@dalmia/calibrate-mcp 0.0.52 → 0.0.54
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/bin/mcp-server.js +397 -216
- package/bin/mcp-server.js.map +19 -16
- package/esm/funcs/agentTestsGetBenchmark.js +1 -0
- package/esm/funcs/agentTestsGetBenchmark.js.map +1 -1
- package/esm/funcs/agentTestsGetRun.js +1 -0
- package/esm/funcs/agentTestsGetRun.js.map +1 -1
- package/esm/funcs/agentTestsGetRunCase.d.ts +16 -0
- package/esm/funcs/agentTestsGetRunCase.d.ts.map +1 -0
- package/esm/funcs/agentTestsGetRunCase.js +100 -0
- package/esm/funcs/agentTestsGetRunCase.js.map +1 -0
- package/esm/landing-page.js +1 -1
- package/esm/lib/config.d.ts +2 -2
- package/esm/lib/config.js +2 -2
- package/esm/mcp-server/mcp-server.js +1 -1
- package/esm/mcp-server/server.d.ts.map +1 -1
- package/esm/mcp-server/server.js +3 -1
- package/esm/mcp-server/server.js.map +1 -1
- package/esm/mcp-server/tools/agentTestsGetRunCase.d.ts +7 -0
- package/esm/mcp-server/tools/agentTestsGetRunCase.d.ts.map +1 -0
- package/esm/mcp-server/tools/agentTestsGetRunCase.js +38 -0
- package/esm/mcp-server/tools/agentTestsGetRunCase.js.map +1 -0
- package/esm/models/agenttestrunlistitem.js +1 -1
- package/esm/models/agenttestrunlistitem.js.map +1 -1
- package/esm/models/benchmarkstatusresponse.d.ts +1 -0
- package/esm/models/benchmarkstatusresponse.d.ts.map +1 -1
- package/esm/models/benchmarkstatusresponse.js +1 -0
- package/esm/models/benchmarkstatusresponse.js.map +1 -1
- package/esm/models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.d.ts +13 -0
- package/esm/models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.d.ts.map +1 -0
- package/esm/models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js +17 -0
- package/esm/models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js.map +1 -0
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.d.ts +17 -0
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.d.ts.map +1 -1
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.js +12 -0
- package/esm/models/getagenttestrunstatusagenttestsruntaskidgetop.js.map +1 -1
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts +17 -0
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts.map +1 -1
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.js +13 -0
- package/esm/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.js.map +1 -1
- package/esm/models/testcaseresult.d.ts +16 -0
- package/esm/models/testcaseresult.d.ts.map +1 -1
- package/esm/models/testcaseresult.js +14 -0
- package/esm/models/testcaseresult.js.map +1 -1
- package/esm/models/testrunstatusresponse.d.ts +4 -0
- package/esm/models/testrunstatusresponse.d.ts.map +1 -1
- package/esm/models/testrunstatusresponse.js +3 -0
- package/esm/models/testrunstatusresponse.js.map +1 -1
- package/esm/tool-names.d.ts.map +1 -1
- package/esm/tool-names.js +4 -0
- package/esm/tool-names.js.map +1 -1
- package/package.json +1 -1
- package/src/funcs/agentTestsGetBenchmark.ts +1 -0
- package/src/funcs/agentTestsGetRun.ts +1 -0
- package/src/funcs/agentTestsGetRunCase.ts +167 -0
- package/src/landing-page.ts +1 -1
- package/src/lib/config.ts +2 -2
- package/src/mcp-server/mcp-server.ts +1 -1
- package/src/mcp-server/server.ts +3 -1
- package/src/mcp-server/tools/agentTestsGetRunCase.ts +47 -0
- package/src/models/agenttestrunlistitem.ts +1 -1
- package/src/models/benchmarkstatusresponse.ts +4 -0
- package/src/models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.ts +44 -0
- package/src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts +30 -0
- package/src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts +29 -0
- package/src/models/testcaseresult.ts +24 -0
- package/src/models/testrunstatusresponse.ts +9 -0
- package/src/tool-names.ts +4 -0
|
@@ -33,6 +33,7 @@ async function $do(client$, request, options) {
|
|
|
33
33
|
const path$ = pathToFunc("/agent-tests/benchmark/{task_id}")(pathParams$);
|
|
34
34
|
const query$ = encodeFormQuery({
|
|
35
35
|
"compact": payload$.compact,
|
|
36
|
+
"mode": payload$.mode,
|
|
36
37
|
"only_failed": payload$.only_failed,
|
|
37
38
|
});
|
|
38
39
|
const headers$ = new Headers(compactMap({
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agentTestsGetBenchmark.js","sourceRoot":"","sources":["../../src/funcs/agentTestsGetBenchmark.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,+DAA+D,GAChE,MAAM,+DAA+D,CAAC;AACvE,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,sBAAsB,CACpC,OAAyB,EACzB,OAA8D,EAC9D,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAA8D,EAC9D,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CACT,+DAA+D,CAAC,KAAK,CACnE,MAAM,CACP,EACH,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,OAAO,EAAE,YAAY,CAAC,SAAS,EAAE,QAAQ,CAAC,OAAO,EAAE;YACjD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,kCAAkC,CAAC,CAC1D,WAAW,CACZ,CAAC;IACF,MAAM,MAAM,GAAG,eAAe,CAAC;QAC7B,SAAS,EAAE,QAAQ,CAAC,OAAO;QAC3B,aAAa,EAAE,QAAQ,CAAC,WAAW;KACpC,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;QAC1B,WAAW,EAAE,YAAY,CAAC,WAAW,EAAE,QAAQ,CAAC,OAAO,EAAE;YACvD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,MAAM;SACrB,CAAC;KACH,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,0DAA0D;QACvE,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,KAAK,EAAE,MAAM;QACb,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
1
|
+
{"version":3,"file":"agentTestsGetBenchmark.js","sourceRoot":"","sources":["../../src/funcs/agentTestsGetBenchmark.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,+DAA+D,GAChE,MAAM,+DAA+D,CAAC;AACvE,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,sBAAsB,CACpC,OAAyB,EACzB,OAA8D,EAC9D,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAA8D,EAC9D,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CACT,+DAA+D,CAAC,KAAK,CACnE,MAAM,CACP,EACH,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,OAAO,EAAE,YAAY,CAAC,SAAS,EAAE,QAAQ,CAAC,OAAO,EAAE;YACjD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,kCAAkC,CAAC,CAC1D,WAAW,CACZ,CAAC;IACF,MAAM,MAAM,GAAG,eAAe,CAAC;QAC7B,SAAS,EAAE,QAAQ,CAAC,OAAO;QAC3B,MAAM,EAAE,QAAQ,CAAC,IAAI;QACrB,aAAa,EAAE,QAAQ,CAAC,WAAW;KACpC,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;QAC1B,WAAW,EAAE,YAAY,CAAC,WAAW,EAAE,QAAQ,CAAC,OAAO,EAAE;YACvD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,MAAM;SACrB,CAAC;KACH,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,0DAA0D;QACvE,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,KAAK,EAAE,MAAM;QACb,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -33,6 +33,7 @@ async function $do(client$, request, options) {
|
|
|
33
33
|
const path$ = pathToFunc("/agent-tests/run/{task_id}")(pathParams$);
|
|
34
34
|
const query$ = encodeFormQuery({
|
|
35
35
|
"compact": payload$.compact,
|
|
36
|
+
"mode": payload$.mode,
|
|
36
37
|
"only_failed": payload$.only_failed,
|
|
37
38
|
});
|
|
38
39
|
const headers$ = new Headers(compactMap({
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agentTestsGetRun.js","sourceRoot":"","sources":["../../src/funcs/agentTestsGetRun.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,4DAA4D,GAC7D,MAAM,4DAA4D,CAAC;AACpE,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAC9B,OAAyB,EACzB,OAA2D,EAC3D,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAA2D,EAC3D,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CACT,4DAA4D,CAAC,KAAK,CAChE,MAAM,CACP,EACH,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,OAAO,EAAE,YAAY,CAAC,SAAS,EAAE,QAAQ,CAAC,OAAO,EAAE;YACjD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,4BAA4B,CAAC,CACpD,WAAW,CACZ,CAAC;IACF,MAAM,MAAM,GAAG,eAAe,CAAC;QAC7B,SAAS,EAAE,QAAQ,CAAC,OAAO;QAC3B,aAAa,EAAE,QAAQ,CAAC,WAAW;KACpC,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;QAC1B,WAAW,EAAE,YAAY,CAAC,WAAW,EAAE,QAAQ,CAAC,OAAO,EAAE;YACvD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,MAAM;SACrB,CAAC;KACH,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,yDAAyD;QACtE,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,KAAK,EAAE,MAAM;QACb,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
1
|
+
{"version":3,"file":"agentTestsGetRun.js","sourceRoot":"","sources":["../../src/funcs/agentTestsGetRun.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,4DAA4D,GAC7D,MAAM,4DAA4D,CAAC;AACpE,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,gBAAgB,CAC9B,OAAyB,EACzB,OAA2D,EAC3D,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAA2D,EAC3D,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CACT,4DAA4D,CAAC,KAAK,CAChE,MAAM,CACP,EACH,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,OAAO,EAAE,YAAY,CAAC,SAAS,EAAE,QAAQ,CAAC,OAAO,EAAE;YACjD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,4BAA4B,CAAC,CACpD,WAAW,CACZ,CAAC;IACF,MAAM,MAAM,GAAG,eAAe,CAAC;QAC7B,SAAS,EAAE,QAAQ,CAAC,OAAO;QAC3B,MAAM,EAAE,QAAQ,CAAC,IAAI;QACrB,aAAa,EAAE,QAAQ,CAAC,WAAW;KACpC,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;QAC1B,WAAW,EAAE,YAAY,CAAC,WAAW,EAAE,QAAQ,CAAC,OAAO,EAAE;YACvD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,MAAM;SACrB,CAAC;KACH,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EAAE,yDAAyD;QACtE,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,KAAK,EAAE,MAAM;QACb,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
|
@@ -0,0 +1,16 @@
|
|
|
1
|
+
import { CalibrateMcpCore } from "../core.js";
|
|
2
|
+
import { RequestOptions } from "../lib/sdks.js";
|
|
3
|
+
import { APIError } from "../models/errors/apierror.js";
|
|
4
|
+
import { ConnectionError, InvalidRequestError, RequestAbortedError, RequestTimeoutError, UnexpectedClientError } from "../models/errors/httpclienterrors.js";
|
|
5
|
+
import { SDKValidationError } from "../models/errors/sdkvalidationerror.js";
|
|
6
|
+
import { GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest } from "../models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js";
|
|
7
|
+
import { APIPromise } from "../types/async.js";
|
|
8
|
+
import { Result } from "../types/fp.js";
|
|
9
|
+
/**
|
|
10
|
+
* Get test case result
|
|
11
|
+
*
|
|
12
|
+
* @remarks
|
|
13
|
+
* Get the full result of one test case in a run
|
|
14
|
+
*/
|
|
15
|
+
export declare function agentTestsGetRunCase(client$: CalibrateMcpCore, request: GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest, options?: RequestOptions): APIPromise<Result<Response, APIError | SDKValidationError | UnexpectedClientError | InvalidRequestError | RequestAbortedError | RequestTimeoutError | ConnectionError>>;
|
|
16
|
+
//# sourceMappingURL=agentTestsGetRunCase.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agentTestsGetRunCase.d.ts","sourceRoot":"","sources":["../../src/funcs/agentTestsGetRunCase.ts"],"names":[],"mappings":"AAIA,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAI9C,OAAO,EAAE,cAAc,EAAE,MAAM,gBAAgB,CAAC;AAGhD,OAAO,EAAE,QAAQ,EAAE,MAAM,8BAA8B,CAAC;AACxD,OAAO,EACL,eAAe,EACf,mBAAmB,EACnB,mBAAmB,EACnB,mBAAmB,EACnB,qBAAqB,EACtB,MAAM,sCAAsC,CAAC;AAC9C,OAAO,EAAE,kBAAkB,EAAE,MAAM,wCAAwC,CAAC;AAC5E,OAAO,EACL,kEAAkE,EAEnE,MAAM,4EAA4E,CAAC;AACpF,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AACxD,OAAO,EAAE,MAAM,EAAE,MAAM,gBAAgB,CAAC;AAExC;;;;;GAKG;AACH,wBAAgB,oBAAoB,CAClC,OAAO,EAAE,gBAAgB,EACzB,OAAO,EAAE,kEAAkE,EAC3E,OAAO,CAAC,EAAE,cAAc,GACvB,UAAU,CACX,MAAM,CACJ,QAAQ,EACN,QAAQ,GACR,kBAAkB,GAClB,qBAAqB,GACrB,mBAAmB,GACnB,mBAAmB,GACnB,mBAAmB,GACnB,eAAe,CAClB,CACF,CAMA"}
|
|
@@ -0,0 +1,100 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
import { encodeFormQuery, encodeSimple } from "../lib/encodings.js";
|
|
5
|
+
import { compactMap } from "../lib/primitives.js";
|
|
6
|
+
import { safeParse } from "../lib/schemas.js";
|
|
7
|
+
import { extractSecurity, resolveGlobalSecurity } from "../lib/security.js";
|
|
8
|
+
import { pathToFunc } from "../lib/url.js";
|
|
9
|
+
import { GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest$zodSchema, } from "../models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js";
|
|
10
|
+
import { APIPromise } from "../types/async.js";
|
|
11
|
+
/**
|
|
12
|
+
* Get test case result
|
|
13
|
+
*
|
|
14
|
+
* @remarks
|
|
15
|
+
* Get the full result of one test case in a run
|
|
16
|
+
*/
|
|
17
|
+
export function agentTestsGetRunCase(client$, request, options) {
|
|
18
|
+
return new APIPromise($do(client$, request, options));
|
|
19
|
+
}
|
|
20
|
+
async function $do(client$, request, options) {
|
|
21
|
+
const parsed$ = safeParse(request, (value$) => GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest$zodSchema
|
|
22
|
+
.parse(value$), "Input validation failed");
|
|
23
|
+
if (!parsed$.ok) {
|
|
24
|
+
return [parsed$, { status: "invalid" }];
|
|
25
|
+
}
|
|
26
|
+
const payload$ = parsed$.value;
|
|
27
|
+
const body$ = null;
|
|
28
|
+
const pathParams$ = {
|
|
29
|
+
task_id: encodeSimple("task_id", payload$.task_id, {
|
|
30
|
+
explode: false,
|
|
31
|
+
charEncoding: "percent",
|
|
32
|
+
}),
|
|
33
|
+
test_uuid: encodeSimple("test_uuid", payload$.test_uuid, {
|
|
34
|
+
explode: false,
|
|
35
|
+
charEncoding: "percent",
|
|
36
|
+
}),
|
|
37
|
+
};
|
|
38
|
+
const path$ = pathToFunc("/agent-tests/run/{task_id}/results/{test_uuid}")(pathParams$);
|
|
39
|
+
const query$ = encodeFormQuery({
|
|
40
|
+
"model": payload$.model,
|
|
41
|
+
});
|
|
42
|
+
const headers$ = new Headers(compactMap({
|
|
43
|
+
Accept: "application/json",
|
|
44
|
+
"X-API-Key": encodeSimple("X-API-Key", payload$.xAPIKey, {
|
|
45
|
+
explode: false,
|
|
46
|
+
charEncoding: "none",
|
|
47
|
+
}),
|
|
48
|
+
}));
|
|
49
|
+
const securityInput = await extractSecurity(client$._options.security);
|
|
50
|
+
const requestSecurity = resolveGlobalSecurity(securityInput);
|
|
51
|
+
const context = {
|
|
52
|
+
options: client$._options,
|
|
53
|
+
baseURL: options?.serverURL ?? client$._baseURL ?? "",
|
|
54
|
+
operationID: "get_agent_test_case_result_agent_tests_run__task_id__results__test_uuid__get",
|
|
55
|
+
oAuth2Scopes: null,
|
|
56
|
+
resolvedSecurity: requestSecurity,
|
|
57
|
+
securitySource: client$._options.security,
|
|
58
|
+
retryConfig: options?.retries
|
|
59
|
+
|| client$._options.retryConfig
|
|
60
|
+
|| { strategy: "none" },
|
|
61
|
+
retryCodes: options?.retryCodes || [
|
|
62
|
+
"429",
|
|
63
|
+
"500",
|
|
64
|
+
"502",
|
|
65
|
+
"503",
|
|
66
|
+
"504",
|
|
67
|
+
],
|
|
68
|
+
};
|
|
69
|
+
const requestRes = client$._createRequest(context, {
|
|
70
|
+
security: requestSecurity,
|
|
71
|
+
method: "GET",
|
|
72
|
+
baseURL: options?.serverURL,
|
|
73
|
+
path: path$,
|
|
74
|
+
headers: headers$,
|
|
75
|
+
query: query$,
|
|
76
|
+
body: body$,
|
|
77
|
+
userAgent: client$._options.userAgent,
|
|
78
|
+
timeoutMs: options?.timeoutMs || client$._options.timeoutMs
|
|
79
|
+
|| -1,
|
|
80
|
+
}, options);
|
|
81
|
+
if (!requestRes.ok) {
|
|
82
|
+
return [requestRes, { status: "invalid" }];
|
|
83
|
+
}
|
|
84
|
+
const req$ = requestRes.value;
|
|
85
|
+
const doResult = await client$._do(req$, {
|
|
86
|
+
context,
|
|
87
|
+
errorCodes: [],
|
|
88
|
+
retryConfig: context.retryConfig,
|
|
89
|
+
retryCodes: context.retryCodes,
|
|
90
|
+
});
|
|
91
|
+
if (!doResult.ok) {
|
|
92
|
+
return [doResult, { status: "request-error", request: req$ }];
|
|
93
|
+
}
|
|
94
|
+
return [doResult, {
|
|
95
|
+
status: "complete",
|
|
96
|
+
"request": req$,
|
|
97
|
+
response: doResult.value,
|
|
98
|
+
}];
|
|
99
|
+
}
|
|
100
|
+
//# sourceMappingURL=agentTestsGetRunCase.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agentTestsGetRunCase.js","sourceRoot":"","sources":["../../src/funcs/agentTestsGetRunCase.ts"],"names":[],"mappings":"AAAA;;GAEG;AAGH,OAAO,EAAE,eAAe,EAAE,YAAY,EAAE,MAAM,qBAAqB,CAAC;AACpE,OAAO,EAAE,UAAU,EAAE,MAAM,sBAAsB,CAAC;AAClD,OAAO,EAAE,SAAS,EAAE,MAAM,mBAAmB,CAAC;AAE9C,OAAO,EAAE,eAAe,EAAE,qBAAqB,EAAE,MAAM,oBAAoB,CAAC;AAC5E,OAAO,EAAE,UAAU,EAAE,MAAM,eAAe,CAAC;AAU3C,OAAO,EAEL,4EAA4E,GAC7E,MAAM,4EAA4E,CAAC;AACpF,OAAO,EAAW,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAGxD;;;;;GAKG;AACH,MAAM,UAAU,oBAAoB,CAClC,OAAyB,EACzB,OAA2E,EAC3E,OAAwB;IAaxB,OAAO,IAAI,UAAU,CAAC,GAAG,CACvB,OAAO,EACP,OAAO,EACP,OAAO,CACR,CAAC,CAAC;AACL,CAAC;AAED,KAAK,UAAU,GAAG,CAChB,OAAyB,EACzB,OAA2E,EAC3E,OAAwB;IAgBxB,MAAM,OAAO,GAAG,SAAS,CACvB,OAAO,EACP,CAAC,MAAM,EAAE,EAAE,CACT,4EAA4E;SACzE,KAAK,CAAC,MAAM,CAAC,EAClB,yBAAyB,CAC1B,CAAC;IACF,IAAI,CAAC,OAAO,CAAC,EAAE,EAAE,CAAC;QAChB,OAAO,CAAC,OAAO,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC1C,CAAC;IACD,MAAM,QAAQ,GAAG,OAAO,CAAC,KAAK,CAAC;IAC/B,MAAM,KAAK,GAAG,IAAI,CAAC;IAEnB,MAAM,WAAW,GAAG;QAClB,OAAO,EAAE,YAAY,CAAC,SAAS,EAAE,QAAQ,CAAC,OAAO,EAAE;YACjD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;QACF,SAAS,EAAE,YAAY,CAAC,WAAW,EAAE,QAAQ,CAAC,SAAS,EAAE;YACvD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,SAAS;SACxB,CAAC;KACH,CAAC;IACF,MAAM,KAAK,GAAG,UAAU,CAAC,gDAAgD,CAAC,CACxE,WAAW,CACZ,CAAC;IACF,MAAM,MAAM,GAAG,eAAe,CAAC;QAC7B,OAAO,EAAE,QAAQ,CAAC,KAAK;KACxB,CAAC,CAAC;IAEH,MAAM,QAAQ,GAAG,IAAI,OAAO,CAAC,UAAU,CAAC;QACtC,MAAM,EAAE,kBAAkB;QAC1B,WAAW,EAAE,YAAY,CAAC,WAAW,EAAE,QAAQ,CAAC,OAAO,EAAE;YACvD,OAAO,EAAE,KAAK;YACd,YAAY,EAAE,MAAM;SACrB,CAAC;KACH,CAAC,CAAC,CAAC;IACJ,MAAM,aAAa,GAAG,MAAM,eAAe,CAAC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IACvE,MAAM,eAAe,GAAG,qBAAqB,CAAC,aAAa,CAAC,CAAC;IAE7D,MAAM,OAAO,GAAG;QACd,OAAO,EAAE,OAAO,CAAC,QAAQ;QACzB,OAAO,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,IAAI,EAAE;QACrD,WAAW,EACT,8EAA8E;QAChF,YAAY,EAAE,IAAI;QAClB,gBAAgB,EAAE,eAAe;QACjC,cAAc,EAAE,OAAO,CAAC,QAAQ,CAAC,QAAQ;QACzC,WAAW,EAAE,OAAO,EAAE,OAAO;eACxB,OAAO,CAAC,QAAQ,CAAC,WAAW;eAC5B,EAAE,QAAQ,EAAE,MAAM,EAAE;QACzB,UAAU,EAAE,OAAO,EAAE,UAAU,IAAI;YACjC,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;YACL,KAAK;SACN;KACF,CAAC;IAEF,MAAM,UAAU,GAAG,OAAO,CAAC,cAAc,CAAC,OAAO,EAAE;QACjD,QAAQ,EAAE,eAAe;QACzB,MAAM,EAAE,KAAK;QACb,OAAO,EAAE,OAAO,EAAE,SAAS;QAC3B,IAAI,EAAE,KAAK;QACX,OAAO,EAAE,QAAQ;QACjB,KAAK,EAAE,MAAM;QACb,IAAI,EAAE,KAAK;QACX,SAAS,EAAE,OAAO,CAAC,QAAQ,CAAC,SAAS;QACrC,SAAS,EAAE,OAAO,EAAE,SAAS,IAAI,OAAO,CAAC,QAAQ,CAAC,SAAS;eACtD,CAAC,CAAC;KACR,EAAE,OAAO,CAAC,CAAC;IACZ,IAAI,CAAC,UAAU,CAAC,EAAE,EAAE,CAAC;QACnB,OAAO,CAAC,UAAU,EAAE,EAAE,MAAM,EAAE,SAAS,EAAE,CAAC,CAAC;IAC7C,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,KAAK,CAAC;IAE9B,MAAM,QAAQ,GAAG,MAAM,OAAO,CAAC,GAAG,CAAC,IAAI,EAAE;QACvC,OAAO;QACP,UAAU,EAAE,EAAE;QACd,WAAW,EAAE,OAAO,CAAC,WAAW;QAChC,UAAU,EAAE,OAAO,CAAC,UAAU;KAC/B,CAAC,CAAC;IACH,IAAI,CAAC,QAAQ,CAAC,EAAE,EAAE,CAAC;QACjB,OAAO,CAAC,QAAQ,EAAE,EAAE,MAAM,EAAE,eAAe,EAAE,OAAO,EAAE,IAAI,EAAE,CAAC,CAAC;IAChE,CAAC;IACD,OAAO,CAAC,QAAQ,EAAE;YAChB,MAAM,EAAE,UAAU;YAClB,SAAS,EAAE,IAAI;YACf,QAAQ,EAAE,QAAQ,CAAC,KAAK;SACzB,CAAC,CAAC;AACL,CAAC"}
|
package/esm/landing-page.js
CHANGED
|
@@ -915,7 +915,7 @@ http_headers = { "api-key-auth" = "YOUR_API_KEY_AUTH" }`;
|
|
|
915
915
|
<h1>Instructions</h1>
|
|
916
916
|
<p>One-click installation for Claude Desktop users</p>
|
|
917
917
|
<div class="instruction-item">
|
|
918
|
-
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.
|
|
918
|
+
<a href="https://github.com/dalmia/calibrate-mcp/releases/download/v0.0.54/mcp-server.mcpb" download="mcp-server.mcpb" class="action-button header-action" style="display: inline-flex; margin-bottom: 16px;">
|
|
919
919
|
📥 Download MCP Bundle
|
|
920
920
|
</a>
|
|
921
921
|
</div>
|
package/esm/lib/config.d.ts
CHANGED
|
@@ -35,8 +35,8 @@ export declare function serverURLFromOptions(options: SDKOptions): URL | null;
|
|
|
35
35
|
export declare const SDK_METADATA: {
|
|
36
36
|
readonly language: "typescript";
|
|
37
37
|
readonly openapiDocVersion: "0.1.0";
|
|
38
|
-
readonly sdkVersion: "0.0.
|
|
38
|
+
readonly sdkVersion: "0.0.54";
|
|
39
39
|
readonly genVersion: "2.915.1";
|
|
40
|
-
readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.
|
|
40
|
+
readonly userAgent: "speakeasy-sdk/mcp-typescript 0.0.54 2.915.1 0.1.0 @dalmia/calibrate-mcp";
|
|
41
41
|
};
|
|
42
42
|
//# sourceMappingURL=config.d.ts.map
|
package/esm/lib/config.js
CHANGED
|
@@ -27,8 +27,8 @@ export function serverURLFromOptions(options) {
|
|
|
27
27
|
export const SDK_METADATA = {
|
|
28
28
|
language: "typescript",
|
|
29
29
|
openapiDocVersion: "0.1.0",
|
|
30
|
-
sdkVersion: "0.0.
|
|
30
|
+
sdkVersion: "0.0.54",
|
|
31
31
|
genVersion: "2.915.1",
|
|
32
|
-
userAgent: "speakeasy-sdk/mcp-typescript 0.0.
|
|
32
|
+
userAgent: "speakeasy-sdk/mcp-typescript 0.0.54 2.915.1 0.1.0 @dalmia/calibrate-mcp",
|
|
33
33
|
};
|
|
34
34
|
//# sourceMappingURL=config.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/mcp-server/server.ts"],"names":[],"mappings":"AAIA,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AACpE,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAC9C,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAMzD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACvC,OAAO,EAEL,uBAAuB,EAExB,MAAM,YAAY,CAAC;
|
|
1
|
+
{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/mcp-server/server.ts"],"names":[],"mappings":"AAIA,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AACpE,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAC9C,OAAO,EAAE,UAAU,EAAE,MAAM,kBAAkB,CAAC;AAC9C,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AAMzD,OAAO,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AACvC,OAAO,EAEL,uBAAuB,EAExB,MAAM,YAAY,CAAC;AA4CpB,wBAAgB,eAAe,CAAC,IAAI,EAAE;IACpC,MAAM,EAAE,aAAa,CAAC;IACtB,YAAY,CAAC,EAAE,MAAM,EAAE,GAAG,SAAS,CAAC;IACpC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,MAAM,CAAC,EAAE,QAAQ,EAAE,GAAG,SAAS,CAAC;IAChC,gBAAgB,CAAC,EAAE,uBAAuB,GAAG,SAAS,CAAC;IACvD,MAAM,CAAC,EAAE,MAAM,gBAAgB,CAAC;IAChC,SAAS,CAAC,EAAE,MAAM,GAAG,SAAS,CAAC;IAC/B,QAAQ,CAAC,EAAE,UAAU,CAAC,UAAU,CAAC,GAAG,SAAS,CAAC;IAC9C,SAAS,CAAC,EAAE,UAAU,CAAC,WAAW,CAAC,GAAG,SAAS,CAAC;CACjD;;;;;;EAgGA"}
|
package/esm/mcp-server/server.js
CHANGED
|
@@ -18,6 +18,7 @@ import { tool$agentTestsBenchmark } from "./tools/agentTestsBenchmark.js";
|
|
|
18
18
|
import { tool$agentTestsBulkUnlink } from "./tools/agentTestsBulkUnlink.js";
|
|
19
19
|
import { tool$agentTestsGetBenchmark } from "./tools/agentTestsGetBenchmark.js";
|
|
20
20
|
import { tool$agentTestsGetRun } from "./tools/agentTestsGetRun.js";
|
|
21
|
+
import { tool$agentTestsGetRunCase } from "./tools/agentTestsGetRunCase.js";
|
|
21
22
|
import { tool$agentTestsLink } from "./tools/agentTestsLink.js";
|
|
22
23
|
import { tool$agentTestsListForAgent } from "./tools/agentTestsListForAgent.js";
|
|
23
24
|
import { tool$agentTestsListRunsForAgent } from "./tools/agentTestsListRunsForAgent.js";
|
|
@@ -50,7 +51,7 @@ import { tool$tracesCreate } from "./tools/tracesCreate.js";
|
|
|
50
51
|
export function createMCPServer(deps) {
|
|
51
52
|
const server = new McpServer({
|
|
52
53
|
name: "CalibrateMcp",
|
|
53
|
-
version: "0.0.
|
|
54
|
+
version: "0.0.54",
|
|
54
55
|
});
|
|
55
56
|
const getClient = deps.getSDK || (() => new CalibrateMcpCore({
|
|
56
57
|
security: deps.security,
|
|
@@ -92,6 +93,7 @@ export function createMCPServer(deps) {
|
|
|
92
93
|
tool(tool$agentTestsRun);
|
|
93
94
|
tool(tool$agentTestsRunBatch);
|
|
94
95
|
tool(tool$agentTestsGetRun);
|
|
96
|
+
tool(tool$agentTestsGetRunCase);
|
|
95
97
|
tool(tool$agentTestsBenchmark);
|
|
96
98
|
tool(tool$agentTestsGetBenchmark);
|
|
97
99
|
tool(tool$evaluatorsCreate);
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"server.js","sourceRoot":"","sources":["../../src/mcp-server/server.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AACpE,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAG9C,OAAO,EAAE,oBAAoB,EAAE,MAAM,cAAc,CAAC;AACpD,OAAO,EACL,sBAAsB,EACtB,8BAA8B,GAC/B,MAAM,gBAAgB,CAAC;AAExB,OAAO,EACL,kBAAkB,EAElB,oBAAoB,GACrB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,yBAAyB,EAAE,MAAM,iCAAiC,CAAC;AAC5E,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,yBAAyB,EAAE,MAAM,iCAAiC,CAAC;AAC5E,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,2BAA2B,EAAE,MAAM,mCAAmC,CAAC;AAChF,OAAO,EAAE,wBAAwB,EAAE,MAAM,gCAAgC,CAAC;AAC1E,OAAO,EAAE,yBAAyB,EAAE,MAAM,iCAAiC,CAAC;AAC5E,OAAO,EAAE,2BAA2B,EAAE,MAAM,mCAAmC,CAAC;AAChF,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,2BAA2B,EAAE,MAAM,mCAAmC,CAAC;AAChF,OAAO,EAAE,+BAA+B,EAAE,MAAM,uCAAuC,CAAC;AACxF,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,4BAA4B,EAAE,MAAM,oCAAoC,CAAC;AAClF,OAAO,EAAE,0BAA0B,EAAE,MAAM,kCAAkC,CAAC;AAC9E,OAAO,EAAE,sCAAsC,EAAE,MAAM,8CAA8C,CAAC;AACtG,OAAO,EAAE,uCAAuC,EAAE,MAAM,+CAA+C,CAAC;AACxG,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,gCAAgC,EAAE,MAAM,wCAAwC,CAAC;AAC1F,OAAO,EAAE,mCAAmC,EAAE,MAAM,2CAA2C,CAAC;AAChG,OAAO,EAAE,8BAA8B,EAAE,MAAM,sCAAsC,CAAC;AACtF,OAAO,EAAE,wBAAwB,EAAE,MAAM,gCAAgC,CAAC;AAC1E,OAAO,EAAE,iCAAiC,EAAE,MAAM,yCAAyC,CAAC;AAC5F,OAAO,EAAE,+BAA+B,EAAE,MAAM,uCAAuC,CAAC;AACxF,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,4BAA4B,EAAE,MAAM,oCAAoC,CAAC;AAClF,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAE5D,MAAM,UAAU,eAAe,CAAC,IAU/B;IACC,MAAM,MAAM,GAAG,IAAI,SAAS,CAAC;QAC3B,IAAI,EAAE,cAAc;QACpB,OAAO,EAAE,QAAQ;KAClB,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,IAAI,CAAC,GAAG,EAAE,CACrC,IAAI,gBAAgB,CAAC;QACnB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,KAAK,OAAO;YACxC,CAAC,CAAC;gBACA,GAAG,EAAE,CAAC,GAAG,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC;gBACtC,KAAK,EAAE,CAAC,GAAG,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC;gBAC1C,QAAQ,EAAE,CAAC,GAAG,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,GAAG,IAAI,CAAC;aACjD;YACD,CAAC,CAAC,SAAS;KACd,CAAC,CAAC,CAAC;IAEN,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAEpC,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IACrE,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,GAAG,kBAAkB,CAC/C,IAAI,CAAC,MAAM,EACX,MAAM,EACN,SAAS,EACT,MAAM,EACN,YAAY,EACZ,IAAI,CAAC,OAAO,EACZ,IAAI,CAAC,gBAAgB,CACtB,CAAC;IACF,MAAM,QAAQ,GAAG,sBAAsB,CACrC,IAAI,CAAC,MAAM,EACX,MAAM,EACN,SAAS,EACT,MAAM,CACP,CAAC;IACF,MAAM,gBAAgB,GAAG,8BAA8B,CACrD,IAAI,CAAC,MAAM,EACX,MAAM,EACN,SAAS,EACT,MAAM,CACP,CAAC;IACF,MAAM,MAAM,GAAG,oBAAoB,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;IAC5E,MAAM,QAAQ,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC9D,KAAK,QAAQ,CAAC,CAAC,2BAA2B;IAE1C,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAClC,IAAI,CAAC,kBAAkB,CAAC,CAAC;IACzB,IAAI,CAAC,iBAAiB,CAAC,CAAC;IACxB,IAAI,CAAC,eAAe,CAAC,CAAC;IACtB,IAAI,CAAC,cAAc,CAAC,CAAC;IACrB,IAAI,CAAC,iBAAiB,CAAC,CAAC;IACxB,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAChC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAChC,IAAI,CAAC,oBAAoB,CAAC,CAAC;IAC3B,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACvB,IAAI,CAAC,cAAc,CAAC,CAAC;IACrB,IAAI,CAAC,aAAa,CAAC,CAAC;IACpB,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACvB,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAC1B,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAClC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACtC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAChC,IAAI,CAAC,kBAAkB,CAAC,CAAC;IACzB,IAAI,CAAC,uBAAuB,CAAC,CAAC;IAC9B,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC/B,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAClC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAC1B,IAAI,CAAC,kBAAkB,CAAC,CAAC;IACzB,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACnC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACjC,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC/B,IAAI,CAAC,uBAAuB,CAAC,CAAC;IAC9B,IAAI,CAAC,iCAAiC,CAAC,CAAC;IACxC,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACnC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACtC,IAAI,CAAC,uCAAuC,CAAC,CAAC;IAC9C,IAAI,CAAC,sCAAsC,CAAC,CAAC;IAC7C,IAAI,CAAC,mCAAmC,CAAC,CAAC;IAC1C,IAAI,CAAC,gCAAgC,CAAC,CAAC;IACvC,IAAI,CAAC,8BAA8B,CAAC,CAAC;IACrC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAC1B,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,iBAAiB,CAAC,CAAC;IAExB,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;QACjB,oBAAoB,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACxE,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;AAC3B,CAAC"}
|
|
1
|
+
{"version":3,"file":"server.js","sourceRoot":"","sources":["../../src/mcp-server/server.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,SAAS,EAAE,MAAM,yCAAyC,CAAC;AACpE,OAAO,EAAE,gBAAgB,EAAE,MAAM,YAAY,CAAC;AAG9C,OAAO,EAAE,oBAAoB,EAAE,MAAM,cAAc,CAAC;AACpD,OAAO,EACL,sBAAsB,EACtB,8BAA8B,GAC/B,MAAM,gBAAgB,CAAC;AAExB,OAAO,EACL,kBAAkB,EAElB,oBAAoB,GACrB,MAAM,YAAY,CAAC;AACpB,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,yBAAyB,EAAE,MAAM,iCAAiC,CAAC;AAC5E,OAAO,EAAE,eAAe,EAAE,MAAM,uBAAuB,CAAC;AACxD,OAAO,EAAE,yBAAyB,EAAE,MAAM,iCAAiC,CAAC;AAC5E,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAC5D,OAAO,EAAE,2BAA2B,EAAE,MAAM,mCAAmC,CAAC;AAChF,OAAO,EAAE,wBAAwB,EAAE,MAAM,gCAAgC,CAAC;AAC1E,OAAO,EAAE,yBAAyB,EAAE,MAAM,iCAAiC,CAAC;AAC5E,OAAO,EAAE,2BAA2B,EAAE,MAAM,mCAAmC,CAAC;AAChF,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,yBAAyB,EAAE,MAAM,iCAAiC,CAAC;AAC5E,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,2BAA2B,EAAE,MAAM,mCAAmC,CAAC;AAChF,OAAO,EAAE,+BAA+B,EAAE,MAAM,uCAAuC,CAAC;AACxF,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,4BAA4B,EAAE,MAAM,oCAAoC,CAAC;AAClF,OAAO,EAAE,0BAA0B,EAAE,MAAM,kCAAkC,CAAC;AAC9E,OAAO,EAAE,sCAAsC,EAAE,MAAM,8CAA8C,CAAC;AACtG,OAAO,EAAE,uCAAuC,EAAE,MAAM,+CAA+C,CAAC;AACxG,OAAO,EAAE,uBAAuB,EAAE,MAAM,+BAA+B,CAAC;AACxE,OAAO,EAAE,gCAAgC,EAAE,MAAM,wCAAwC,CAAC;AAC1F,OAAO,EAAE,mCAAmC,EAAE,MAAM,2CAA2C,CAAC;AAChG,OAAO,EAAE,8BAA8B,EAAE,MAAM,sCAAsC,CAAC;AACtF,OAAO,EAAE,wBAAwB,EAAE,MAAM,gCAAgC,CAAC;AAC1E,OAAO,EAAE,iCAAiC,EAAE,MAAM,yCAAyC,CAAC;AAC5F,OAAO,EAAE,+BAA+B,EAAE,MAAM,uCAAuC,CAAC;AACxF,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,qBAAqB,EAAE,MAAM,6BAA6B,CAAC;AACpE,OAAO,EAAE,4BAA4B,EAAE,MAAM,oCAAoC,CAAC;AAClF,OAAO,EAAE,kBAAkB,EAAE,MAAM,0BAA0B,CAAC;AAC9D,OAAO,EAAE,mBAAmB,EAAE,MAAM,2BAA2B,CAAC;AAChE,OAAO,EAAE,oBAAoB,EAAE,MAAM,4BAA4B,CAAC;AAClE,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,aAAa,EAAE,MAAM,qBAAqB,CAAC;AACpD,OAAO,EAAE,cAAc,EAAE,MAAM,sBAAsB,CAAC;AACtD,OAAO,EAAE,gBAAgB,EAAE,MAAM,wBAAwB,CAAC;AAC1D,OAAO,EAAE,iBAAiB,EAAE,MAAM,yBAAyB,CAAC;AAE5D,MAAM,UAAU,eAAe,CAAC,IAU/B;IACC,MAAM,MAAM,GAAG,IAAI,SAAS,CAAC;QAC3B,IAAI,EAAE,cAAc;QACpB,OAAO,EAAE,QAAQ;KAClB,CAAC,CAAC;IAEH,MAAM,SAAS,GAAG,IAAI,CAAC,MAAM,IAAI,CAAC,GAAG,EAAE,CACrC,IAAI,gBAAgB,CAAC;QACnB,QAAQ,EAAE,IAAI,CAAC,QAAQ;QACvB,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,SAAS,EAAE,IAAI,CAAC,SAAS;QACzB,WAAW,EAAE,IAAI,CAAC,MAAM,CAAC,KAAK,KAAK,OAAO;YACxC,CAAC,CAAC;gBACA,GAAG,EAAE,CAAC,GAAG,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,GAAG,CAAC,GAAG,IAAI,CAAC;gBACtC,KAAK,EAAE,CAAC,GAAG,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,GAAG,IAAI,CAAC;gBAC1C,QAAQ,EAAE,CAAC,GAAG,IAAI,EAAE,EAAE,CAAC,OAAO,CAAC,QAAQ,CAAC,GAAG,IAAI,CAAC;aACjD;YACD,CAAC,CAAC,SAAS;KACd,CAAC,CAAC,CAAC;IAEN,MAAM,MAAM,GAAG,IAAI,GAAG,CAAC,IAAI,CAAC,MAAM,CAAC,CAAC;IAEpC,MAAM,YAAY,GAAG,IAAI,CAAC,YAAY,IAAI,IAAI,GAAG,CAAC,IAAI,CAAC,YAAY,CAAC,CAAC;IACrE,MAAM,CAAC,IAAI,EAAE,KAAK,EAAE,OAAO,CAAC,GAAG,kBAAkB,CAC/C,IAAI,CAAC,MAAM,EACX,MAAM,EACN,SAAS,EACT,MAAM,EACN,YAAY,EACZ,IAAI,CAAC,OAAO,EACZ,IAAI,CAAC,gBAAgB,CACtB,CAAC;IACF,MAAM,QAAQ,GAAG,sBAAsB,CACrC,IAAI,CAAC,MAAM,EACX,MAAM,EACN,SAAS,EACT,MAAM,CACP,CAAC;IACF,MAAM,gBAAgB,GAAG,8BAA8B,CACrD,IAAI,CAAC,MAAM,EACX,MAAM,EACN,SAAS,EACT,MAAM,CACP,CAAC;IACF,MAAM,MAAM,GAAG,oBAAoB,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,MAAM,CAAC,CAAC;IAC5E,MAAM,QAAQ,GAAG,EAAE,IAAI,EAAE,QAAQ,EAAE,gBAAgB,EAAE,MAAM,EAAE,CAAC;IAC9D,KAAK,QAAQ,CAAC,CAAC,2BAA2B;IAE1C,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAClC,IAAI,CAAC,kBAAkB,CAAC,CAAC;IACzB,IAAI,CAAC,iBAAiB,CAAC,CAAC;IACxB,IAAI,CAAC,eAAe,CAAC,CAAC;IACtB,IAAI,CAAC,cAAc,CAAC,CAAC;IACrB,IAAI,CAAC,iBAAiB,CAAC,CAAC;IACxB,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAChC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAChC,IAAI,CAAC,oBAAoB,CAAC,CAAC;IAC3B,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACvB,IAAI,CAAC,cAAc,CAAC,CAAC;IACrB,IAAI,CAAC,aAAa,CAAC,CAAC;IACpB,IAAI,CAAC,gBAAgB,CAAC,CAAC;IACvB,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAC1B,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAClC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACtC,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAChC,IAAI,CAAC,kBAAkB,CAAC,CAAC;IACzB,IAAI,CAAC,uBAAuB,CAAC,CAAC;IAC9B,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,yBAAyB,CAAC,CAAC;IAChC,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC/B,IAAI,CAAC,2BAA2B,CAAC,CAAC;IAClC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAC1B,IAAI,CAAC,kBAAkB,CAAC,CAAC;IACzB,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACnC,IAAI,CAAC,0BAA0B,CAAC,CAAC;IACjC,IAAI,CAAC,wBAAwB,CAAC,CAAC;IAC/B,IAAI,CAAC,uBAAuB,CAAC,CAAC;IAC9B,IAAI,CAAC,iCAAiC,CAAC,CAAC;IACxC,IAAI,CAAC,4BAA4B,CAAC,CAAC;IACnC,IAAI,CAAC,+BAA+B,CAAC,CAAC;IACtC,IAAI,CAAC,uCAAuC,CAAC,CAAC;IAC9C,IAAI,CAAC,sCAAsC,CAAC,CAAC;IAC7C,IAAI,CAAC,mCAAmC,CAAC,CAAC;IAC1C,IAAI,CAAC,gCAAgC,CAAC,CAAC;IACvC,IAAI,CAAC,8BAA8B,CAAC,CAAC;IACrC,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,mBAAmB,CAAC,CAAC;IAC1B,IAAI,CAAC,qBAAqB,CAAC,CAAC;IAC5B,IAAI,CAAC,iBAAiB,CAAC,CAAC;IAExB,IAAI,IAAI,CAAC,OAAO,EAAE,CAAC;QACjB,oBAAoB,CAAC,IAAI,CAAC,MAAM,EAAE,MAAM,EAAE,SAAS,EAAE,OAAO,EAAE,MAAM,CAAC,CAAC;IACxE,CAAC;IAED,OAAO,EAAE,MAAM,EAAE,KAAK,EAAE,CAAC;AAC3B,CAAC"}
|
|
@@ -0,0 +1,7 @@
|
|
|
1
|
+
import { ToolDefinition } from "../tools.js";
|
|
2
|
+
declare const args: {
|
|
3
|
+
request: import("zod").ZodType<import("../../models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js").GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest, unknown, import("zod/v4/core").$ZodTypeInternals<import("../../models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js").GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest, unknown>>;
|
|
4
|
+
};
|
|
5
|
+
export declare const tool$agentTestsGetRunCase: ToolDefinition<typeof args>;
|
|
6
|
+
export {};
|
|
7
|
+
//# sourceMappingURL=agentTestsGetRunCase.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agentTestsGetRunCase.d.ts","sourceRoot":"","sources":["../../../src/mcp-server/tools/agentTestsGetRunCase.ts"],"names":[],"mappings":"AAMA,OAAO,EAAgB,cAAc,EAAE,MAAM,aAAa,CAAC;AAE3D,QAAA,MAAM,IAAI;;CAGT,CAAC;AAEF,eAAO,MAAM,yBAAyB,EAAE,cAAc,CAAC,OAAO,IAAI,CAiCjE,CAAC"}
|
|
@@ -0,0 +1,38 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
import { agentTestsGetRunCase } from "../../funcs/agentTestsGetRunCase.js";
|
|
5
|
+
import { GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest$zodSchema } from "../../models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js";
|
|
6
|
+
import { formatResult } from "../tools.js";
|
|
7
|
+
const args = {
|
|
8
|
+
request: GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest$zodSchema,
|
|
9
|
+
};
|
|
10
|
+
export const tool$agentTestsGetRunCase = {
|
|
11
|
+
name: "get-agent-test-case-result",
|
|
12
|
+
description: `Fetch the full result of one test case in a run, including the
|
|
13
|
+
conversation, the agent's output and each evaluator's reasoning. Pass
|
|
14
|
+
the model query param for a benchmark, which runs every test once per
|
|
15
|
+
model. Use with mode=summary on get-agent-test-run, which returns
|
|
16
|
+
light rows without this detail.
|
|
17
|
+
`,
|
|
18
|
+
scopes: ["read"],
|
|
19
|
+
annotations: {
|
|
20
|
+
"title": "Get agent test case result",
|
|
21
|
+
"destructiveHint": false,
|
|
22
|
+
"idempotentHint": false,
|
|
23
|
+
"openWorldHint": true,
|
|
24
|
+
"readOnlyHint": true,
|
|
25
|
+
},
|
|
26
|
+
args,
|
|
27
|
+
tool: async (client, args, ctx) => {
|
|
28
|
+
const [result] = await agentTestsGetRunCase(client, args.request, { fetchOptions: { signal: ctx.signal } }).$inspect();
|
|
29
|
+
if (!result.ok) {
|
|
30
|
+
return {
|
|
31
|
+
content: [{ type: "text", text: result.error.message }],
|
|
32
|
+
isError: true,
|
|
33
|
+
};
|
|
34
|
+
}
|
|
35
|
+
return formatResult(result.value);
|
|
36
|
+
},
|
|
37
|
+
};
|
|
38
|
+
//# sourceMappingURL=agentTestsGetRunCase.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"agentTestsGetRunCase.js","sourceRoot":"","sources":["../../../src/mcp-server/tools/agentTestsGetRunCase.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,EAAE,oBAAoB,EAAE,MAAM,qCAAqC,CAAC;AAC3E,OAAO,EAAE,4EAA4E,EAAE,MAAM,+EAA+E,CAAC;AAC7K,OAAO,EAAE,YAAY,EAAkB,MAAM,aAAa,CAAC;AAE3D,MAAM,IAAI,GAAG;IACX,OAAO,EACL,4EAA4E;CAC/E,CAAC;AAEF,MAAM,CAAC,MAAM,yBAAyB,GAAgC;IACpE,IAAI,EAAE,4BAA4B;IAClC,WAAW,EAAE;;;;;CAKd;IACC,MAAM,EAAE,CAAC,MAAM,CAAC;IAChB,WAAW,EAAE;QACX,OAAO,EAAE,4BAA4B;QACrC,iBAAiB,EAAE,KAAK;QACxB,gBAAgB,EAAE,KAAK;QACvB,eAAe,EAAE,IAAI;QACrB,cAAc,EAAE,IAAI;KACrB;IACD,IAAI;IACJ,IAAI,EAAE,KAAK,EAAE,MAAM,EAAE,IAAI,EAAE,GAAG,EAAE,EAAE;QAChC,MAAM,CAAC,MAAM,CAAC,GAAG,MAAM,oBAAoB,CACzC,MAAM,EACN,IAAI,CAAC,OAAO,EACZ,EAAE,YAAY,EAAE,EAAE,MAAM,EAAE,GAAG,CAAC,MAAM,EAAE,EAAE,CACzC,CAAC,QAAQ,EAAE,CAAC;QAEb,IAAI,CAAC,MAAM,CAAC,EAAE,EAAE,CAAC;YACf,OAAO;gBACL,OAAO,EAAE,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,CAAC,KAAK,CAAC,OAAO,EAAE,CAAC;gBACvD,OAAO,EAAE,IAAI;aACd,CAAC;QACJ,CAAC;QAED,OAAO,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC,CAAC;IACpC,CAAC;CACF,CAAC"}
|
|
@@ -33,7 +33,7 @@ export const AgentTestRunListItem$zodSchema = z
|
|
|
33
33
|
latency_ms: z.record(z.string(), z.any()).nullable().optional().describe("Aggregated latency in milliseconds, as `{p50, p95, p99, count}`"),
|
|
34
34
|
model_results: z.array(ModelRunSummary$zodSchema).nullable().optional()
|
|
35
35
|
.describe("Flat summary for each model in a benchmark run (fetch the benchmark detail for full results)"),
|
|
36
|
-
name: z.string().describe("
|
|
36
|
+
name: z.string().describe("Name of the run. A run nobody has renamed shows its number instead, such as `Run 1` for a test run or `Benchmark 1` for a benchmark"),
|
|
37
37
|
passed: z.int().nullable().optional().describe("Number of test cases that passed"),
|
|
38
38
|
results: z.array(TestRunCaseSummary$zodSchema).nullable().optional()
|
|
39
39
|
.describe("Flat pass/fail summary for each test case (fetch the run detail for full results)"),
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"agenttestrunlistitem.js","sourceRoot":"","sources":["../../src/models/agenttestrunlistitem.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,yBAAyB,GAC1B,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,4BAA4B,GAC7B,MAAM,yBAAyB,CAAC;AAEjC;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG;IACtC,WAAW,EAAE,eAAe;IAC5B,YAAY,EAAE,eAAe;CACrB,CAAC;AAYX,MAAM,CAAC,MAAM,kCAAkC,GAAG,CAAC,CAAC,IAAI,CAAC;IACvD,eAAe;IACf,eAAe;CAChB,CAAC,CAAC,QAAQ,CACT,8HAA8H,CAC/H,CAAC;AAyBF,MAAM,CAAC,MAAM,8BAA8B,GAAoC,CAAC;KAC7E,MAAM,CAAC;IACN,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAChE,oDAAoD,CACrD;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IAC1E,KAAK,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,wBAAwB,CAAC;IACpE,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,0BAA0B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACjE,sQAAsQ,CACvQ;IACD,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5C,kCAAkC,CACnC;IACD,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtE,iEAAiE,CAClE;IACD,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,yBAAyB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SACpE,QAAQ,CACP,8FAA8F,CAC/F;IACH,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACvB,
|
|
1
|
+
{"version":3,"file":"agenttestrunlistitem.js","sourceRoot":"","sources":["../../src/models/agenttestrunlistitem.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,yBAAyB,GAC1B,MAAM,sBAAsB,CAAC;AAC9B,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAC/B,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,4BAA4B,GAC7B,MAAM,yBAAyB,CAAC;AAEjC;;;;;;GAMG;AACH,MAAM,CAAC,MAAM,wBAAwB,GAAG;IACtC,WAAW,EAAE,eAAe;IAC5B,YAAY,EAAE,eAAe;CACrB,CAAC;AAYX,MAAM,CAAC,MAAM,kCAAkC,GAAG,CAAC,CAAC,IAAI,CAAC;IACvD,eAAe;IACf,eAAe;CAChB,CAAC,CAAC,QAAQ,CACT,8HAA8H,CAC/H,CAAC;AAyBF,MAAM,CAAC,MAAM,8BAA8B,GAAoC,CAAC;KAC7E,MAAM,CAAC;IACN,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAChE,oDAAoD,CACrD;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IAC1E,KAAK,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,wBAAwB,CAAC;IACpE,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,0BAA0B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACjE,sQAAsQ,CACvQ;IACD,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5C,kCAAkC,CACnC;IACD,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtE,iEAAiE,CAClE;IACD,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,yBAAyB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SACpE,QAAQ,CACP,8FAA8F,CAC/F;IACH,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACvB,qIAAqI,CACtI;IACD,MAAM,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5C,kCAAkC,CACnC;IACD,OAAO,EAAE,CAAC,CAAC,KAAK,CAAC,4BAA4B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SACjE,QAAQ,CACP,mFAAmF,CACpF;IACH,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,yCAAyC,CAC1C;IACD,MAAM,EAAE,oBAAoB;IAC5B,WAAW,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACjD,4BAA4B,CAC7B;IACD,YAAY,EAAE,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACxE,qDAAqD,CACtD;IACD,IAAI,EAAE,kCAAkC,CAAC,QAAQ,CAC/C,8HAA8H,CAC/H;IACD,gBAAgB,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACtD,4JAA4J,CAC7J;IACD,UAAU,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC7B,8CAA8C,CAC/C;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,iBAAiB,CAAC;CAC7C,CAAC,CAAC"}
|
|
@@ -4,6 +4,7 @@ import { TaskStatus } from "./taskstatus.js";
|
|
|
4
4
|
import { TestRunEvaluator } from "./testrunevaluator.js";
|
|
5
5
|
export type BenchmarkStatusResponse = {
|
|
6
6
|
task_id: string;
|
|
7
|
+
name: string;
|
|
7
8
|
status: TaskStatus;
|
|
8
9
|
test_uuids?: Array<string> | null | undefined;
|
|
9
10
|
evaluators?: Array<TestRunEvaluator> | null | undefined;
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"benchmarkstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAE/B,MAAM,MAAM,uBAAuB,GAAG;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,aAAa,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACtD,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACrE,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,KAAK,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC5B,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACzC,CAAC;AAEF,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,
|
|
1
|
+
{"version":3,"file":"benchmarkstatusresponse.d.ts","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,WAAW,EAAyB,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAE,UAAU,EAAwB,MAAM,iBAAiB,CAAC;AACnE,OAAO,EACL,gBAAgB,EAEjB,MAAM,uBAAuB,CAAC;AAE/B,MAAM,MAAM,uBAAuB,GAAG;IACpC,OAAO,EAAE,MAAM,CAAC;IAChB,IAAI,EAAE,MAAM,CAAC;IACb,MAAM,EAAE,UAAU,CAAC;IACnB,UAAU,CAAC,EAAE,KAAK,CAAC,MAAM,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IAC9C,UAAU,CAAC,EAAE,KAAK,CAAC,gBAAgB,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACxD,aAAa,CAAC,EAAE,KAAK,CAAC,WAAW,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACtD,mBAAmB,CAAC,EAAE,KAAK,CAAC;QAAE,CAAC,CAAC,EAAE,MAAM,GAAG,GAAG,CAAA;KAAE,CAAC,GAAG,IAAI,GAAG,SAAS,CAAC;IACrE,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,KAAK,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC5B,SAAS,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAChC,WAAW,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACzC,CAAC;AAEF,eAAO,MAAM,iCAAiC,EAAE,CAAC,CAAC,OAAO,CACvD,uBAAuB,CA+BvB,CAAC"}
|
|
@@ -14,6 +14,7 @@ export const BenchmarkStatusResponse$zodSchema = z.object({
|
|
|
14
14
|
leaderboard_summary: z.array(z.record(z.string(), z.any())).nullable()
|
|
15
15
|
.optional().describe("Leaderboard comparing the models, one row per model. Columns vary by benchmark: a `model` column plus pass/fail counts, latency, cost, and one score column per evaluator, keyed by evaluator name"),
|
|
16
16
|
model_results: z.array(ModelResult$zodSchema).nullable().optional().describe("Results for each model"),
|
|
17
|
+
name: z.string().describe("Name of the run. A run nobody has renamed shows its number instead, such as `Run 1` for a test run or `Benchmark 1` for a benchmark"),
|
|
17
18
|
share_token: z.string().nullable().optional().describe("Token for building the public share URL"),
|
|
18
19
|
status: TaskStatus$zodSchema,
|
|
19
20
|
task_id: z.string().describe("Benchmark run job ID"),
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"benchmarkstatusresponse.js","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;
|
|
1
|
+
{"version":3,"file":"benchmarkstatusresponse.js","sourceRoot":"","sources":["../../src/models/benchmarkstatusresponse.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAe,qBAAqB,EAAE,MAAM,kBAAkB,CAAC;AACtE,OAAO,EAAc,oBAAoB,EAAE,MAAM,iBAAiB,CAAC;AACnE,OAAO,EAEL,0BAA0B,GAC3B,MAAM,uBAAuB,CAAC;AAgB/B,MAAM,CAAC,MAAM,iCAAiC,GAE1C,CAAC,CAAC,MAAM,CAAC;IACX,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gLAAgL,CACjL;IACD,KAAK,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAAC,wBAAwB,CAAC;IACpE,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,0BAA0B,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;SAClE,QAAQ,CACP,4GAA4G,CAC7G;IACH,SAAS,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC5C,oCAAoC,CACrC;IACD,mBAAmB,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,CAAC,GAAG,EAAE,CAAC,CAAC,CAAC,QAAQ,EAAE;SACnE,QAAQ,EAAE,CAAC,QAAQ,CAClB,oMAAoM,CACrM;IACH,aAAa,EAAE,CAAC,CAAC,KAAK,CAAC,qBAAqB,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC1E,wBAAwB,CACzB;IACD,IAAI,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACvB,qIAAqI,CACtI;IACD,WAAW,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CACpD,yCAAyC,CAC1C;IACD,MAAM,EAAE,oBAAoB;IAC5B,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,sBAAsB,CAAC;IACpD,UAAU,EAAE,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,MAAM,EAAE,CAAC,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,CAC5D,wDAAwD,CACzD;CACF,CAAC,CAAC"}
|
|
@@ -0,0 +1,13 @@
|
|
|
1
|
+
import * as z from "zod";
|
|
2
|
+
import { HTTPValidationError } from "./httpvalidationerror.js";
|
|
3
|
+
import { TestCaseResult } from "./testcaseresult.js";
|
|
4
|
+
export type GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest = {
|
|
5
|
+
task_id: string;
|
|
6
|
+
test_uuid: string;
|
|
7
|
+
model?: string | null | undefined;
|
|
8
|
+
xAPIKey?: string | null | undefined;
|
|
9
|
+
};
|
|
10
|
+
export declare const GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest$zodSchema: z.ZodType<GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest>;
|
|
11
|
+
export type GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetResponse = TestCaseResult | HTTPValidationError;
|
|
12
|
+
export declare const GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetResponse$zodSchema: z.ZodType<GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetResponse>;
|
|
13
|
+
//# sourceMappingURL=getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.d.ts.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAAE,cAAc,EAA4B,MAAM,qBAAqB,CAAC;AAE/E,MAAM,MAAM,kEAAkE,GAC5E;IACE,OAAO,EAAE,MAAM,CAAC;IAChB,SAAS,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;IAClC,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEJ,eAAO,MAAM,4EAA4E,EACvF,CAAC,CAAC,OAAO,CACP,kEAAkE,CAUlE,CAAC;AAEL,MAAM,MAAM,mEAAmE,GAC3E,cAAc,GACd,mBAAmB,CAAC;AAExB,eAAO,MAAM,6EAA6E,EACxF,CAAC,CAAC,OAAO,CACP,mEAAmE,CAInE,CAAC"}
|
|
@@ -0,0 +1,17 @@
|
|
|
1
|
+
/*
|
|
2
|
+
* Code generated by Speakeasy (https://speakeasy.com). DO NOT EDIT.
|
|
3
|
+
*/
|
|
4
|
+
import * as z from "zod";
|
|
5
|
+
import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
|
|
6
|
+
import { TestCaseResult$zodSchema } from "./testcaseresult.js";
|
|
7
|
+
export const GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetRequest$zodSchema = z.object({
|
|
8
|
+
model: z.string().describe("Which model's answer to read. Required for a benchmark, which runs every test once per model").nullable().optional(),
|
|
9
|
+
task_id: z.string().describe("Test run or benchmark the case was run in"),
|
|
10
|
+
test_uuid: z.string().describe("The test whose result to read, as `test_uuid` on the case"),
|
|
11
|
+
xAPIKey: z.string().nullable().optional(),
|
|
12
|
+
});
|
|
13
|
+
export const GetAgentTestCaseResultAgentTestsRunTaskIdResultsTestUuidGetResponse$zodSchema = z.union([
|
|
14
|
+
TestCaseResult$zodSchema,
|
|
15
|
+
HTTPValidationError$zodSchema,
|
|
16
|
+
]);
|
|
17
|
+
//# sourceMappingURL=getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js.map
|
|
@@ -0,0 +1 @@
|
|
|
1
|
+
{"version":3,"file":"getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.js","sourceRoot":"","sources":["../../src/models/getagenttestcaseresultagenttestsruntaskidresultstestuuidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAAkB,wBAAwB,EAAE,MAAM,qBAAqB,CAAC;AAU/E,MAAM,CAAC,MAAM,4EAA4E,GAGnF,CAAC,CAAC,MAAM,CAAC;IACX,KAAK,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CACxB,8FAA8F,CAC/F,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;IACvB,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,2CAA2C,CAAC;IACzE,SAAS,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC5B,2DAA2D,CAC5D;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,6EAA6E,GAGpF,CAAC,CAAC,KAAK,CAAC;IACV,wBAAwB;IACxB,6BAA6B;CAC9B,CAAC,CAAC"}
|
|
@@ -1,9 +1,26 @@
|
|
|
1
1
|
import * as z from "zod";
|
|
2
|
+
import { ClosedEnum } from "../types/enums.js";
|
|
2
3
|
import { HTTPValidationError } from "./httpvalidationerror.js";
|
|
3
4
|
import { TestRunStatusResponse } from "./testrunstatusresponse.js";
|
|
5
|
+
/**
|
|
6
|
+
* How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`
|
|
7
|
+
*/
|
|
8
|
+
export declare const GetAgentTestRunStatusAgentTestsRunTaskIdGetMode: {
|
|
9
|
+
readonly Full: "full";
|
|
10
|
+
readonly Summary: "summary";
|
|
11
|
+
};
|
|
12
|
+
/**
|
|
13
|
+
* How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`
|
|
14
|
+
*/
|
|
15
|
+
export type GetAgentTestRunStatusAgentTestsRunTaskIdGetMode = ClosedEnum<typeof GetAgentTestRunStatusAgentTestsRunTaskIdGetMode>;
|
|
16
|
+
export declare const GetAgentTestRunStatusAgentTestsRunTaskIdGetMode$zodSchema: z.ZodEnum<{
|
|
17
|
+
full: "full";
|
|
18
|
+
summary: "summary";
|
|
19
|
+
}>;
|
|
4
20
|
export type GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest = {
|
|
5
21
|
task_id: string;
|
|
6
22
|
only_failed?: boolean | undefined;
|
|
23
|
+
mode?: GetAgentTestRunStatusAgentTestsRunTaskIdGetMode | undefined;
|
|
7
24
|
compact?: boolean | undefined;
|
|
8
25
|
xAPIKey?: string | null | undefined;
|
|
9
26
|
};
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,qBAAqB,EAEtB,MAAM,4BAA4B,CAAC;AAEpC,MAAM,MAAM,kDAAkD,GAAG;IAC/D,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,4DAA4D,EACvE,CAAC,CAAC,OAAO,CAAC,kDAAkD,
|
|
1
|
+
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAClC,OAAO,EACL,qBAAqB,EAEtB,MAAM,4BAA4B,CAAC;AAEpC;;GAEG;AACH,eAAO,MAAM,+CAA+C;;;CAGlD,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,+CAA+C,GAAG,UAAU,CACtE,OAAO,+CAA+C,CACvD,CAAC;AAEF,eAAO,MAAM,yDAAyD;;;EAOrE,CAAC;AAEF,MAAM,MAAM,kDAAkD,GAAG;IAC/D,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,IAAI,CAAC,EAAE,+CAA+C,GAAG,SAAS,CAAC;IACnE,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,4DAA4D,EACvE,CAAC,CAAC,OAAO,CAAC,kDAAkD,CAc1D,CAAC;AAEL,MAAM,MAAM,mDAAmD,GAC3D,qBAAqB,GACrB,mBAAmB,CAAC;AAExB,eAAO,MAAM,6DAA6D,EACxE,CAAC,CAAC,OAAO,CAAC,mDAAmD,CAG3D,CAAC"}
|
|
@@ -4,8 +4,20 @@
|
|
|
4
4
|
import * as z from "zod";
|
|
5
5
|
import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
|
|
6
6
|
import { TestRunStatusResponse$zodSchema, } from "./testrunstatusresponse.js";
|
|
7
|
+
/**
|
|
8
|
+
* How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`
|
|
9
|
+
*/
|
|
10
|
+
export const GetAgentTestRunStatusAgentTestsRunTaskIdGetMode = {
|
|
11
|
+
Full: "full",
|
|
12
|
+
Summary: "summary",
|
|
13
|
+
};
|
|
14
|
+
export const GetAgentTestRunStatusAgentTestsRunTaskIdGetMode$zodSchema = z.enum([
|
|
15
|
+
"full",
|
|
16
|
+
"summary",
|
|
17
|
+
]).describe("How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`");
|
|
7
18
|
export const GetAgentTestRunStatusAgentTestsRunTaskIdGetRequest$zodSchema = z.object({
|
|
8
19
|
compact: z.boolean().default(false).describe("Return a compact response that omits heavy detail fields (`results.output`, `results.test_case`, `results.judge_results`, `results.reasoning`, `evaluators.output_config`), keeping only the lightweight decision fields. Omit for full detail"),
|
|
20
|
+
mode: GetAgentTestRunStatusAgentTestsRunTaskIdGetMode$zodSchema.default("full").describe("How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`"),
|
|
9
21
|
only_failed: z.boolean().default(false).describe("Return only failing test cases. Omit to return every case"),
|
|
10
22
|
task_id: z.string().describe("Test run to poll for status and results"),
|
|
11
23
|
xAPIKey: z.string().nullable().optional(),
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.js","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;
|
|
1
|
+
{"version":3,"file":"getagenttestrunstatusagenttestsruntaskidgetop.js","sourceRoot":"","sources":["../../src/models/getagenttestrunstatusagenttestsruntaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAClC,OAAO,EAEL,+BAA+B,GAChC,MAAM,4BAA4B,CAAC;AAEpC;;GAEG;AACH,MAAM,CAAC,MAAM,+CAA+C,GAAG;IAC7D,IAAI,EAAE,MAAM;IACZ,OAAO,EAAE,SAAS;CACV,CAAC;AAQX,MAAM,CAAC,MAAM,yDAAyD,GAAG,CAAC,CAAC,IAAI,CAC7E;IACE,MAAM;IACN,SAAS;CACV,CACF,CAAC,QAAQ,CACR,6UAA6U,CAC9U,CAAC;AAUF,MAAM,CAAC,MAAM,4DAA4D,GACP,CAAC,CAAC,MAAM,CAAC;IACvE,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,gPAAgP,CACjP;IACD,IAAI,EAAE,yDAAyD,CAAC,OAAO,CACrE,MAAM,CACP,CAAC,QAAQ,CACR,6UAA6U,CAC9U;IACD,WAAW,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC9C,2DAA2D,CAC5D;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAAC,yCAAyC,CAAC;IACvE,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,6DAA6D,GACP,CAAC,CAAC,KAAK,CAAC;IACvE,+BAA+B;IAC/B,6BAA6B;CAC9B,CAAC,CAAC"}
|
|
@@ -1,9 +1,26 @@
|
|
|
1
1
|
import * as z from "zod";
|
|
2
|
+
import { ClosedEnum } from "../types/enums.js";
|
|
2
3
|
import { BenchmarkStatusResponse } from "./benchmarkstatusresponse.js";
|
|
3
4
|
import { HTTPValidationError } from "./httpvalidationerror.js";
|
|
5
|
+
/**
|
|
6
|
+
* How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`
|
|
7
|
+
*/
|
|
8
|
+
export declare const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode: {
|
|
9
|
+
readonly Full: "full";
|
|
10
|
+
readonly Summary: "summary";
|
|
11
|
+
};
|
|
12
|
+
/**
|
|
13
|
+
* How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`
|
|
14
|
+
*/
|
|
15
|
+
export type GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode = ClosedEnum<typeof GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode>;
|
|
16
|
+
export declare const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode$zodSchema: z.ZodEnum<{
|
|
17
|
+
full: "full";
|
|
18
|
+
summary: "summary";
|
|
19
|
+
}>;
|
|
4
20
|
export type GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest = {
|
|
5
21
|
task_id: string;
|
|
6
22
|
only_failed?: boolean | undefined;
|
|
23
|
+
mode?: GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode | undefined;
|
|
7
24
|
compact?: boolean | undefined;
|
|
8
25
|
xAPIKey?: string | null | undefined;
|
|
9
26
|
};
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EACL,uBAAuB,EAExB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC,MAAM,MAAM,qDAAqD,GAAG;IAClE,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,+DAA+D,EAC1E,CAAC,CAAC,OAAO,CAAC,qDAAqD,
|
|
1
|
+
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.d.ts","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAIA,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AACzB,OAAO,EAAE,UAAU,EAAE,MAAM,mBAAmB,CAAC;AAC/C,OAAO,EACL,uBAAuB,EAExB,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,mBAAmB,EAEpB,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,eAAO,MAAM,kDAAkD;;;CAGrD,CAAC;AACX;;GAEG;AACH,MAAM,MAAM,kDAAkD,GAAG,UAAU,CACzE,OAAO,kDAAkD,CAC1D,CAAC;AAEF,eAAO,MAAM,4DAA4D;;;EAMtE,CAAC;AAEJ,MAAM,MAAM,qDAAqD,GAAG;IAClE,OAAO,EAAE,MAAM,CAAC;IAChB,WAAW,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAClC,IAAI,CAAC,EAAE,kDAAkD,GAAG,SAAS,CAAC;IACtE,OAAO,CAAC,EAAE,OAAO,GAAG,SAAS,CAAC;IAC9B,OAAO,CAAC,EAAE,MAAM,GAAG,IAAI,GAAG,SAAS,CAAC;CACrC,CAAC;AAEF,eAAO,MAAM,+DAA+D,EAC1E,CAAC,CAAC,OAAO,CAAC,qDAAqD,CAgB7D,CAAC;AAEL,MAAM,MAAM,sDAAsD,GAC9D,uBAAuB,GACvB,mBAAmB,CAAC;AAExB,eAAO,MAAM,gEAAgE,EAC3E,CAAC,CAAC,OAAO,CAAC,sDAAsD,CAG9D,CAAC"}
|
|
@@ -4,8 +4,21 @@
|
|
|
4
4
|
import * as z from "zod";
|
|
5
5
|
import { BenchmarkStatusResponse$zodSchema, } from "./benchmarkstatusresponse.js";
|
|
6
6
|
import { HTTPValidationError$zodSchema, } from "./httpvalidationerror.js";
|
|
7
|
+
/**
|
|
8
|
+
* How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`
|
|
9
|
+
*/
|
|
10
|
+
export const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode = {
|
|
11
|
+
Full: "full",
|
|
12
|
+
Summary: "summary",
|
|
13
|
+
};
|
|
14
|
+
export const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode$zodSchema = z
|
|
15
|
+
.enum([
|
|
16
|
+
"full",
|
|
17
|
+
"summary",
|
|
18
|
+
]).describe("How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`");
|
|
7
19
|
export const GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetRequest$zodSchema = z.object({
|
|
8
20
|
compact: z.boolean().default(false).describe("Return a compact response that omits heavy detail fields (`model_results.test_results`, `evaluators.output_config`), keeping only the lightweight decision fields. Omit for full detail"),
|
|
21
|
+
mode: GetBenchmarkStatusAgentTestsBenchmarkTaskIdGetMode$zodSchema.default("full").describe("How much of each test case to return. `full` returns every field of every case. `summary` returns one light row per case, with its ID, name, verdict and short reason, leaving out the conversation, the agent's output and the evaluator verdicts. Read those one case at a time from `GET /agent-tests/run/{task_id}/results/{test_uuid}`"),
|
|
9
22
|
only_failed: z.boolean().default(false).describe("Return only failing test cases for each model. Omit to return every case"),
|
|
10
23
|
task_id: z.string().describe("Benchmark run to poll for status and results"),
|
|
11
24
|
xAPIKey: z.string().nullable().optional(),
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.js","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;
|
|
1
|
+
{"version":3,"file":"getbenchmarkstatusagenttestsbenchmarktaskidgetop.js","sourceRoot":"","sources":["../../src/models/getbenchmarkstatusagenttestsbenchmarktaskidgetop.ts"],"names":[],"mappings":"AAAA;;GAEG;AAEH,OAAO,KAAK,CAAC,MAAM,KAAK,CAAC;AAEzB,OAAO,EAEL,iCAAiC,GAClC,MAAM,8BAA8B,CAAC;AACtC,OAAO,EAEL,6BAA6B,GAC9B,MAAM,0BAA0B,CAAC;AAElC;;GAEG;AACH,MAAM,CAAC,MAAM,kDAAkD,GAAG;IAChE,IAAI,EAAE,MAAM;IACZ,OAAO,EAAE,SAAS;CACV,CAAC;AAQX,MAAM,CAAC,MAAM,4DAA4D,GAAG,CAAC;KAC1E,IAAI,CAAC;IACJ,MAAM;IACN,SAAS;CACV,CAAC,CAAC,QAAQ,CACT,6UAA6U,CAC9U,CAAC;AAUJ,MAAM,CAAC,MAAM,+DAA+D,GACP,CAAC,CAAC,MAAM,CAAC;IAC1E,OAAO,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC1C,yLAAyL,CAC1L;IACD,IAAI,EAAE,4DAA4D,CAAC,OAAO,CACxE,MAAM,CACP,CAAC,QAAQ,CACR,6UAA6U,CAC9U;IACD,WAAW,EAAE,CAAC,CAAC,OAAO,EAAE,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,QAAQ,CAC9C,0EAA0E,CAC3E;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,CAC1B,8CAA8C,CAC/C;IACD,OAAO,EAAE,CAAC,CAAC,MAAM,EAAE,CAAC,QAAQ,EAAE,CAAC,QAAQ,EAAE;CAC1C,CAAC,CAAC;AAML,MAAM,CAAC,MAAM,gEAAgE,GACP,CAAC,CAAC,KAAK,CAAC;IAC1E,iCAAiC;IACjC,6BAA6B;CAC9B,CAAC,CAAC"}
|