE,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;QAAC,wBAAwB,CAAC,EAAE,OAAO,CAAA;KAAE,GAAG,IAAI,CA2JtG;IAED,OAAO,CAAC,4BAA4B;IAepC,OAAO,CAAC,6BAA6B;IAiBrC,OAAO,CAAC,oBAAoB;IA2B5B,YAAY,CAAC,OAAO,EAAE;QACrB,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC;QAC3B,UAAU,CAAC,EAAE,GAAG,CAAC,MAAM,EAAE,OAAO,GAAG,MAAM,CAAC,CAAC;QAC3C,wBAAwB,CAAC,EAAE,OAAO,CAAC;KACnC,GAAG,IAAI,CAsHP;IAEK,MAAM,IAAI,OAAO,CAAC,IAAI,CAAC,CAyE5B;IAED;;;;;OAKG;IACG,mBAAmB,CAAC,aAAa,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAmD9D;IAED;;;;;OAKG;IACG,iBAAiB,CAAC,aAAa,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC,CAuC5D;IAED;;;;OAIG;IACG,yBAAyB,IAAI,OAAO,CAAC,IAAI,CAAC,CA+E/C;CACD","sourcesContent":["/**\n * Runtime build & reload: the session's tool-registry assembly and the self-modification-safe\n * extension reload path. Owns building the base tool definitions, wrapping them into the live tool\n * registry (`_refreshToolRegistry`), constructing the {@link ExtensionRunner} for a rebuilt runtime\n * (`_buildRuntime`), and the repo's self-modification safety crown jewel: the preflight-guarded\n * `reload()` with its snapshot / doctor / commit-or-rollback sequence, plus the single-extension\n * live load/unload/reconcile operations that rebuild the runtime in place.\n *\n * Extracted verbatim from agent-session.ts (god-file decomposition). Owns ONLY the tool-registry\n * state — the base tool definitions, the wrapped tool registry, the definition registry, and the\n * per-tool prompt snippet/guideline maps. Everything else the build & reload touch — the live\n * {@link ExtensionRunner}, the agent's tool/system-prompt state, the base system prompt, the\n * resource loader, the session/settings managers, the model registry, the profile tool filter and\n * its warning sinks, the requested-active-tool-names request, the memory subsystem, and the many\n * host callbacks the rebuilt tools/runtime are wired to — is reached through narrow deps accessors\n * that read and write the SAME storage the host owns today.\n *\n * Snapshot-ownership boundary (deliberate, load-bearing for reload safety): the reload snapshot\n * spans state owned by OTHER collaborators — the extension runner and its ref, `agent.state.tools`\n * / `agent.state.systemPrompt`, and `_baseSystemPrompt`. Those are captured and restored through\n * {@link RuntimeBuilderDeps} get/set accessors so save/restore mutate exactly the host/agent fields\n * they mutated before extraction; only the five tool-registry maps are captured by direct field\n * reference here (this builder owns them). The `_extensionRunnerRef.current` update folds into\n * {@link RuntimeBuilderDeps.setExtensionRunner} (the same field-then-ref pattern the host used at\n * every assignment site), so no observer runs between the two writes.\n *\n * Host-binding boundary (deliberate): `_bindExtensionCore` — which exposes the SESSION's own public\n * surface (sendMessage, setModel, compact, abort, reload, …) to the extension runner — stays\n * host-side and is invoked from {@link buildRuntime} via {@link RuntimeBuilderDeps.bindExtensionCore};\n * it is host identity, not build logic, and moving it would only re-export ~30 host methods through\n * deps to hand them straight back. The host keeps one-line delegations for the public reload /\n * load / unload / reconcile API and for `getAllTools` / `getToolDefinition`, and a private\n * `_refreshToolRegistry` delegation for its internal callers.\n */\n\nimport type { Agent, AgentContext, AgentMessage, AgentTool } from \"@caupulican/pi-agent-core\";\nimport type { SessionManager } from \"@caupulican/pi-agent-core/node\";\nimport type { Api, Model, Usage } from \"@caupulican/pi-ai\";\nimport { resetApiProviders } from \"@caupulican/pi-ai\";\nimport type {\n\tIsolatedCompletionOptions,\n\tIsolatedCompletionResult,\n\tWorkerDelegationRunOutcome,\n} from \"./agent-session.ts\";\nimport type { ArtifactStore } from \"./context/context-artifacts.ts\";\nimport type { MemoryPromptInclusionReport, MemoryRetrievalDiagnostics } from \"./context/memory-diagnostics.ts\";\nimport type { ContextGcReport } from \"./context-gc.ts\";\nimport { createCoreDiagnosticsToolDefinitions } from \"./extensions/builtin.ts\";\nimport {\n\ttype ContextUsage,\n\ttype Extension,\n\ttype ExtensionCommandContextActions,\n\ttype ExtensionErrorListener,\n\tExtensionRunner,\n\ttype ExtensionUIContext,\n\ttype ShutdownHandler,\n\ttype ToolDefinition,\n\ttype ToolInfo,\n\twrapRegisteredTools,\n} from \"./extensions/index.ts\";\nimport { disposeExtensionEventSubscriptions } from \"./extensions/loader.ts\";\nimport { emitSessionShutdownEvent } from \"./extensions/runner.ts\";\nimport type { GoalState } from \"./goals/goal-state.ts\";\nimport type { MemoryManager } from \"./memory/memory-manager.ts\";\nimport type { ModelRegistry } from \"./model-registry.ts\";\nimport type { ModelFitnessReport } from \"./research/model-fitness.ts\";\nimport type { ResourceLoader } from \"./resource-loader.ts\";\nimport {\n\tmatchesResourceProfilePattern,\n\ttype ResourceProfileFilterSettings,\n\ttype SettingsManager,\n} from \"./settings-manager.ts\";\nimport { createSyntheticSourceInfo, type SourceInfo } from \"./source-info.ts\";\nimport {\n\tbuildReflexUserPrompt,\n\tparseReflexPlan,\n\tREFLEX_INTERPRETER_SYSTEM_PROMPT,\n} from \"./toolkit/reflex-interpreter.ts\";\nimport { executeToolkitScript } from \"./toolkit/script-runner.ts\";\nimport { createDelegateToolDefinition } from \"./tools/delegate.ts\";\nimport { createGoalToolDefinition } from \"./tools/goal.ts\";\nimport { createAllToolDefinitions } from \"./tools/index.ts\";\nimport { createModelFitnessToolDefinition } from \"./tools/model-fitness.ts\";\nimport { createRunToolkitScriptToolDefinition } from \"./tools/run-toolkit-script.ts\";\nimport { createToolDefinitionFromAgentTool } from \"./tools/tool-definition-wrapper.ts\";\n\ninterface ToolDefinitionEntry {\n\tdefinition: ToolDefinition;\n\tsourceInfo: SourceInfo;\n}\n\ninterface ReloadRuntimeSnapshot {\n\textensionRunner: ExtensionRunner;\n\tbaseToolDefinitions: Map<string, ToolDefinition>;\n\ttoolRegistry: Map<string, AgentTool>;\n\ttoolDefinitions: Map<string, ToolDefinitionEntry>;\n\ttoolPromptSnippets: Map<string, string>;\n\ttoolPromptGuidelines: Map<string, string[]>;\n\tagentTools: AgentTool[];\n\tagentSystemPrompt: string;\n\tbaseSystemPrompt: string;\n}\n\nexport interface RuntimeBuilderDeps {\n\t/** Live agent — the snapshot/doctor read and restore its `state.tools` / `state.systemPrompt`. */\n\tgetAgent(): Agent;\n\t/** Workspace root, passed to the tool-definition factory and the extension runner. */\n\tgetCwd(): string;\n\t/** Session log, passed to the extension runner. */\n\tgetSessionManager(): SessionManager;\n\t/** Tool/shell/toolkit/resource settings + reload target (settingsManager.reload()). */\n\tgetSettingsManager(): SettingsManager;\n\t/** Model registry, passed to the extension runner and profile model re-resolution. */\n\tgetModelRegistry(): ModelRegistry;\n\t/** Extension/skill/prompt/theme discovery + the reload/commit/rollback generation swap. */\n\tgetResourceLoader(): ResourceLoader;\n\n\t/** Live extension runner (host-owned; pervasive). */\n\tgetExtensionRunner(): ExtensionRunner;\n\t/** Store the extension runner AND update the Agent's mutable `_extensionRunnerRef.current` (both together). */\n\tsetExtensionRunner(runner: ExtensionRunner): void;\n\t/** Base (extension-free) system prompt; captured/restored by the reload snapshot. */\n\tgetBaseSystemPrompt(): string;\n\tsetBaseSystemPrompt(prompt: string): void;\n\n\t/** SDK-provided plain tools, synthesized into the registry alongside built-ins. */\n\tgetCustomTools(): ToolDefinition[];\n\t/** Optional plain-tool override; when set, the built-in factory + core diagnostics are skipped. */\n\tgetBaseToolsOverride(): Record<string, AgentTool> | undefined;\n\t/** Pre-filter tool REQUEST (never the capability/profile-filtered active set) preserved across a rebuild. */\n\tgetRequestedActiveToolNames(): string[] | undefined;\n\tsetRequestedActiveToolNames(names: string[] | undefined): void;\n\n\t/** Resource-profile tool allow/block filter + the raw allow/exclude sets applied during registry build. */\n\tgetToolProfileFilter(): Required<ResourceProfileFilterSettings> | undefined;\n\tsetToolProfileFilter(filter: Required<ResourceProfileFilterSettings> | undefined): void;\n\tgetAllowedToolNames(): Set<string> | undefined;\n\tgetExcludedToolNames(): Set<string> | undefined;\n\t/** Re-derive the profile tool filter from freshly reloaded settings (reload only). */\n\tderiveToolProfileFilter(): Required<ResourceProfileFilterSettings>;\n\t/** True when a tool/command name survives the active profile's allow/block + user allow/exclude. */\n\tisToolOrCommandAllowedByProfile(name: string): boolean;\n\t/** Filter the loaded extensions through the active resource profile (records inert/denied warnings host-side). */\n\tfilterExtensionsForRuntime(extensions: Extension[]): Extension[];\n\t/** Sink for the G13 unbound-profile-tool-grant warnings surfaced in /context. */\n\tsetUnboundToolGrantWarnings(warnings: string[]): void;\n\n\t/** Currently-active tool names (reads agent.state.tools; the pre-filter fallback for a rebuild). */\n\tgetActiveToolNames(): string[];\n\t/** Apply the recomputed active set (capability filter + companion auto-activation live here). */\n\tsetActiveToolsByName(toolNames: string[]): void;\n\t/** Normalize a tool's prompt snippet / guidelines through the system-prompt builder. */\n\tnormalizePromptSnippet(text: string | undefined): string | undefined;\n\tnormalizePromptGuidelines(guidelines: string[] | undefined): string[];\n\n\t/** Wire the session's own public surface into a freshly-built runner (host identity; stays host-side). */\n\tbindExtensionCore(runner: ExtensionRunner): void;\n\t/** Re-apply UI context / mode / command-context / error subscription to a runner. */\n\tapplyExtensionBindings(runner: ExtensionRunner): void;\n\t/** Re-run resource discovery for the active extensions (reload only). */\n\textendResourcesFromExtensions(reason: \"startup\" | \"reload\"): Promise<void>;\n\t/** Re-apply the active profile's model/thinking from reloaded settings (reload only). */\n\treapplyActiveProfileModelSettings(): Promise<void>;\n\t/** Notify extensions-changed listeners after a single-extension live op. */\n\tnotifyExtensionsChanged(): void;\n\n\t/** Session-scoped tool-output artifact store for grep/find/artifact_retrieve (gated on the profile). */\n\tgetToolArtifactStore(): ArtifactStore;\n\n\t/** Live memory manager — its provider tools join the registry. */\n\tgetMemoryManager(): MemoryManager;\n\t/** Memory retrieval + prompt-inclusion diagnostics for the core diagnostics tool. */\n\tgetMemoryAuditDiagnostics(): { retrieval: MemoryRetrievalDiagnostics; promptInclusion: MemoryPromptInclusionReport };\n\t/** Drop extension-contributed pending memory providers before a reload re-registers them. */\n\tclearPendingMemoryProviders(): void;\n\t/** (Re)derive the memory subsystem from reloaded settings/providers. */\n\tinitializeMemory(): Promise<void>;\n\n\t/** Goal-tool state accessors. */\n\tgetGoalStateSnapshot(): GoalState | undefined;\n\tsaveGoalStateSnapshot(state: GoalState): string;\n\t/** Context-gc report for the core diagnostics tool. */\n\tgetContextGcReport(messages: AgentMessage[]): ContextGcReport;\n\t/** Worker-delegation runner for the delegate tool. */\n\trunWorkerDelegationOnce(request: {\n\t\tinstructions: string;\n\t\tsystemPrompt?: string;\n\t}): Promise<WorkerDelegationRunOutcome>;\n\t/** Model-fitness probe for the model_fitness tool. */\n\trunModelFitness(args: {\n\t\tmodel: string;\n\t\ttrials?: number;\n\t}): Promise<{ started: true; model: string; report: ModelFitnessReport } | { started: false; skipReason: string }>;\n\t/** Fitness-gated reflex-brain model resolver (run_toolkit_script interpretation). */\n\tresolveCurationModelIfFit(): Model<Api> | undefined;\n\t/** One-shot, tool-less LLM call — the reflex-brain interpreter rides this. */\n\trunIsolatedCompletion(opts: IsolatedCompletionOptions): Promise<IsolatedCompletionResult>;\n\t/** Roll reflex-brain spend into spawned-usage accounting. */\n\taddSpawnedUsage(\n\t\tusage: Usage,\n\t\topts?: { label?: string; sourceSessionId?: string; reportId?: string },\n\t): string | undefined;\n\n\t/** Post-rebuild doctor helpers (validate the rebuilt runtime can render a context). */\n\tcreateAgentContextSnapshot(): AgentContext;\n\tgetContextUsage(): ContextUsage | undefined;\n\n\t/** Reload/live-op preflight refusal guards (identical refusal behavior; host-checked). */\n\tisStreaming(): boolean;\n\tisCompacting(): boolean;\n\n\t/** Extension bindings present — gate for re-emitting session_start on reload. */\n\tgetExtensionUIContext(): ExtensionUIContext | undefined;\n\tgetExtensionCommandContextActions(): ExtensionCommandContextActions | undefined;\n\tgetExtensionShutdownHandler(): ShutdownHandler | undefined;\n\tgetExtensionErrorListener(): ExtensionErrorListener | undefined;\n}\n\n/**\n * Owns the tool-registry build and the self-modification-safe extension reload extracted from\n * {@link AgentSession}. See the module header for the snapshot-ownership and host-binding boundaries.\n */\nexport class RuntimeBuilder {\n\t// Tool registry for extension getTools/setTools\n\tprivate _toolRegistry: Map<string, AgentTool> = new Map();\n\tprivate _toolDefinitions: Map<string, ToolDefinitionEntry> = new Map();\n\tprivate _toolPromptSnippets: Map<string, string> = new Map();\n\tprivate _toolPromptGuidelines: Map<string, string[]> = new Map();\n\tprivate _baseToolDefinitions: Map<string, ToolDefinition> = new Map();\n\n\tprivate readonly deps: RuntimeBuilderDeps;\n\n\tconstructor(deps: RuntimeBuilderDeps) {\n\t\tthis.deps = deps;\n\t}\n\n\t/** Whether a tool name is present in the live wrapped registry. */\n\thasTool(name: string): boolean {\n\t\treturn this._toolRegistry.has(name);\n\t}\n\n\t/** The live wrapped tool for a name, if registered (activation lookup). */\n\tgetRegisteredTool(name: string): AgentTool | undefined {\n\t\treturn this._toolRegistry.get(name);\n\t}\n\n\t/** A registered tool's normalized prompt snippet, if any. */\n\tgetToolPromptSnippet(name: string): string | undefined {\n\t\treturn this._toolPromptSnippets.get(name);\n\t}\n\n\t/** A registered tool's normalized prompt guidelines, if any. */\n\tgetToolPromptGuidelines(name: string): string[] | undefined {\n\t\treturn this._toolPromptGuidelines.get(name);\n\t}\n\n\t/**\n\t * Get all configured tools with name, description, parameter schema, prompt guidelines, and source metadata.\n\t */\n\tgetAllTools(): ToolInfo[] {\n\t\treturn Array.from(this._toolDefinitions.values()).map(({ definition, sourceInfo }) => ({\n\t\t\tname: definition.name,\n\t\t\tdescription: definition.description,\n\t\t\tparameters: definition.parameters,\n\t\t\tpromptGuidelines: definition.promptGuidelines,\n\t\t\tsourceInfo,\n\t\t}));\n\t}\n\n\tgetToolDefinition(name: string): ToolDefinition | undefined {\n\t\treturn this._toolDefinitions.get(name)?.definition;\n\t}\n\n\trefreshToolRegistry(options?: { activeToolNames?: string[]; includeAllExtensionTools?: boolean }): void {\n\t\tconst previousRegistryNames = new Set(this._toolRegistry.keys());\n\t\t// Re-derive from the pre-filter REQUEST, never from agent.state.tools: the active set is\n\t\t// capability/profile-filtered, so feeding it back through setActiveToolsByName would\n\t\t// permanently shrink what a later switch to a larger model (or permissive profile) restores.\n\t\tconst previousActiveToolNames = this.deps.getRequestedActiveToolNames() ?? this.deps.getActiveToolNames();\n\t\tconst allowedToolNames = this.deps.getAllowedToolNames();\n\t\tconst excludedToolNames = this.deps.getExcludedToolNames();\n\t\tconst toolProfileFilter = this.deps.getToolProfileFilter();\n\t\tconst isAllowedTool = (name: string): boolean => {\n\t\t\tif (allowedToolNames && !allowedToolNames.has(name)) return false;\n\t\t\tif (excludedToolNames?.has(name)) return false;\n\t\t\tif (!toolProfileFilter) return true;\n\t\t\tif (toolProfileFilter.allow.length > 0 && !matchesResourceProfilePattern(name, toolProfileFilter.allow)) {\n\t\t\t\treturn false;\n\t\t\t}\n\t\t\tif (matchesResourceProfilePattern(name, toolProfileFilter.block)) return false;\n\t\t\treturn true;\n\t\t};\n\n\t\tconst registeredTools = this.deps.getExtensionRunner().getAllRegisteredTools();\n\t\tconst allCustomTools = [\n\t\t\t...registeredTools,\n\t\t\t...this.deps.getCustomTools().map((definition) => ({\n\t\t\t\tdefinition,\n\t\t\t\tsourceInfo: createSyntheticSourceInfo(`<sdk:${definition.name}>`, { source: \"sdk\" }),\n\t\t\t})),\n\t\t\t// Memory subsystem provider tools (e.g. file-store's `memory` tool).\n\t\t\t...this.deps\n\t\t\t\t.getMemoryManager()\n\t\t\t\t.getToolDefinitions()\n\t\t\t\t.map((definition) => ({\n\t\t\t\t\tdefinition,\n\t\t\t\t\tsourceInfo: createSyntheticSourceInfo(`<memory:${definition.name}>`, { source: \"sdk\" }),\n\t\t\t\t})),\n\t\t].filter((tool) => isAllowedTool(tool.definition.name));\n\t\tconst definitionRegistry = new Map<string, ToolDefinitionEntry>(\n\t\t\tArray.from(this._baseToolDefinitions.entries())\n\t\t\t\t.filter(([name]) => isAllowedTool(name))\n\t\t\t\t.map(([name, definition]) => [\n\t\t\t\t\tname,\n\t\t\t\t\t{\n\t\t\t\t\t\tdefinition,\n\t\t\t\t\t\tsourceInfo: createSyntheticSourceInfo(`<builtin:${name}>`, { source: \"builtin\" }),\n\t\t\t\t\t},\n\t\t\t\t]),\n\t\t);\n\t\tfor (const tool of allCustomTools) {\n\t\t\tdefinitionRegistry.set(tool.definition.name, {\n\t\t\t\tdefinition: tool.definition,\n\t\t\t\tsourceInfo: tool.sourceInfo,\n\t\t\t});\n\t\t}\n\t\tthis._toolDefinitions = definitionRegistry;\n\t\tthis._toolPromptSnippets = new Map(\n\t\t\tArray.from(definitionRegistry.values())\n\t\t\t\t.map(({ definition }) => {\n\t\t\t\t\tconst snippet = this.deps.normalizePromptSnippet(definition.promptSnippet);\n\t\t\t\t\treturn snippet ? ([definition.name, snippet] as const) : undefined;\n\t\t\t\t})\n\t\t\t\t.filter((entry): entry is readonly [string, string] => entry !== undefined),\n\t\t);\n\t\tthis._toolPromptGuidelines = new Map(\n\t\t\tArray.from(definitionRegistry.values())\n\t\t\t\t.map(({ definition }) => {\n\t\t\t\t\tconst guidelines = this.deps.normalizePromptGuidelines(definition.promptGuidelines);\n\t\t\t\t\treturn guidelines.length > 0 ? ([definition.name, guidelines] as const) : undefined;\n\t\t\t\t})\n\t\t\t\t.filter((entry): entry is readonly [string, string[]] => entry !== undefined),\n\t\t);\n\t\tconst runner = this.deps.getExtensionRunner();\n\t\tconst wrappedExtensionTools = wrapRegisteredTools(allCustomTools, runner);\n\t\tconst wrappedBuiltInTools = wrapRegisteredTools(\n\t\t\tArray.from(this._baseToolDefinitions.values())\n\t\t\t\t.filter((definition) => isAllowedTool(definition.name))\n\t\t\t\t.map((definition) => ({\n\t\t\t\t\tdefinition,\n\t\t\t\t\tsourceInfo: createSyntheticSourceInfo(`<builtin:${definition.name}>`, { source: \"builtin\" }),\n\t\t\t\t})),\n\t\t\trunner,\n\t\t);\n\n\t\tconst toolRegistry = new Map(wrappedBuiltInTools.map((tool) => [tool.name, tool]));\n\t\tfor (const tool of wrappedExtensionTools as AgentTool[]) {\n\t\t\ttoolRegistry.set(tool.name, tool);\n\t\t}\n\t\tthis._toolRegistry = toolRegistry;\n\n\t\tconst requestedBase = options?.activeToolNames ? [...options.activeToolNames] : [...previousActiveToolNames];\n\t\tconst nextActiveToolNames = requestedBase.filter((name) => isAllowedTool(name));\n\n\t\tconst autoActivated: string[] = [];\n\t\tif (allowedToolNames) {\n\t\t\tfor (const toolName of this._toolRegistry.keys()) {\n\t\t\t\tif (allowedToolNames.has(toolName)) {\n\t\t\t\t\tnextActiveToolNames.push(toolName);\n\t\t\t\t\tautoActivated.push(toolName);\n\t\t\t\t}\n\t\t\t}\n\t\t} else if (options?.includeAllExtensionTools) {\n\t\t\tfor (const tool of wrappedExtensionTools) {\n\t\t\t\tnextActiveToolNames.push(tool.name);\n\t\t\t\tautoActivated.push(tool.name);\n\t\t\t}\n\t\t} else if (!options?.activeToolNames) {\n\t\t\tfor (const toolName of this._toolRegistry.keys()) {\n\t\t\t\tif (!previousRegistryNames.has(toolName)) {\n\t\t\t\t\tnextActiveToolNames.push(toolName);\n\t\t\t\t\tautoActivated.push(toolName);\n\t\t\t\t}\n\t\t\t}\n\t\t}\n\n\t\t// Strict UAC: the active profile is the COMPLETE grant, so a tool the profile names\n\t\t// explicitly is itself a request for that tool — it must ACTIVATE from the registry even\n\t\t// if the session never requested it. Without this, activation is only ever the requested\n\t\t// defaults ∩ allow-list, and a profile granting non-default tools (a search-only profile's\n\t\t// grep/find) yields an empty or truncated tool set on load and /reload. A blanket \"*\"\n\t\t// stays grant-only: activation then still derives from the request/defaults above.\n\t\tconst explicitAllowPatterns = toolProfileFilter?.allow.filter((pattern) => pattern !== \"*\") ?? [];\n\t\tif (explicitAllowPatterns.length > 0) {\n\t\t\tconst boundPatterns = new Set<string>();\n\t\t\tfor (const toolName of this._toolRegistry.keys()) {\n\t\t\t\tif (!isAllowedTool(toolName)) continue;\n\t\t\t\tfor (const pattern of explicitAllowPatterns) {\n\t\t\t\t\tif (matchesResourceProfilePattern(toolName, [pattern])) boundPatterns.add(pattern);\n\t\t\t\t}\n\t\t\t\tif (matchesResourceProfilePattern(toolName, explicitAllowPatterns)) {\n\t\t\t\t\tnextActiveToolNames.push(toolName);\n\t\t\t\t\tautoActivated.push(toolName);\n\t\t\t\t}\n\t\t\t}\n\t\t\t// G13: an explicit grant that binds to NO registered tool is a silent no-op — typo'd\n\t\t\t// name, or the owning extension is not granted/loaded. Surface it.\n\t\t\tthis.deps.setUnboundToolGrantWarnings(\n\t\t\t\texplicitAllowPatterns\n\t\t\t\t\t.filter((pattern) => !boundPatterns.has(pattern))\n\t\t\t\t\t.map(\n\t\t\t\t\t\t(pattern) =>\n\t\t\t\t\t\t\t`profile tool grant \"${pattern}\" binds to no registered tool (typo, or the owning extension is not granted/loaded)`,\n\t\t\t\t\t),\n\t\t\t);\n\t\t} else {\n\t\t\tthis.deps.setUnboundToolGrantWarnings([]);\n\t\t}\n\n\t\t// artifact_retrieve companion auto-activation is enforced inside\n\t\t// setActiveToolsByName() itself (not duplicated here), so every activation path --\n\t\t// including the public, extension-exposed setActiveTools() -- gets the same\n\t\t// guarantee, not just this settings/profile refresh flow.\n\t\tthis.deps.setActiveToolsByName([...new Set(nextActiveToolNames)]);\n\t\t// setActiveToolsByName just stored the profile-filtered ACTIVE set as the request; restore\n\t\t// the true pre-filter request (plus this refresh's auto-activations) so an internal refresh\n\t\t// can never permanently narrow it.\n\t\tthis.deps.setRequestedActiveToolNames([...new Set([...requestedBase, ...autoActivated])]);\n\t}\n\n\tprivate _createReloadRuntimeSnapshot(): ReloadRuntimeSnapshot {\n\t\tconst agent = this.deps.getAgent();\n\t\treturn {\n\t\t\textensionRunner: this.deps.getExtensionRunner(),\n\t\t\tbaseToolDefinitions: this._baseToolDefinitions,\n\t\t\ttoolRegistry: this._toolRegistry,\n\t\t\ttoolDefinitions: this._toolDefinitions,\n\t\t\ttoolPromptSnippets: this._toolPromptSnippets,\n\t\t\ttoolPromptGuidelines: this._toolPromptGuidelines,\n\t\t\tagentTools: agent.state.tools,\n\t\t\tagentSystemPrompt: agent.state.systemPrompt,\n\t\t\tbaseSystemPrompt: this.deps.getBaseSystemPrompt(),\n\t\t};\n\t}\n\n\tprivate _restoreReloadRuntimeSnapshot(snapshot: ReloadRuntimeSnapshot): void {\n\t\t// setExtensionRunner restores both _extensionRunner and _extensionRunnerRef.current together\n\t\t// (the same field-then-ref pair the host wrote at every assignment site); nothing reads the\n\t\t// ref between the two writes, so folding them is unobservable.\n\t\tthis.deps.setExtensionRunner(snapshot.extensionRunner);\n\t\tthis._baseToolDefinitions = snapshot.baseToolDefinitions;\n\t\tthis._toolRegistry = snapshot.toolRegistry;\n\t\tthis._toolDefinitions = snapshot.toolDefinitions;\n\t\tthis._toolPromptSnippets = snapshot.toolPromptSnippets;\n\t\tthis._toolPromptGuidelines = snapshot.toolPromptGuidelines;\n\t\tconst agent = this.deps.getAgent();\n\t\tagent.state.tools = snapshot.agentTools;\n\t\tagent.state.systemPrompt = snapshot.agentSystemPrompt;\n\t\tthis.deps.setBaseSystemPrompt(snapshot.baseSystemPrompt);\n\t\tthis.deps.applyExtensionBindings(snapshot.extensionRunner);\n\t}\n\n\tprivate _doctorReloadRuntime(): void {\n\t\tconst extensionErrors = this.deps.getResourceLoader().getExtensions().errors;\n\t\tif (extensionErrors.length > 0) {\n\t\t\tconst summary = extensionErrors\n\t\t\t\t.slice(0, 6)\n\t\t\t\t.map((error) => `${error.path}: ${error.error}`)\n\t\t\t\t.join(\"; \");\n\t\t\tthrow new Error(`Extension reload failed doctor: ${summary}`);\n\t\t}\n\n\t\tconst missingActiveTools = this.deps.getActiveToolNames().filter((name) => !this._toolRegistry.has(name));\n\t\tif (missingActiveTools.length > 0) {\n\t\t\tthrow new Error(\n\t\t\t\t`Extension reload failed doctor: active tool(s) missing after reload: ${missingActiveTools.join(\", \")}`,\n\t\t\t);\n\t\t}\n\n\t\tfor (const tool of this.deps.getAgent().state.tools) {\n\t\t\tif (!this._toolDefinitions.has(tool.name)) {\n\t\t\t\tthrow new Error(`Extension reload failed doctor: tool ${tool.name} missing from definition registry`);\n\t\t\t}\n\t\t}\n\n\t\tthis.deps.createAgentContextSnapshot();\n\t\tthis.deps.getContextUsage();\n\t}\n\n\tbuildRuntime(options: {\n\t\tactiveToolNames?: string[];\n\t\tflagValues?: Map<string, boolean | string>;\n\t\tincludeAllExtensionTools?: boolean;\n\t}): void {\n\t\tconst settingsManager = this.deps.getSettingsManager();\n\t\tconst autoResizeImages = settingsManager.getImageAutoResize();\n\t\tconst shellCommandPrefix = settingsManager.getShellCommandPrefix();\n\t\tconst shellPath = settingsManager.getShellPath();\n\t\tconst baseToolsOverride = this.deps.getBaseToolsOverride();\n\t\t// grep/find must not emit a \"Full output: artifact tool-output:<id>\" handle that\n\t\t// nothing can resolve. If artifact_retrieve is explicitly excluded/blocked/outside\n\t\t// an active allowlist, don't hand grep/find an artifact store at all: they fall\n\t\t// back to their pre-existing bounded preview/truncation behavior, with no\n\t\t// payload/meta files ever written and no retrieval promise made.\n\t\tconst toolArtifactStore = this.deps.isToolOrCommandAllowedByProfile(\"artifact_retrieve\")\n\t\t\t? this.deps.getToolArtifactStore()\n\t\t\t: undefined;\n\t\tconst baseToolDefinitions = baseToolsOverride\n\t\t\t? Object.fromEntries(\n\t\t\t\t\tObject.entries(baseToolsOverride).map(([name, tool]) => [name, createToolDefinitionFromAgentTool(tool)]),\n\t\t\t\t)\n\t\t\t: createAllToolDefinitions(this.deps.getCwd(), {\n\t\t\t\t\tread: { autoResizeImages },\n\t\t\t\t\tbash: { commandPrefix: shellCommandPrefix, shellPath },\n\t\t\t\t\tgrep: { artifactStore: toolArtifactStore },\n\t\t\t\t\tfind: { artifactStore: toolArtifactStore },\n\t\t\t\t\tartifact_retrieve: { artifactStore: toolArtifactStore },\n\t\t\t\t});\n\n\t\tthis._baseToolDefinitions = new Map(\n\t\t\tObject.entries(baseToolDefinitions).map(([name, tool]) => [name, tool as ToolDefinition]),\n\t\t);\n\t\tif (!baseToolsOverride) {\n\t\t\tfor (const definition of createCoreDiagnosticsToolDefinitions(\n\t\t\t\t() => this.deps.getActiveToolNames(),\n\t\t\t\t() => this.getAllTools(),\n\t\t\t\t(messages) => this.deps.getContextGcReport(messages),\n\t\t\t\t() => this.deps.getMemoryAuditDiagnostics(),\n\t\t\t)) {\n\t\t\t\tthis._baseToolDefinitions.set(definition.name, definition);\n\t\t\t}\n\t\t\tconst goalToolDefinition = createGoalToolDefinition({\n\t\t\t\tgetGoalState: () => this.deps.getGoalStateSnapshot(),\n\t\t\t\tsaveGoalState: (state) => {\n\t\t\t\t\tthis.deps.saveGoalStateSnapshot(state);\n\t\t\t\t},\n\t\t\t});\n\t\t\tthis._baseToolDefinitions.set(goalToolDefinition.name, goalToolDefinition);\n\t\t\tconst delegateToolDefinition = createDelegateToolDefinition({\n\t\t\t\trunWorkerDelegation: (args) => this.deps.runWorkerDelegationOnce(args),\n\t\t\t});\n\t\t\tthis._baseToolDefinitions.set(delegateToolDefinition.name, delegateToolDefinition);\n\t\t\t// Registered but not default-active: probes spend tokens on the probed model, so\n\t\t\t// activation is an explicit choice (settings/profile/setActiveTools or /autonomy fitness).\n\t\t\tconst modelFitnessToolDefinition = createModelFitnessToolDefinition({\n\t\t\t\trunProbe: (args) => this.deps.runModelFitness(args),\n\t\t\t});\n\t\t\tthis._baseToolDefinitions.set(modelFitnessToolDefinition.name, modelFitnessToolDefinition);\n\t\t\tconst runToolkitScriptToolDefinition = createRunToolkitScriptToolDefinition({\n\t\t\t\tgetScripts: () => this.deps.getSettingsManager().getToolkitScripts(),\n\t\t\t\texecute: (script, scriptArgs) => executeToolkitScript({ script, scriptArgs, cwd: this.deps.getCwd() }),\n\t\t\t\t// Reflex brain (fitness-gated local model): resolves ambiguous requests into a\n\t\t\t\t// registry pick. Best-effort — absent/unfit brain keeps the shortlist behavior.\n\t\t\t\tinterpret: async (request, scripts) => {\n\t\t\t\t\tconst model = this.deps.resolveCurationModelIfFit();\n\t\t\t\t\tif (!model) return undefined;\n\t\t\t\t\tconst completion = await this.deps.runIsolatedCompletion({\n\t\t\t\t\t\tsystemPrompt: REFLEX_INTERPRETER_SYSTEM_PROMPT,\n\t\t\t\t\t\tmessages: [\n\t\t\t\t\t\t\t{\n\t\t\t\t\t\t\t\trole: \"user\",\n\t\t\t\t\t\t\t\tcontent: [{ type: \"text\", text: buildReflexUserPrompt(request, scripts) }],\n\t\t\t\t\t\t\t\ttimestamp: Date.now(),\n\t\t\t\t\t\t\t},\n\t\t\t\t\t\t],\n\t\t\t\t\t\tmodel,\n\t\t\t\t\t\tthinkingLevel: \"off\",\n\t\t\t\t\t\tmaxTokens: 256,\n\t\t\t\t\t\tcacheRetention: \"short\",\n\t\t\t\t\t});\n\t\t\t\t\tif (completion.usage.cost.total > 0 || completion.usage.totalTokens > 0) {\n\t\t\t\t\t\tthis.deps.addSpawnedUsage(completion.usage, { label: \"toolkit-brain\" });\n\t\t\t\t\t}\n\t\t\t\t\treturn parseReflexPlan(completion.text);\n\t\t\t\t},\n\t\t\t});\n\t\t\tthis._baseToolDefinitions.set(runToolkitScriptToolDefinition.name, runToolkitScriptToolDefinition);\n\t\t}\n\n\t\tconst extensionsResult = this.deps.getResourceLoader().getExtensions();\n\t\tif (options.flagValues) {\n\t\t\tfor (const [name, value] of options.flagValues) {\n\t\t\t\textensionsResult.runtime.flagValues.set(name, value);\n\t\t\t}\n\t\t}\n\t\tconst extensions = this.deps.filterExtensionsForRuntime(extensionsResult.extensions);\n\t\tconst runtimeExtensionPaths = new Set(extensions.map((extension) => extension.path));\n\t\textensionsResult.runtime.pendingProviderRegistrations =\n\t\t\textensionsResult.runtime.pendingProviderRegistrations.filter((registration) =>\n\t\t\t\truntimeExtensionPaths.has(registration.extensionPath),\n\t\t\t);\n\n\t\tconst runner = new ExtensionRunner(\n\t\t\textensions,\n\t\t\textensionsResult.runtime,\n\t\t\tthis.deps.getCwd(),\n\t\t\tthis.deps.getSessionManager(),\n\t\t\tthis.deps.getModelRegistry(),\n\t\t);\n\t\tthis.deps.setExtensionRunner(runner);\n\t\tthis.deps.bindExtensionCore(runner);\n\t\tthis.deps.applyExtensionBindings(runner);\n\n\t\tconst defaultActiveToolNames = baseToolsOverride\n\t\t\t? Object.keys(baseToolsOverride)\n\t\t\t: [\"read\", \"bash\", \"edit\", \"write\", \"context_audit\", \"goal\", \"delegate\", \"run_toolkit_script\"];\n\t\tconst baseActiveToolNames = options.activeToolNames ?? defaultActiveToolNames;\n\t\tthis.refreshToolRegistry({\n\t\t\tactiveToolNames: baseActiveToolNames,\n\t\t\tincludeAllExtensionTools: options.includeAllExtensionTools,\n\t\t});\n\t}\n\n\tasync reload(): Promise<void> {\n\t\tif (this.deps.isStreaming()) {\n\t\t\tthrow new Error(\"Cannot reload while the agent is streaming or a tool call is active\");\n\t\t}\n\t\tif (this.deps.isCompacting()) {\n\t\t\tthrow new Error(\"Cannot reload while context compaction or branch summarization is active\");\n\t\t}\n\t\tconst previousRunner = this.deps.getExtensionRunner();\n\t\tconst snapshot = this._createReloadRuntimeSnapshot();\n\t\t// Preserve the pre-filter tool REQUEST across the rebuild, not the capability/profile-filtered\n\t\t// active set — otherwise a reload under a small model permanently shrinks the restorable set.\n\t\tconst activeToolNames = this.deps.getRequestedActiveToolNames() ?? this.deps.getActiveToolNames();\n\t\tconst previousFlagValues = previousRunner.getFlagValues();\n\t\tconst reloadErrors: string[] = [];\n\t\tlet newRunner: ExtensionRunner | undefined;\n\t\ttry {\n\t\t\tawait this.deps.getSettingsManager().reload();\n\t\t\t// Re-derive the resource-profile tool filter from the freshly reloaded settings.\n\t\t\t// Unlike skills/prompts/themes (which re-filter through the resource loader on every\n\t\t\t// reload), the tool filter is held on the session, so without this a live edit to the\n\t\t\t// active profile's tools allow/block — or switching the active profile — would not\n\t\t\t// apply on /reload and allowed tools would stay missing.\n\t\t\tthis.deps.setToolProfileFilter(this.deps.deriveToolProfileFilter());\n\t\t\t// Re-apply the active profile's model/thinking from the freshly reloaded settings, so a live\n\t\t\t// profile edit (or switch) takes effect on /reload. Skipped when the launch used an explicit\n\t\t\t// --model/--thinking flag, which must win over the profile across reloads.\n\t\t\tawait this.deps.reapplyActiveProfileModelSettings();\n\t\t\tawait this.deps.getResourceLoader().reload({ failOnExtensionErrors: true, deferExtensionDispose: true });\n\t\t\tresetApiProviders();\n\t\t\tthis.buildRuntime({\n\t\t\t\tactiveToolNames,\n\t\t\t\tflagValues: previousFlagValues,\n\t\t\t\tincludeAllExtensionTools: true,\n\t\t\t});\n\t\t\tnewRunner = this.deps.getExtensionRunner();\n\t\t\tconst offDoctorErrors = newRunner.onError((error) => {\n\t\t\t\treloadErrors.push(`${error.extensionPath} ${error.event}: ${error.error}`);\n\t\t\t});\n\t\t\ttry {\n\t\t\t\tthis._doctorReloadRuntime();\n\t\t\t\t// Reload starts memory providers fresh; loaded extensions re-register below.\n\t\t\t\tthis.deps.clearPendingMemoryProviders();\n\t\t\t\tconst hasBindings =\n\t\t\t\t\tthis.deps.getExtensionUIContext() ||\n\t\t\t\t\tthis.deps.getExtensionCommandContextActions() ||\n\t\t\t\t\tthis.deps.getExtensionShutdownHandler() ||\n\t\t\t\t\tthis.deps.getExtensionErrorListener();\n\t\t\t\tif (hasBindings) {\n\t\t\t\t\tawait newRunner.emit({ type: \"session_start\", reason: \"reload\" });\n\t\t\t\t\tawait this.deps.extendResourcesFromExtensions(\"reload\");\n\t\t\t\t\tthis._doctorReloadRuntime();\n\t\t\t\t}\n\t\t\t} finally {\n\t\t\t\toffDoctorErrors();\n\t\t\t}\n\t\t\tif (reloadErrors.length > 0) {\n\t\t\t\tthrow new Error(`Extension reload failed doctor: ${reloadErrors.slice(0, 6).join(\"; \")}`);\n\t\t\t}\n\t\t\tawait emitSessionShutdownEvent(previousRunner, { type: \"session_shutdown\", reason: \"reload\" });\n\t\t\tpreviousRunner.invalidate();\n\t\t\tthis.deps.getResourceLoader().commitReload?.();\n\t\t\t// Re-derive the memory subsystem from the reloaded settings/providers.\n\t\t\tawait this.deps.initializeMemory();\n\t\t} catch (error) {\n\t\t\tif (newRunner && newRunner !== previousRunner) {\n\t\t\t\tnewRunner.invalidate(\n\t\t\t\t\t\"This extension ctx was discarded because reload failed and Pi restored the previous valid runtime.\",\n\t\t\t\t);\n\t\t\t}\n\t\t\tthis.deps.getResourceLoader().rollbackReload?.();\n\t\t\tthis._restoreReloadRuntimeSnapshot(snapshot);\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\t/**\n\t * Unload a single extension without full reload.\n\t * Runs the extension's session_shutdown lifecycle, unregisters its providers,\n\t * disposes its event subscriptions, and rebuilds the runtime.\n\t * Falls back to full reload on error.\n\t */\n\tasync unloadExtensionLive(extensionPath: string): Promise<void> {\n\t\tif (this.deps.isStreaming()) {\n\t\t\tthrow new Error(\"Cannot unload extension while the agent is streaming or a tool call is active\");\n\t\t}\n\t\tif (this.deps.isCompacting()) {\n\t\t\tthrow new Error(\"Cannot unload extension while context compaction or branch summarization is active\");\n\t\t}\n\n\t\tconst ext = this.deps.getResourceLoader().getLoadedExtension(extensionPath);\n\t\tif (!ext) {\n\t\t\treturn; // Nothing to unload\n\t\t}\n\n\t\tconst previousRunner = this.deps.getExtensionRunner();\n\t\ttry {\n\t\t\t// Run session_shutdown lifecycle for this extension only\n\t\t\tawait this.deps.getExtensionRunner().emitToExtension(ext, { type: \"session_shutdown\", reason: \"unload\" });\n\n\t\t\t// Unregister its providers (keyed by the extension's own path, as registered)\n\t\t\tconst runtime = this.deps.getResourceLoader().getExtensions().runtime;\n\t\t\tfor (const name of runtime.getProvidersForExtension(ext.path)) {\n\t\t\t\truntime.unregisterProvider(name, ext.path);\n\t\t\t}\n\n\t\t\t// Dispose its event subscriptions and run disposers\n\t\t\tawait disposeExtensionEventSubscriptions([ext]);\n\n\t\t\t// Remove from loaded extensions\n\t\t\tthis.deps.getResourceLoader().removeLoadedExtension(extensionPath);\n\n\t\t\t// Rebuild runtime with new extension set\n\t\t\tconst activeToolNames = this.deps.getRequestedActiveToolNames() ?? this.deps.getActiveToolNames();\n\t\t\tconst previousFlagValues = previousRunner.getFlagValues();\n\t\t\tthis.buildRuntime({\n\t\t\t\tactiveToolNames,\n\t\t\t\tflagValues: previousFlagValues,\n\t\t\t\tincludeAllExtensionTools: true,\n\t\t\t});\n\t\t\tpreviousRunner.invalidate();\n\n\t\t\t// Notify extensions-changed listeners\n\t\t\tthis.deps.notifyExtensionsChanged();\n\t\t} catch (error) {\n\t\t\t// Fall back to full reload on error\n\t\t\ttry {\n\t\t\t\tawait this.reload();\n\t\t\t} catch {\n\t\t\t\t// Suppress nested error; original error will be thrown below\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\t/**\n\t * Load a single extension without full reload.\n\t * Loads the extension with fresh import, rebuilds the runtime,\n\t * and runs the extension's session_start lifecycle.\n\t * Falls back to full reload on error.\n\t */\n\tasync loadExtensionLive(extensionPath: string): Promise<void> {\n\t\tif (this.deps.isStreaming()) {\n\t\t\tthrow new Error(\"Cannot load extension while the agent is streaming or a tool call is active\");\n\t\t}\n\t\tif (this.deps.isCompacting()) {\n\t\t\tthrow new Error(\"Cannot load extension while context compaction or branch summarization is active\");\n\t\t}\n\n\t\tconst previousRunner = this.deps.getExtensionRunner();\n\t\ttry {\n\t\t\t// Load the extension with fresh import\n\t\t\tconst { extension, error } = await this.deps.getResourceLoader().loadSingleExtension(extensionPath);\n\t\t\tif (error || !extension) {\n\t\t\t\tthrow new Error(error || `Failed to load extension: ${extensionPath}`);\n\t\t\t}\n\n\t\t\t// Rebuild runtime to aggregate tools/commands/handlers/providers\n\t\t\tconst activeToolNames = this.deps.getRequestedActiveToolNames() ?? this.deps.getActiveToolNames();\n\t\t\tconst previousFlagValues = previousRunner.getFlagValues();\n\t\t\tthis.buildRuntime({\n\t\t\t\tactiveToolNames,\n\t\t\t\tflagValues: previousFlagValues,\n\t\t\t\tincludeAllExtensionTools: true,\n\t\t\t});\n\n\t\t\t// Run session_start lifecycle for the new extension only\n\t\t\tawait this.deps.getExtensionRunner().emitToExtension(extension, { type: \"session_start\", reason: \"load\" });\n\n\t\t\t// Notify extensions-changed listeners\n\t\t\tthis.deps.notifyExtensionsChanged();\n\t\t} catch (error) {\n\t\t\t// Fall back to full reload on error\n\t\t\ttry {\n\t\t\t\tawait this.reload();\n\t\t\t} catch {\n\t\t\t\t// Suppress nested error; original error will be thrown below\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n\n\t/**\n\t * Reconcile loaded extensions with the active profile.\n\t * Loads extensions that should be enabled but aren't, and unloads extensions that shouldn't be.\n\t * Falls back to full reload if any individual load/unload fails.\n\t */\n\tasync reconcileLoadedExtensions(): Promise<void> {\n\t\tif (this.deps.isStreaming()) {\n\t\t\tthrow new Error(\"Cannot reconcile extensions while the agent is streaming or a tool call is active\");\n\t\t}\n\t\tif (this.deps.isCompacting()) {\n\t\t\tthrow new Error(\"Cannot reconcile extensions while context compaction or branch summarization is active\");\n\t\t}\n\n\t\ttry {\n\t\t\t// Get all discoverable extension paths\n\t\t\tconst allDiscoverablePaths = await this.deps.getResourceLoader().getDiscoverableExtensionPaths();\n\n\t\t\t// Get the target enabled set based on profile filters\n\t\t\tconst targetEnabledSet = new Set<string>();\n\t\t\tfor (const path of allDiscoverablePaths) {\n\t\t\t\tif (this.deps.getSettingsManager().isResourceAllowedByProfile(\"extensions\", path)) {\n\t\t\t\t\ttargetEnabledSet.add(path);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Get currently loaded set\n\t\t\tconst loadedExtensions = this.deps.getResourceLoader().getExtensions().extensions;\n\t\t\tconst loadedSet = new Set<string>();\n\t\t\tfor (const ext of loadedExtensions) {\n\t\t\t\tloadedSet.add(ext.path);\n\t\t\t}\n\n\t\t\t// Collect unloads and loads\n\t\t\tconst toUnload: string[] = [];\n\t\t\tconst toLoad: string[] = [];\n\n\t\t\tfor (const path of loadedSet) {\n\t\t\t\tif (!targetEnabledSet.has(path)) {\n\t\t\t\t\ttoUnload.push(path);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tfor (const path of targetEnabledSet) {\n\t\t\t\tif (!loadedSet.has(path)) {\n\t\t\t\t\ttoLoad.push(path);\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// Apply unloads first, then loads, to minimize churn\n\t\t\t// Collect errors but continue through all operations\n\t\t\tconst errors: Error[] = [];\n\n\t\t\tfor (const path of toUnload) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.unloadExtensionLive(path);\n\t\t\t\t} catch (error) {\n\t\t\t\t\terrors.push(error instanceof Error ? error : new Error(String(error)));\n\t\t\t\t}\n\t\t\t}\n\n\t\t\tfor (const path of toLoad) {\n\t\t\t\ttry {\n\t\t\t\t\tawait this.loadExtensionLive(path);\n\t\t\t\t} catch (error) {\n\t\t\t\t\terrors.push(error instanceof Error ? error : new Error(String(error)));\n\t\t\t\t}\n\t\t\t}\n\n\t\t\t// If any errors occurred, throw the first one (already fell back to full reload in load/unload)\n\t\t\tif (errors.length > 0) {\n\t\t\t\tthrow errors[0];\n\t\t\t}\n\n\t\t\t// Single notification at the end\n\t\t\tthis.deps.notifyExtensionsChanged();\n\t\t} catch (error) {\n\t\t\t// Fall back to full reload on error\n\t\t\ttry {\n\t\t\t\tawait this.reload();\n\t\t\t} catch {\n\t\t\t\t// Suppress nested error; original error will be thrown below\n\t\t\t}\n\t\t\tthrow error;\n\t\t}\n\t}\n}\n"]}