@nodetool-ai/agents 0.7.0-rc.23 → 0.8.2

package/dist/evals/surfaces/timeline.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"timeline.js","sourceRoot":"","sources":["../../../src/evals/surfaces/timeline.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,wBAAwB,CAAC;AAEzD,OAAO,EACL,iBAAiB,EAEjB,WAAW,EACX,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EACL,qBAAqB,EAEtB,MAAM,6BAA6B,CAAC;AACrC,OAAO,EACL,wBAAwB,EAEzB,MAAM,uCAAuC,CAAC;AAC/C,cAAc,uCAAuC,CAAC;AACtD,MAAM,CAAC,MAAM,wBAAwB,GAAsB;IACzD,uBAAuB;IACvB,oCAAoC;IACpC,yBAAyB;IACzB,0CAA0C;IAC1C,kBAAkB;IAClB,wBAAwB;CACzB,CAAC;AAEF,2EAA2E;AAC3E,MAAM,UAAU,sBAAsB,CAAC,OAA0B;IAC/D,MAAM,QAAQ,GAAG,OAAO,CAAC,MAAM,CAC7B,CAAC,KAAK,EAAE,IAAI,EAAE,CAAC,EAAE,EAAE,CAAC,CAAC,wBAAwB,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,CAAC,EACzE,CAAC,CAAC,CACH,CAAC;IACF,IAAI,QAAQ,KAAK,CAAC,CAAC;QAAE,OAAO,KAAK,CAAC;IAClC,OAAO,OAAO,CAAC,OAAO,CAAC,wBAAwB,EAAE,QAAQ,GAAG,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC;AACxE,CAAC;AAED;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,kBAAkB,CAChC,cAAiC,EACjC,KAA8B,EAC9B,MAAoB;IAEpB,IAAI,cAAc,CAAC,MAAM,KAAK,CAAC;QAAE,OAAO,KAAK,CAAC;IAC9C,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CACzB,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,SAAS,EAAE,EAAE;QACzC,MAAM,MAAM,GAAG,eAAe,CAAC,IAAI,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;QACxD,IAAI,CAAC,CAAC,MAAM,CAAC,KAAK,GAAG,MAAM,CAAC,OAAO,CAAC;YAAE,OAAO,KAAK,CAAC;QACnD,OAAO,cAAc,CAAC,IAAI,CACxB,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,GAAG,MAAM,CAAC,OAAO,IAAI,MAAM,GAAG,MAAM,CAAC,KAAK,CAC7D,CAAC;IACJ,CAAC,CAAC,CACH,CAAC;AACJ,CAAC;AAED;;;GAGG;AACH,SAAS,YAAY,CACnB,IAAkB,EAClB,SAAwB,EACxB,MAAoB;IAEpB,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,SAAS,CAAC,MAAM,CAAC,CAAC;IACxE,MAAM,UAAU,GAAG,SAAS,CAAC,UAAU,IAAI,MAAM,EAAE,iBAAiB,IAAI,CAAC,CAAC;IAC1E,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC;IAClC,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;QAAE,OAAO,UAAU,CAAC;IAC3D,MAAM,KAAK,GAAG,cAAc,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACzD,IAAI,KAAK,GAAG,CAAC;QAAE,OAAO,UAAU,CAAC;IACjC,OAAO,UAAU,GAAG,OAAO,CAAC,QAAQ,GAAG,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;AACrD,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,eAAe,CAC7B,IAAkB,EAClB,SAAwB,EACxB,MAAoB;IAEpB,MAAM,SAAS,GAAG,IAAI,CAAC,OAAO,GAAG,IAAI,CAAC,UAAU,CAAC;IACjD,IAAI,SAAS,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QAC9B,OAAO,EAAE,OAAO,EAAE,IAAI,CAAC,OAAO,EAAE,KAAK,EAAE,SAAS,EAAE,CAAC;IACrD,CAAC;IACD,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,IAAI,CAAC,CAAC;IACvC,MAAM,MAAM,GAAG,YAAY,CAAC,IAAI,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;IACrD,IAAI,SAAS,CAAC,IAAI,KAAK,KAAK,EAAE,CAAC;QAC7B,MAAM,KAAK,GAAG,SAAS,GAAG,OAAO,CAAC;QAClC,OAAO,EAAE,OAAO,EAAE,KAAK,GAAG,MAAM,EAAE,KAAK,EAAE,CAAC;IAC5C,CAAC;IACD,MAAM,OAAO,GAAG,IAAI,CAAC,OAAO,GAAG,OAAO,CAAC;IACvC,OAAO,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,GAAG,MAAM,EAAE,CAAC;AAC9C,CAAC;AAED;;;;;GAKG;AACH,MAAM,UAAU,mBAAmB,CACjC,IAAkB,EAClB,SAAwB,EACxB,MAAoB;IAEpB,MAAM,OAAO,GAAG,SAAS,CAAC,OAAO,CAAC;IAClC,IAAI,CAAC,OAAO,IAAI,CAAC,CAAC,OAAO,CAAC,QAAQ,GAAG,CAAC,CAAC;QAAE,OAAO,IAAI,CAAC;IACrD,MAAM,KAAK,GAAG,cAAc,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,MAAM,CAAC,CAAC;IACzD,IAAI,KAAK,GAAG,CAAC;QAAE,OAAO,IAAI,CAAC;IAC3B,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,SAAS,CAAC,MAAM,CAAC,CAAC;IACxE,MAAM,UAAU,GAAG,SAAS,CAAC,UAAU,IAAI,MAAM,EAAE,iBAAiB,IAAI,CAAC,CAAC;IAC1E,MAAM,IAAI,GACR,CAAC,SAAS,CAAC,OAAO,IAAI,CAAC,CAAC,GAAG,UAAU,GAAG,OAAO,CAAC,QAAQ,GAAG,CAAC,KAAK,GAAG,CAAC,CAAC,CAAC;IACzE,OAAO,IAAI,IAAI,IAAI,CAAC,UAAU,CAAC;AACjC,CAAC;AAED;;;;GAIG;AACH,MAAM,UAAU,cAAc,CAC5B,IAAkB,EAClB,IAAY,EACZ,MAAoB;IAEpB,MAAM,KAAK,GAAG,IAAI,CAAC,SAAS,CAAC;IAC7B,MAAM,MAAM,GAAG,gBAAgB,CAAC,IAAI,CAAC,CAAC;IACtC,yEAAyE;IACzE,gDAAgD;IAChD,IAAI,CAAC,KAAK,IAAI,CAAC,MAAM;QAAE,OAAO,CAAC,CAAC;IAChC,OAAO,qBAAqB,CAAC,KAAK,EAAE,MAAM,EAAE,MAAM,CAAC,CAAC;AACtD,CAAC;AAED;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,SAAwB;IACtD,IAAI,SAAS,CAAC,MAAM;QAAE,OAAO,SAAS,CAAC,MAAM,CAAC;IAC9C,MAAM,MAAM,GAAG,iBAAiB,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,SAAS,CAAC,MAAM,CAAC,CAAC;IACxE,IAAI,MAAM,EAAE,aAAa;QAAE,OAAO,MAAM,CAAC,aAAa,CAAC;IACvD,QAAQ,SAAS,CAAC,IAAI,EAAE,CAAC;QACvB,KAAK,IAAI;YACP,OAAO,SAAS,CAAC;QACnB,KAAK,KAAK;YACR,OAAO,QAAQ,CAAC;QAClB,KAAK,UAAU;YACb,OAAO,WAAW,CAAC;QACrB;YACE,OAAO,QAAQ,CAAC;IACpB,CAAC;AACH,CAAC;AAED,iEAAiE;AACjE,MAAM,iBAAiB,GAAG,IAAI,CAAC;AAE/B;;;;;;;;;;;GAWG;AACH,MAAM,UAAU,iBAAiB,CAAC,MAAc;IAC9C,MAAM,IAAI,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC;IAC3B,IAAI,UAAU,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC;IACvC,MAAM,KAAK,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC;IAChC,IAAI,CAAC,KAAK;QAAE,OAAO,KAAK,CAAC;IACzB,MAAM,KAAK,GAAG,CAAC,KAAK,CAAC,iBAAiB,CAAC,GAAG,KAAK,CAAC,CAAC,CAAC,CAAC,GAAG,iBAAiB,CAAC;IACxE,MAAM,OAAO,GAAG,CAAC,KAAK,CAAC,CAAC,CAAC,GAAG,KAAK,CAAC,CAAC,GAAG,iBAAiB,CAAC,CAAC,GAAG,iBAAiB,CAAC;IAC9E,OAAO,OAAO,GAAG,KAAK,CAAC;AACzB,CAAC;AAED,MAAM,sBAAsB,GAAG;;;;;;;;;;;;;;;;;;;;EAoB7B,qBAAqB,EAAE,EAAE,CAAC;AAE5B;;;;;;GAMG;AACH,SAAS,qBAAqB;IAC5B,MAAM,KAAK,GAAG,eAAe,CAAC,iBAAiB,CAAC,EAAE,OAAO,CAAC;IAC1D,IAAI,CAAC,KAAK;QAAE,OAAO,EAAE,CAAC;IACtB,OAAO;;;;;EAKP,KAAK,EAAE,CAAC;AACV,CAAC;AAED,MAAM,CAAC,MAAM,wBAAwB,GACnC;IACE;QACE,EAAE,EAAE,oBAAoB;QACxB,WAAW,EAAE,0DAA0D;QACvE,SAAS,EACP,6EAA6E;QAC/E,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,EAAE;QAC9C,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE;gBACb,2BAA2B;gBAC3B,0BAA0B;aAC3B;YACD,QAAQ,EAAE,CAAC,CAAC,2BAA2B,EAAE,0BAA0B,CAAC,CAAC;YACrE,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,qBAAqB;oBAC3B,MAAM,EAAE,qCAAqC;oBAC7C,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,KAAK,CAAC,IAAI,CACV,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,SAAS,KAAK,MAAM;wBACtB,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CAC5C;iBACJ;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,sBAAsB;QAC1B,WAAW,EAAE,4DAA4D;QACzE,SAAS,EACP,gHAAgH;QAClH,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,EAAE;QAC9C,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE,CAAC,2BAA2B,EAAE,uBAAuB,CAAC;YACrE,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,6BAA6B;oBACnC,MAAM,EAAE,uDAAuD;oBAC/D,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,KAAK,CAAC,IAAI,CACV,CAAC,CAAC,EAAE,EAAE,CACJ,CAAC,CAAC,SAAS,KAAK,OAAO,IAAI,CAAC,CAAC,OAAO,KAAK,IAAI,IAAI,CAAC,CAAC,CAAC,CAAC,MAAM,CAC9D;iBACJ;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,cAAc;QAClB,WAAW,EAAE,yCAAyC;QACtD,SAAS,EACP,yHAAyH;QAC3H,YAAY,EAAE,GAAG,EAAE,CACjB,wBAAwB,CAAC;YACvB,MAAM,EAAE,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,CAAC;YAC3B,KAAK,EAAE;gBACL;oBACE,IAAI,EAAE,MAAM;oBACZ,UAAU,EAAE,CAAC;oBACb,SAAS,EAAE,OAAO;oBAClB,OAAO,EAAE,CAAC;oBACV,UAAU,EAAE,IAAI;iBACjB;aACF;SACF,CAAC;QACJ,YAAY,EAAE,sBAAsB;QACpC,UAAU,EACR,oIAAoI;QACtI,MAAM,EAAE;YACN,aAAa,EAAE,CAAC,wBAAwB,EAAE,yBAAyB,CAAC;YACpE,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,qBAAqB;oBAC3B,MAAM,EAAE,8CAA8C;oBACtD,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,IAAI,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,CAAC,UAAU,KAAK,IAAI;iBACpE;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,iBAAiB;QACrB,WAAW,EACT,kEAAkE;QACpE,SAAS,EACP,qNAAqN;QACvN,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,EAAE;QAC9C,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE;gBACb,2BAA2B;gBAC3B,0BAA0B;aAC3B;YACD,QAAQ,EAAE,CAAC,CAAC,2BAA2B,EAAE,0BAA0B,CAAC,CAAC;YACrE,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,uBAAuB;oBAC7B,MAAM,EACJ,6EAA6E;oBAC/E,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE;wBAC5B,IAAI,IAAI,CAAC,SAAS,KAAK,MAAM;4BAAE,OAAO,KAAK,CAAC;wBAC5C,MAAM,SAAS,GAAG,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,IAAI,CAC5C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,CAAC,MAAM,KAAK,QAAQ,CAChD,CAAC;wBACF,MAAM,KAAK,GAAG,SAAS,EAAE,MAAM,EAAE,MAAM,CAAC,IAAI,CAC1C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,QAAQ,KAAK,SAAS,CAChC,CAAC;wBACF,IAAI,CAAC,KAAK;4BAAE,OAAO,KAAK,CAAC;wBACzB,MAAM,SAAS,GAAG,KAAK,CAAC,SAAS,CAAC;wBAClC,MAAM,KAAK,GAAG,SAAS,CAAC,CAAC,CAAC,CAAC;wBAC3B,MAAM,IAAI,GAAG,SAAS,CAAC,SAAS,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC;wBAC7C,uDAAuD;wBACvD,4DAA4D;wBAC5D,OAAO,CACL,KAAK,CAAC,CAAC,KAAK,CAAC;4BACb,IAAI,CAAC,CAAC,KAAK,CAAC;4BACZ,KAAK,CAAC,KAAK,IAAI,GAAG;4BAClB,IAAI,CAAC,KAAK,KAAK,CAAC;4BAChB,SAAS,EAAE,UAAU,KAAK,GAAG,CAC9B,CAAC;oBACJ,CAAC,CAAC;iBACL;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,yBAAyB;QAC7B,WAAW,EACT,wEAAwE;QAC1E,SAAS,EACP,qLAAqL;QACvL,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;QAC/D,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE;gBACb,2BAA2B;gBAC3B,0BAA0B;aAC3B;YACD,QAAQ,EAAE,CAAC,CAAC,2BAA2B,EAAE,0BAA0B,CAAC,CAAC;YACrE,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,wBAAwB;oBAC9B,MAAM,EACJ,2FAA2F;oBAC7F,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,aAAa,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE;wBAC5B,IAAI,IAAI,CAAC,SAAS,KAAK,MAAM;4BAAE,OAAO,KAAK,CAAC;wBAC5C,MAAM,QAAQ,GAAG,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,IAAI,CAC3C,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,IAAI,CAAC,CAAC,OAAO,CACpC,CAAC;wBACF,MAAM,OAAO,GAAG,QAAQ,EAAE,OAAO,CAAC;wBAClC,IAAI,CAAC,QAAQ,IAAI,CAAC,OAAO;4BAAE,OAAO,KAAK,CAAC;wBACxC,OAAO,CACL,cAAc,CAAC,IAAI,EAAE,OAAO,CAAC,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC;4BAC1C,mBAAmB,CAAC,IAAI,EAAE,QAAQ,EAAE,CAAC,CAAC,CACvC,CAAC;oBACJ,CAAC,CAAC;iBACL;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,qBAAqB;QACzB,WAAW,EACT,+DAA+D;QACjE,SAAS,EACP,oOAAoO;QACtO,YAAY,EAAE,GAAG,EAAE,CACjB,wBAAwB,CAAC;YACvB,OAAO,EAAE,IAAI;YACb,MAAM,EAAE,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC;YAC5C,KAAK,EAAE;gBACL;oBACE,IAAI,EAAE,MAAM;oBACZ,UAAU,EAAE,CAAC;oBACb,SAAS,EAAE,OAAO;oBAClB,OAAO,EAAE,CAAC;oBACV,UAAU,EAAE,IAAI;iBACjB;aACF;SACF,CAAC;QACJ,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE;gBACb,2BAA2B;gBAC3B,4BAA4B;aAC7B;YACD,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,8BAA8B;oBACpC,MAAM,EACJ,iGAAiG;oBACnG,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE;wBACV,MAAM,OAAO,GAAG,CAAC,OAAe,EAAU,EAAE,CAC1C,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,EAAE,KAAK,OAAO,CAAC,EAAE,KAAK,IAAI,CAAC,CAAC,CAAC;wBACtD,MAAM,MAAM,GAAG,CAAC,CAA0C,EAAE,EAAE,CAC5D,CAAC,CAAC,OAAO,IAAI,CAAC,IAAI,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,UAAU,IAAI,IAAI,CAAC;wBACrD,MAAM,QAAQ,GAAG,CACf,CAA0C,EAC1C,CAA0C,EAC1C,EAAE,CACF,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,UAAU;4BACpC,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,UAAU,CAAC;wBACvC,MAAM,KAAK,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,CAC1B,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,MAAM,IAAI,MAAM,CAAC,CAAC,CAAC,CAC3C,CAAC;wBACF,MAAM,MAAM,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,CAC3B,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,OAAO,IAAI,MAAM,CAAC,CAAC,CAAC,CAC5C,CAAC;wBACF,MAAM,OAAO,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,OAAO,CAAC,CAAC;wBAC/D,gEAAgE;wBAChE,8DAA8D;wBAC9D,yDAAyD;wBACzD,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CACzB,MAAM,CAAC,IAAI,CACT,CAAC,KAAK,EAAE,EAAE,CACR,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,GAAG,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC;4BAC9C,QAAQ,CAAC,KAAK,EAAE,IAAI,CAAC;4BACrB,OAAO,CAAC,IAAI,CACV,CAAC,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,IAAI,CAAC,OAAO,CAAC,GAAG,OAAO,CAAC,KAAK,CAAC,OAAO,CAAC,CACzD,CACJ,CACF,CAAC;oBACJ,CAAC;iBACF;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,sBAAsB;QAC1B,WAAW,EAAE,2DAA2D;QACxE,SAAS,EACP,4IAA4I;QAC9I,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;QAC/D,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE,CAAC,0BAA0B,CAAC;YAC3C,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,0BAA0B;oBAChC,MAAM,EACJ,qFAAqF;oBACvF,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE;wBACV,MAAM,SAAS,GAAG,CAAC,CAAC,aAAa,CAAC,OAAO,CAAC,CAAC,IAAI,EAAE,EAAE,CACjD,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,IAAI,CAAC,CACvD,CAAC;wBACF,OAAO,CACL,SAAS,CAAC,MAAM,IAAI,CAAC;4BACrB,SAAS,CAAC,KAAK,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,iBAAiB,CAAC,eAAe,CAAC,CAAC,CAAC,CAAC,CAAC,CAC9D,CAAC;oBACJ,CAAC;iBACF;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,UAAU;QACd,WAAW,EAAE,wDAAwD;QACrE,SAAS,EACP,mNAAmN;QACrN,YAAY,EAAE,GAAG,EAAE,CACjB,wBAAwB,CAAC;YACvB,OAAO,EAAE,IAAI;YACb,MAAM,EAAE,CAAC,EAAE,IAAI,EAAE,OAAO,EAAE,IAAI,EAAE,SAAS,EAAE,CAAC;YAC5C,KAAK,EAAE;gBACL;oBACE,IAAI,EAAE,GAAG;oBACT,UAAU,EAAE,CAAC;oBACb,SAAS,EAAE,OAAO;oBAClB,OAAO,EAAE,CAAC;oBACV,UAAU,EAAE,IAAI;iBACjB;gBACD;oBACE,IAAI,EAAE,GAAG;oBACT,UAAU,EAAE,CAAC;oBACb,SAAS,EAAE,OAAO;oBAClB,OAAO,EAAE,IAAI;oBACb,UAAU,EAAE,IAAI;iBACjB;gBACD;oBACE,IAAI,EAAE,GAAG;oBACT,UAAU,EAAE,CAAC;oBACb,SAAS,EAAE,OAAO;oBAClB,OAAO,EAAE,IAAI;oBACb,UAAU,EAAE,IAAI;iBACjB;aACF;SACF,CAAC;QACJ,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,wBAAwB;oBAC9B,MAAM,EACJ,4EAA4E;oBAC9E,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE;wBACV,MAAM,MAAM,GAAG,CAAC,CAAC,EAAE,IAAI,EAAE,IAAI,EAAE,IAAI,CAAC,CAAC;wBACrC,MAAM,MAAM,GAAG,CAAC,EAAU,EAAE,EAAE,CAC5B,MAAM,CAAC,IAAI,CAAC,CAAC,KAAK,EAAE,EAAE,CAAC,IAAI,CAAC,GAAG,CAAC,EAAE,GAAG,KAAK,CAAC,IAAI,EAAE,CAAC,CAAC;wBACrD,MAAM,OAAO,GAAG,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,OAAO,CAAC,CAAC;wBAC/D,OAAO,CACL,OAAO,CAAC,MAAM,KAAK,CAAC;4BACpB,OAAO,CAAC,KAAK,CACX,CAAC,CAAC,EAAE,EAAE,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,CAAC,IAAI,MAAM,CAAC,CAAC,CAAC,OAAO,GAAG,CAAC,CAAC,UAAU,CAAC,CAC7D,CACF,CAAC;oBACJ,CAAC;iBACF;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,oBAAoB;QACxB,WAAW,EAAE,4DAA4D;QACzE,SAAS,EACP,sIAAsI;QACxI,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;QAC/D,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE,CAAC,2BAA2B,EAAE,wBAAwB,CAAC;YACtE,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,2BAA2B;oBACjC,MAAM,EACJ,sEAAsE;oBACxE,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,MAAM,CAAC;wBAC3C,sBAAsB,CAAC,CAAC,CAAC,OAAO,CAAC;wBACjC,kBAAkB,CAAC,CAAC,CAAC,cAAc,EAAE,CAAC,CAAC,aAAa,EAAE,CAAC,CAAC;iBAC3D;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,+BAA+B;QACnC,WAAW,EACT,uEAAuE;QACzE,SAAS,EACP,+OAA+O;QACjP,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,EAAE;QAC9C,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE;gBACb,wBAAwB;gBACxB,iCAAiC;aAClC;YACD,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,uBAAuB;oBAC7B,MAAM,EACJ,mEAAmE;oBACrE,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,IAAI,CAAC,CAAC,GAAG,CAAC;wBACjC,CAAC,CAAC,KAAK,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,SAAS,KAAK,OAAO,CAAC,CAAC,MAAM;4BACnD,CAAC,CAAC,CAAC,KAAK,EAAE,KAAK,EAAE,MAAM,IAAI,CAAC,CAAC;iBAClC;gBACD;oBACE,IAAI,EAAE,+BAA+B;oBACrC,MAAM,EACJ,8DAA8D;oBAChE,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE;wBACV,MAAM,MAAM,GAAG,CAAC,CAAC,CAAC,KAAK,EAAE,KAAK,IAAI,EAAE,CAAC,CAAC,MAAM,CAC1C,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,IAAI,CAAC,SAAS,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC,MAAM,GAAG,CAAC,CACnD,CAAC,MAAM,CAAC;wBACT,MAAM,KAAK,GAAG,CAAC,CAAC,aAAa,CAAC,MAAM,CAClC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,WAAW,KAAK,eAAe,IAAI,CAAC,CAAC,MAAM,CACrD,CAAC,MAAM,CAAC;wBACT,OAAO,MAAM,GAAG,CAAC,IAAI,KAAK,KAAK,MAAM,CAAC;oBACxC,CAAC;iBACF;gBACD;oBACE,IAAI,EAAE,cAAc;oBACpB,MAAM,EAAE,+CAA+C;oBACvD,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,KAAK,EAAE,KAAK,KAAK,MAAM;iBACvC;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,mBAAmB;QACvB,WAAW,EAAE,wDAAwD;QACrE,SAAS,EACP,iOAAiO;QACnO,YAAY,EAAE,GAAG,EAAE,CAAC,wBAAwB,CAAC,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC;QAC/D,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE;gBACb,8BAA8B;gBAC9B,0BAA0B;gBAC1B,wBAAwB;aACzB;YACD,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,EAAE;YAChB,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,yBAAyB;oBAC/B,MAAM,EACJ,4EAA4E;oBAC9E,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CACV,CAAC,CAAC,aAAa,CAAC,IAAI,CAClB,CAAC,IAAI,EAAE,EAAE,CACP,IAAI,CAAC,SAAS,KAAK,SAAS;wBAC5B,IAAI,CAAC,cAAc,KAAK,mBAAmB;wBAC3C,IAAI,CAAC,YAAY,KAAK,SAAS;wBAC/B,CAAC,IAAI,CAAC,UAAU,IAAI,EAAE,CAAC,CAAC,IAAI,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,IAAI,KAAK,MAAM,CAAC,CACzD;iBACJ;gBACD;oBACE,kEAAkE;oBAClE,mEAAmE;oBACnE,IAAI,EAAE,yBAAyB;oBAC/B,MAAM,EAAE,qCAAqC;oBAC7C,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC,CAAC,mBAAmB,CAAC,QAAQ,CAAC,SAAS,CAAC;iBACvD;aACF;SACF;KACF;IACD;QACE,EAAE,EAAE,oCAAoC;QACxC,WAAW,EACT,6EAA6E;QAC/E,SAAS,EACP,yNAAyN;QAC3N,YAAY,EAAE,GAAG,EAAE,CACjB,wBAAwB,CAAC;YACvB,UAAU,EAAE,uBAAuB;YACnC,QAAQ,EAAE;gBACR,MAAM,EAAE;oBACN;wBACE,EAAE,EAAE,aAAa;wBACjB,IAAI,EAAE,OAAO;wBACb,IAAI,EAAE,OAAO;wBACb,KAAK,EAAE,CAAC;wBACR,OAAO,EAAE,IAAI;wBACb,MAAM,EAAE,KAAK;qBACd;iBACF;gBACD,KAAK,EAAE;oBACL;wBACE,EAAE,EAAE,cAAc;wBAClB,OAAO,EAAE,aAAa;wBACtB,IAAI,EAAE,SAAS;wBACf,OAAO,EAAE,IAAI;wBACb,UAAU,EAAE,IAAI;wBAChB,SAAS,EAAE,KAAK;wBAChB,UAAU,EAAE,KAAK;wBACjB,SAAS,EAAE,OAAO;wBAClB,UAAU,EAAE,UAAU;wBACtB,MAAM,EAAE,WAAW;wBACnB,MAAM,EAAE,KAAK;wBACb,cAAc,EAAE,gBAAgB;wBAChC,YAAY,EAAE,eAAe;wBAC7B,QAAQ,EAAE;4BACR;gCACE,EAAE,EAAE,eAAe;gCACnB,SAAS,EAAE,0BAA0B;gCACrC,iBAAiB,EAAE,0BAA0B;gCAC7C,KAAK,EAAE,cAAc;gCACrB,OAAO,EAAE,gBAAgB;gCACzB,cAAc,EAAE,EAAE;gCAClB,sBAAsB,EAAE,EAAE;gCAC1B,UAAU,EAAE,IAAI;gCAChB,MAAM,EAAE,SAAS;gCACjB,MAAM,EAAE,UAAU;gCAClB,aAAa,EAAE;oCACb,SAAS,EAAE,KAAK;oCAChB,UAAU,EAAE,KAAK;oCACjB,eAAe,EAAE,CAAC;oCAClB,UAAU,EAAE,KAAK;iCAClB;6BACF;yBACF;qBACF;iBACF;aACF;SACF,CAAC;QACJ,YAAY,EAAE,sBAAsB;QACpC,MAAM,EAAE;YACN,aAAa,EAAE;gBACb,oCAAoC;gBACpC,wBAAwB;gBACxB,wBAAwB;aACzB;YACD,QAAQ,EAAE;gBACR,CAAC,oCAAoC,EAAE,wBAAwB,CAAC;gBAChE,CAAC,wBAAwB,EAAE,wBAAwB,CAAC;aACrD;YACD,cAAc,EAAE,IAAI;YACpB,YAAY,EAAE,CAAC;YACf,YAAY,EAAE,CAAC;YACf,UAAU,EAAE;gBACV;oBACE,IAAI,EAAE,oCAAoC;oBAC1C,MAAM,EACJ,2FAA2F;oBAC7F,IAAI,EAAE,CAAC,CAAC,EAAE,EAAE;wBACV,MAAM,IAAI,GAAG,CAAC,CAAC,aAAa,CAAC,IAAI,CAC/B,CAAC,SAAS,EAAE,EAAE,CAAC,SAAS,CAAC,EAAE,KAAK,cAAc,CAC/C,CAAC;wBACF,MAAM,MAAM,GAAG,IAAI,EAAE,QAAQ,EAAE,IAAI,CACjC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,EAAE,KAAK,IAAI,CAAC,YAAY,CACxC,CAAC;wBACF,OAAO,CACL,IAAI,EAAE,OAAO,KAAK,IAAI;4BACtB,IAAI,CAAC,UAAU,KAAK,IAAI;4BACxB,IAAI,CAAC,cAAc,EAAE,UAAU,CAC7B,4BAA4B,CAC7B,KAAK,IAAI;4BACV,IAAI,CAAC,SAAS,KAAK,CAAC;4BACpB,IAAI,CAAC,UAAU,KAAK,IAAI;4BACxB,MAAM,EAAE,SAAS,EAAE,aAAa,CAAC,aAAa,KAAK,KAAK;4BACxD,MAAM,CAAC,SAAS,CAAC,aAAa,CAAC,WAAW,KAAK,KAAK,CACrD,CAAC;oBACJ,CAAC;iBACF;aACF;SACF;KACF;CACF,CAAC"}

package/dist/evals/task-planner-cases.d.ts

@@ -0,0 +1,57 @@
+/**
+ * Built-in evaluation cases for TaskPlanner multi-task (plan mode) planning.
+ *
+ * Each case is an objective plus machine-checkable expectations on the
+ * committed plan. `PlanBuilder` already guarantees the plan is structurally
+ * sound, so the expectations here target judgment calls it cannot make:
+ * how wide the plan opens, whether a genuine dependency is modelled as one,
+ * whether decomposition stays proportional to the work, and which tool a step
+ * routes to.
+ */
+export interface TaskPlannerEvalExpectations {
+    minTasks?: number;
+    maxTasks?: number;
+    minSteps?: number;
+    maxSteps?: number;
+    /** Cap on the widest task — the prompt's step-granularity rule. */
+    maxStepsPerTask?: number;
+    /** Tasks with an empty `dependsOn`; the plan's parallel width. */
+    minIndependentTasks?: number;
+    /** Tasks with a non-empty `dependsOn`; proves the DAG is not flat. */
+    minDependentTasks?: number;
+    /** No task may depend on another — the work is genuinely independent. */
+    requireFlat?: boolean;
+    /** Tool names some step must route to (via `tools` or its instructions). */
+    requiredTools?: string[];
+    /** Tool names no step may route to. */
+    forbiddenTools?: string[];
+    /** Regex sources; each must match the concatenated step instructions. */
+    requiredInstructionPatterns?: string[];
+    /** Opt out of the universal "no synthesis task" check. */
+    allowSynthesisTask?: boolean;
+    /**
+     * No step may instruct work the toolbelt cannot do. Checked against the
+     * instruction text rather than the `tools` array, because a planner with no
+     * tools writes "search the web for …" in prose instead of routing.
+     */
+    forbidToolWork?: boolean;
+}
+export interface TaskPlannerEvalCase {
+    id: string;
+    description: string;
+    objective: string;
+    inputs?: Record<string, unknown>;
+    outputSchema?: Record<string, unknown>;
+    /** Case needs configured model providers — skipped when there are none. */
+    needsModelProviders?: boolean;
+    /**
+     * The toolbelt this case plans against. `"all"` offers the declarative
+     * library; `"none"` offers nothing, which is the shape an `AgentNode` with
+     * no tools selected actually runs — the majority of real runs, and the one
+     * the suite could not see before.
+     */
+    toolbelt?: "all" | "none";
+    expect: TaskPlannerEvalExpectations;
+}
+export declare const TASK_PLANNER_EVAL_CASES: readonly TaskPlannerEvalCase[];
+//# sourceMappingURL=task-planner-cases.d.ts.map

package/dist/evals/task-planner-cases.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-planner-cases.d.ts","sourceRoot":"","sources":["../../src/evals/task-planner-cases.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAEH,MAAM,WAAW,2BAA2B;IAC1C,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,mEAAmE;IACnE,eAAe,CAAC,EAAE,MAAM,CAAC;IACzB,kEAAkE;IAClE,mBAAmB,CAAC,EAAE,MAAM,CAAC;IAC7B,sEAAsE;IACtE,iBAAiB,CAAC,EAAE,MAAM,CAAC;IAC3B,yEAAyE;IACzE,WAAW,CAAC,EAAE,OAAO,CAAC;IACtB,4EAA4E;IAC5E,aAAa,CAAC,EAAE,MAAM,EAAE,CAAC;IACzB,uCAAuC;IACvC,cAAc,CAAC,EAAE,MAAM,EAAE,CAAC;IAC1B,yEAAyE;IACzE,2BAA2B,CAAC,EAAE,MAAM,EAAE,CAAC;IACvC,0DAA0D;IAC1D,kBAAkB,CAAC,EAAE,OAAO,CAAC;IAC7B;;;;OAIG;IACH,cAAc,CAAC,EAAE,OAAO,CAAC;CAC1B;AAED,MAAM,WAAW,mBAAmB;IAClC,EAAE,EAAE,MAAM,CAAC;IACX,WAAW,EAAE,MAAM,CAAC;IACpB,SAAS,EAAE,MAAM,CAAC;IAClB,MAAM,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACjC,YAAY,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;IACvC,2EAA2E;IAC3E,mBAAmB,CAAC,EAAE,OAAO,CAAC;IAC9B;;;;;OAKG;IACH,QAAQ,CAAC,EAAE,KAAK,GAAG,MAAM,CAAC;IAC1B,MAAM,EAAE,2BAA2B,CAAC;CACrC;AAID,eAAO,MAAM,uBAAuB,EAAE,SAAS,mBAAmB,EAmHjE,CAAC"}

package/dist/evals/task-planner-cases.js

@@ -0,0 +1,117 @@
+/**
+ * Built-in evaluation cases for TaskPlanner multi-task (plan mode) planning.
+ *
+ * Each case is an objective plus machine-checkable expectations on the
+ * committed plan. `PlanBuilder` already guarantees the plan is structurally
+ * sound, so the expectations here target judgment calls it cannot make:
+ * how wide the plan opens, whether a genuine dependency is modelled as one,
+ * whether decomposition stays proportional to the work, and which tool a step
+ * routes to.
+ */
+import { PLANNER_TOOL_NAMES } from "./planner-tools.js";
+export const TASK_PLANNER_EVAL_CASES = [
+    {
+        id: "parallel-research",
+        description: "Three unrelated topics must become three concurrent tasks",
+        objective: "Research three separate topics — retrieval-augmented generation, vector databases, and prompt caching. For each one, find recent sources and write down the key findings. The three topics are unrelated to each other.",
+        outputSchema: {
+            type: "object",
+            properties: { findings: { type: "string" } },
+            required: ["findings"]
+        },
+        expect: {
+            minTasks: 3,
+            minIndependentTasks: 3,
+            requireFlat: true,
+            requiredTools: ["web_search"],
+            maxStepsPerTask: 3
+        }
+    },
+    {
+        id: "real-dependency",
+        description: "A genuine data dependency must be modelled as one, not flattened",
+        objective: "Download the page at the input url, then extract every section heading from the downloaded text, and separately count how many words the page contains. The extraction and the word count both need the downloaded page.",
+        inputs: { url: "https://example.com/article" },
+        expect: {
+            minTasks: 3,
+            minDependentTasks: 2,
+            requiredTools: ["fetch_page"],
+            requiredInstructionPatterns: ["heading"]
+        }
+    },
+    {
+        id: "no-over-decomposition",
+        description: "A one-shot objective must not be split into a task graph",
+        objective: "Write a single four-line poem about autumn rain. No research, no files, no images — just the poem.",
+        expect: {
+            maxTasks: 2,
+            maxSteps: 2,
+            forbiddenTools: ["web_search", "fetch_page", "generate_image"]
+        }
+    },
+    {
+        id: "deterministic-routing",
+        description: "Arithmetic must be routed to run_python, not to a reasoning step",
+        objective: "Compute the first 40 Fibonacci numbers and their sum exactly. This is deterministic computation — it must be executed as code, not reasoned out. Write the result to results.txt.",
+        expect: {
+            maxTasks: 3,
+            requiredTools: ["run_python", "write_file"],
+            forbiddenTools: ["web_search", "generate_image"]
+        }
+    },
+    {
+        id: "media-fanout",
+        description: "One image per scene, planned as concurrent generation tasks",
+        objective: "Generate one illustration for each of these three scenes: a lighthouse at dawn, a market at noon, a harbour at night. Each illustration is independent of the others.",
+        needsModelProviders: true,
+        expect: {
+            minTasks: 3,
+            minIndependentTasks: 3,
+            requiredTools: ["generate_image"],
+            requiredInstructionPatterns: ["lighthouse", "harbour"]
+        }
+    },
+    {
+        id: "no-tools-research",
+        description: "With an empty toolbelt the plan must not instruct steps to search or fetch",
+        objective: "Plan a research project on node-based AI workflow tools: what exists, how they differ architecturally, and where they fall short.",
+        toolbelt: "none",
+        expect: {
+            minTasks: 3,
+            minIndependentTasks: 3,
+            forbidToolWork: true,
+            forbiddenTools: [...PLANNER_TOOL_NAMES]
+        }
+    },
+    {
+        id: "gather-not-assemble",
+        description: "Under an output schema the plan must gather facts and leave assembly to the calling loop",
+        objective: "Produce a competitive brief on three note-taking apps: Obsidian, Notion, and Logseq. Gather pricing, platform support, and the standout feature for each.",
+        outputSchema: {
+            type: "object",
+            properties: {
+                apps: {
+                    type: "array",
+                    items: {
+                        type: "object",
+                        properties: {
+                            name: { type: "string" },
+                            pricing: { type: "string" },
+                            platforms: { type: "string" },
+                            standoutFeature: { type: "string" }
+                        },
+                        required: ["name", "pricing", "platforms", "standoutFeature"]
+                    }
+                }
+            },
+            required: ["apps"]
+        },
+        expect: {
+            minTasks: 3,
+            minIndependentTasks: 3,
+            requiredTools: ["web_search"],
+            requiredInstructionPatterns: ["pricing"]
+        }
+    }
+];
+//# sourceMappingURL=task-planner-cases.js.map

package/dist/evals/task-planner-cases.js.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-planner-cases.js","sourceRoot":"","sources":["../../src/evals/task-planner-cases.ts"],"names":[],"mappings":"AAAA;;;;;;;;;GASG;AAiDH,OAAO,EAAE,kBAAkB,EAAE,MAAM,oBAAoB,CAAC;AAExD,MAAM,CAAC,MAAM,uBAAuB,GAAmC;IACrE;QACE,EAAE,EAAE,mBAAmB;QACvB,WAAW,EAAE,2DAA2D;QACxE,SAAS,EACP,yNAAyN;QAC3N,YAAY,EAAE;YACZ,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE,EAAE,QAAQ,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE,EAAE;YAC5C,QAAQ,EAAE,CAAC,UAAU,CAAC;SACvB;QACD,MAAM,EAAE;YACN,QAAQ,EAAE,CAAC;YACX,mBAAmB,EAAE,CAAC;YACtB,WAAW,EAAE,IAAI;YACjB,aAAa,EAAE,CAAC,YAAY,CAAC;YAC7B,eAAe,EAAE,CAAC;SACnB;KACF;IACD;QACE,EAAE,EAAE,iBAAiB;QACrB,WAAW,EACT,kEAAkE;QACpE,SAAS,EACP,0NAA0N;QAC5N,MAAM,EAAE,EAAE,GAAG,EAAE,6BAA6B,EAAE;QAC9C,MAAM,EAAE;YACN,QAAQ,EAAE,CAAC;YACX,iBAAiB,EAAE,CAAC;YACpB,aAAa,EAAE,CAAC,YAAY,CAAC;YAC7B,2BAA2B,EAAE,CAAC,SAAS,CAAC;SACzC;KACF;IACD;QACE,EAAE,EAAE,uBAAuB;QAC3B,WAAW,EAAE,0DAA0D;QACvE,SAAS,EACP,oGAAoG;QACtG,MAAM,EAAE;YACN,QAAQ,EAAE,CAAC;YACX,QAAQ,EAAE,CAAC;YACX,cAAc,EAAE,CAAC,YAAY,EAAE,YAAY,EAAE,gBAAgB,CAAC;SAC/D;KACF;IACD;QACE,EAAE,EAAE,uBAAuB;QAC3B,WAAW,EACT,kEAAkE;QACpE,SAAS,EACP,mLAAmL;QACrL,MAAM,EAAE;YACN,QAAQ,EAAE,CAAC;YACX,aAAa,EAAE,CAAC,YAAY,EAAE,YAAY,CAAC;YAC3C,cAAc,EAAE,CAAC,YAAY,EAAE,gBAAgB,CAAC;SACjD;KACF;IACD;QACE,EAAE,EAAE,cAAc;QAClB,WAAW,EAAE,6DAA6D;QAC1E,SAAS,EACP,uKAAuK;QACzK,mBAAmB,EAAE,IAAI;QACzB,MAAM,EAAE;YACN,QAAQ,EAAE,CAAC;YACX,mBAAmB,EAAE,CAAC;YACtB,aAAa,EAAE,CAAC,gBAAgB,CAAC;YACjC,2BAA2B,EAAE,CAAC,YAAY,EAAE,SAAS,CAAC;SACvD;KACF;IACD;QACE,EAAE,EAAE,mBAAmB;QACvB,WAAW,EACT,4EAA4E;QAC9E,SAAS,EACP,mIAAmI;QACrI,QAAQ,EAAE,MAAM;QAChB,MAAM,EAAE;YACN,QAAQ,EAAE,CAAC;YACX,mBAAmB,EAAE,CAAC;YACtB,cAAc,EAAE,IAAI;YACpB,cAAc,EAAE,CAAC,GAAG,kBAAkB,CAAC;SACxC;KACF;IACD;QACE,EAAE,EAAE,qBAAqB;QACzB,WAAW,EACT,0FAA0F;QAC5F,SAAS,EACP,2JAA2J;QAC7J,YAAY,EAAE;YACZ,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,IAAI,EAAE;oBACJ,IAAI,EAAE,OAAO;oBACb,KAAK,EAAE;wBACL,IAAI,EAAE,QAAQ;wBACd,UAAU,EAAE;4BACV,IAAI,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;4BACxB,OAAO,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;4BAC3B,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;4BAC7B,eAAe,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;yBACpC;wBACD,QAAQ,EAAE,CAAC,MAAM,EAAE,SAAS,EAAE,WAAW,EAAE,iBAAiB,CAAC;qBAC9D;iBACF;aACF;YACD,QAAQ,EAAE,CAAC,MAAM,CAAC;SACnB;QACD,MAAM,EAAE;YACN,QAAQ,EAAE,CAAC;YACX,mBAAmB,EAAE,CAAC;YACtB,aAAa,EAAE,CAAC,YAAY,CAAC;YAC7B,2BAA2B,EAAE,CAAC,SAAS,CAAC;SACzC;KACF;CACF,CAAC"}

package/dist/evals/task-planner-eval.d.ts

@@ -0,0 +1,101 @@
+/**
+ * TaskPlanner (plan mode) evaluation harness — provider-agnostic.
+ *
+ * Runs {@link TaskPlanner.planMultiTask} for each case and scores the resulting
+ * {@link TaskPlan}. `PlanBuilder` already rejects structurally broken plans —
+ * duplicate step ids, dangling dependencies, cycles — so a plan that comes back
+ * at all is valid by construction. What it cannot judge is plan *quality*, and
+ * that is what this suite measures: does the planner exploit parallelism, keep
+ * decomposition proportional to the objective, route work to the right tool,
+ * respect the step-granularity and id-prefix conventions, and leave final
+ * synthesis to the calling loop instead of planning an "assemble" task?
+ *
+ * Scoring is structural (see {@link checkTaskPlanExpectations}) — task/step
+ * counts, DAG shape, tool routing, instruction patterns — never an exact plan,
+ * so many valid decompositions pass.
+ */
+import type { BaseProvider } from "@nodetool-ai/runtime";
+import type { EvalCheck } from "./graph-planner-eval.js";
+import type { Task, TaskPlan } from "../types.js";
+import { type TaskPlannerEvalCase, type TaskPlannerEvalExpectations } from "./task-planner-cases.js";
+export interface TaskPlanCaseResult {
+    caseId: string;
+    description: string;
+    skipped: boolean;
+    /** The planner returned a committed plan. */
+    accepted: boolean;
+    score: number;
+    checks: EvalCheck[];
+    tasks: number;
+    steps: number;
+    /** Tasks with no `dependsOn` — how wide the plan opens. */
+    parallelWidth: number;
+    /** Longest chain of task dependencies (1 for a flat fan-out). */
+    criticalPath: number;
+    /** Planner tool calls by name (`add_task`, `remove_task`, `finish_plan`). */
+    toolCalls: Record<string, number>;
+    /** `add_task`/`finish_plan` calls the builder rejected. */
+    validationFailures: number;
+    durationMs: number;
+    costUsd: number;
+    error?: string;
+}
+export interface TaskPlanEvalReport {
+    provider: string;
+    model: string;
+    startedAt: string;
+    cases: TaskPlanCaseResult[];
+    summary: {
+        total: number;
+        skipped: number;
+        accepted: number;
+        /** accepted / (total - skipped) */
+        successRate: number;
+        /** Mean expectation score over non-skipped cases. */
+        meanScore: number;
+        /** Fraction of accepted plans built without a single rejected tool call. */
+        cleanRate: number;
+        avgTasks: number;
+        avgSteps: number;
+        avgParallelWidth: number;
+        avgDurationMs: number;
+        totalCostUsd: number;
+    };
+}
+/**
+ * The `AgentNode`'s default system prompt, mirrored here so the suite plans
+ * behind the same preamble a real run does. Kept as a literal rather than
+ * imported: `packages/agents` sits below `llm-nodes` in the dependency order.
+ */
+export declare const AGENT_NODE_SYSTEM_PROMPT = "You are a friendly assistant";
+export interface RunTaskPlannerEvalOptions {
+    provider: BaseProvider;
+    model: string;
+    /** Configured providers; enables model-dependent cases (else they skip). */
+    providers?: Record<string, BaseProvider>;
+    /** Cases to run; defaults to the built-in suite. */
+    cases?: readonly TaskPlannerEvalCase[];
+    /**
+     * The caller preamble every case plans behind. Defaults to the `AgentNode`'s
+     * own default system prompt, because that is what the product sends; pass
+     * `""` to score the bare planner contract instead.
+     */
+    systemPrompt?: string;
+    signal?: AbortSignal;
+    /** Progress callback (one line per event, for CLI display). */
+    onEvent?: (line: string) => void;
+}
+/**
+ * Longest chain of task-level dependencies. 1 means a flat fan-out (every task
+ * independent); N means the plan serializes N stages.
+ */
+export declare function criticalPathDepth(tasks: readonly Task[]): number;
+/**
+ * Score a committed plan against a case's expectations. Pure and fully
+ * unit-testable: takes a {@link TaskPlan}, never calls a provider.
+ */
+export declare function checkTaskPlanExpectations(plan: TaskPlan, expect: TaskPlannerEvalExpectations): EvalCheck[];
+export declare function runTaskPlannerEval(opts: RunTaskPlannerEvalOptions): Promise<TaskPlanEvalReport>;
+/** Text summary table for terminal output. */
+export declare function formatTaskPlanReport(report: TaskPlanEvalReport): string;
+//# sourceMappingURL=task-planner-eval.d.ts.map

package/dist/evals/task-planner-eval.d.ts.map

@@ -0,0 +1 @@
+{"version":3,"file":"task-planner-eval.d.ts","sourceRoot":"","sources":["../../src/evals/task-planner-eval.ts"],"names":[],"mappings":"AAAA;;;;;;;;;;;;;;;GAeG;AAGH,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,sBAAsB,CAAC;AAEzD,OAAO,KAAK,EAAE,SAAS,EAAE,MAAM,yBAAyB,CAAC;AAEzD,OAAO,KAAK,EAAE,IAAI,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAElD,OAAO,EAEL,KAAK,mBAAmB,EACxB,KAAK,2BAA2B,EACjC,MAAM,yBAAyB,CAAC;AAiBjC,MAAM,WAAW,kBAAkB;IACjC,MAAM,EAAE,MAAM,CAAC;IACf,WAAW,EAAE,MAAM,CAAC;IACpB,OAAO,EAAE,OAAO,CAAC;IACjB,6CAA6C;IAC7C,QAAQ,EAAE,OAAO,CAAC;IAClB,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,SAAS,EAAE,CAAC;IACpB,KAAK,EAAE,MAAM,CAAC;IACd,KAAK,EAAE,MAAM,CAAC;IACd,2DAA2D;IAC3D,aAAa,EAAE,MAAM,CAAC;IACtB,iEAAiE;IACjE,YAAY,EAAE,MAAM,CAAC;IACrB,6EAA6E;IAC7E,SAAS,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;IAClC,2DAA2D;IAC3D,kBAAkB,EAAE,MAAM,CAAC;IAC3B,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;IAChB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AAED,MAAM,WAAW,kBAAkB;IACjC,QAAQ,EAAE,MAAM,CAAC;IACjB,KAAK,EAAE,MAAM,CAAC;IACd,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,EAAE,kBAAkB,EAAE,CAAC;IAC5B,OAAO,EAAE;QACP,KAAK,EAAE,MAAM,CAAC;QACd,OAAO,EAAE,MAAM,CAAC;QAChB,QAAQ,EAAE,MAAM,CAAC;QACjB,mCAAmC;QACnC,WAAW,EAAE,MAAM,CAAC;QACpB,qDAAqD;QACrD,SAAS,EAAE,MAAM,CAAC;QAClB,4EAA4E;QAC5E,SAAS,EAAE,MAAM,CAAC;QAClB,QAAQ,EAAE,MAAM,CAAC;QACjB,QAAQ,EAAE,MAAM,CAAC;QACjB,gBAAgB,EAAE,MAAM,CAAC;QACzB,aAAa,EAAE,MAAM,CAAC;QACtB,YAAY,EAAE,MAAM,CAAC;KACtB,CAAC;CACH;AAED;;;;GAIG;AACH,eAAO,MAAM,wBAAwB,iCAAiC,CAAC;AAEvE,MAAM,WAAW,yBAAyB;IACxC,QAAQ,EAAE,YAAY,CAAC;IACvB,KAAK,EAAE,MAAM,CAAC;IACd,4EAA4E;IAC5E,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,YAAY,CAAC,CAAC;IACzC,oDAAoD;IACpD,KAAK,CAAC,EAAE,SAAS,mBAAmB,EAAE,CAAC;IACvC;;;;OAIG;IACH,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,MAAM,CAAC,EAAE,WAAW,CAAC;IACrB,+DAA+D;IAC/D,OAAO,CAAC,EAAE,CAAC,IAAI,EAAE,MAAM,KAAK,IAAI,CAAC;CAClC;AAgBD;;;GAGG;AACH,wBAAgB,iBAAiB,CAAC,KAAK,EAAE,SAAS,IAAI,EAAE,GAAG,MAAM,CA0BhE;AAED;;;GAGG;AACH,wBAAgB,yBAAyB,CACvC,IAAI,EAAE,QAAQ,EACd,MAAM,EAAE,2BAA2B,GAClC,SAAS,EAAE,CAwIb;AAkFD,wBAAsB,kBAAkB,CACtC,IAAI,EAAE,yBAAyB,GAC9B,OAAO,CAAC,kBAAkB,CAAC,CAwE7B;AAED,8CAA8C;AAC9C,wBAAgB,oBAAoB,CAAC,MAAM,EAAE,kBAAkB,GAAG,MAAM,CAmDvE"}