@skybridge/test 0.0.0-dev.50c10e1 → 0.0.0-dev.df0cfdd
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/dist/chat.d.ts +4 -2
- package/dist/chat.d.ts.map +1 -1
- package/dist/chat.js +27 -6
- package/dist/chat.js.map +1 -1
- package/dist/global-setup.d.ts.map +1 -1
- package/dist/global-setup.js +4 -2
- package/dist/global-setup.js.map +1 -1
- package/dist/index.d.ts +0 -1
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js.map +1 -1
- package/dist/matchers.js +3 -3
- package/dist/matchers.js.map +1 -1
- package/dist/model.d.ts +1 -1
- package/dist/model.d.ts.map +1 -1
- package/dist/model.js +8 -12
- package/dist/model.js.map +1 -1
- package/dist/server-harness.d.ts +1 -2
- package/dist/server-harness.d.ts.map +1 -1
- package/dist/server-harness.js +3 -2
- package/dist/server-harness.js.map +1 -1
- package/dist/session-registry.d.ts +3 -3
- package/dist/session-registry.d.ts.map +1 -1
- package/dist/session-registry.js +6 -9
- package/dist/session-registry.js.map +1 -1
- package/dist/setup.d.ts.map +1 -1
- package/dist/setup.js +0 -5
- package/dist/setup.js.map +1 -1
- package/package.json +4 -4
package/dist/chat.d.ts
CHANGED
|
@@ -8,7 +8,7 @@ export type ToolCall<App> = {
|
|
|
8
8
|
[Name in ToolNames<App>]: {
|
|
9
9
|
name: Name;
|
|
10
10
|
arguments: ToolInput<App, Name>;
|
|
11
|
-
|
|
11
|
+
failed?: string;
|
|
12
12
|
};
|
|
13
13
|
}[ToolNames<App>];
|
|
14
14
|
/**
|
|
@@ -16,7 +16,7 @@ export type ToolCall<App> = {
|
|
|
16
16
|
* provider and its credentials are chosen by the project rather than by a
|
|
17
17
|
* base URL the runner has to know about.
|
|
18
18
|
*/
|
|
19
|
-
|
|
19
|
+
interface HostConfig {
|
|
20
20
|
model: LanguageModel;
|
|
21
21
|
temperature?: number;
|
|
22
22
|
systemPrompt?: string;
|
|
@@ -35,6 +35,7 @@ export declare class Chat<App = unknown> {
|
|
|
35
35
|
/** Every call the model made, in order, as it crossed the wire. */
|
|
36
36
|
readonly toolCalls: ToolCall<App>[];
|
|
37
37
|
private readonly messages;
|
|
38
|
+
private readonly failures;
|
|
38
39
|
private readonly client;
|
|
39
40
|
private readonly transport;
|
|
40
41
|
private readonly host;
|
|
@@ -46,6 +47,7 @@ export declare class Chat<App = unknown> {
|
|
|
46
47
|
/** Takes a turn and returns the model's answer. */
|
|
47
48
|
send(prompt: string): Promise<string>;
|
|
48
49
|
close(): Promise<void>;
|
|
50
|
+
private failureFor;
|
|
49
51
|
private toolSet;
|
|
50
52
|
}
|
|
51
53
|
export {};
|
package/dist/chat.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAEA,OAAO,EAIL,KAAK,aAAa,EAGnB,MAAM,IAAI,CAAC;AACZ,OAAO,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAE7D;;;GAGG;AACH,MAAM,MAAM,QAAQ,CAAC,GAAG,IAAI;KACzB,IAAI,IAAI,SAAS,CAAC,GAAG,CAAC,GAAG;QACxB,IAAI,EAAE,IAAI,CAAC;QACX,SAAS,EAAE,SAAS,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAChC,
|
|
1
|
+
{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAEA,OAAO,EAIL,KAAK,aAAa,EAGnB,MAAM,IAAI,CAAC;AACZ,OAAO,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAE7D;;;GAGG;AACH,MAAM,MAAM,QAAQ,CAAC,GAAG,IAAI;KACzB,IAAI,IAAI,SAAS,CAAC,GAAG,CAAC,GAAG;QACxB,IAAI,EAAE,IAAI,CAAC;QACX,SAAS,EAAE,SAAS,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC;QAChC,MAAM,CAAC,EAAE,MAAM,CAAC;KACjB;CACF,CAAC,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC;AAElB;;;;GAIG;AACH,UAAU,UAAU;IAClB,KAAK,EAAE,aAAa,CAAC;IACrB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAED,UAAU,iBAAiB;IACzB,IAAI,EAAE,MAAM,CAAC;IACb,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACtC;AAID;;;GAGG;AACH,qBAAa,IAAI,CAAC,GAAG,GAAG,OAAO;IAC7B,mEAAmE;IACnE,QAAQ,CAAC,SAAS,EAAE,QAAQ,CAAC,GAAG,CAAC,EAAE,CAAM;IAEzC,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAAsB;IAC/C,OAAO,CAAC,QAAQ,CAAC,QAAQ,CAA6B;IACtD,OAAO,CAAC,QAAQ,CAAC,MAAM,CAAS;IAChC,OAAO,CAAC,QAAQ,CAAC,SAAS,CAAgC;IAC1D,OAAO,CAAC,QAAQ,CAAC,IAAI,CAAa;IAClC,OAAO,CAAC,KAAK,CAA2B;IAExC,OAAO;WAUM,IAAI,CAAC,GAAG,EACnB,SAAS,EAAE,MAAM,EACjB,IAAI,EAAE,UAAU,GACf,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IAerB,0EAA0E;IAC1E,IAAI,eAAe,IAAI,iBAAiB,EAAE,CAEzC;IAED,mDAAmD;IAC7C,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC;IA4BrC,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IAI5B,OAAO,CAAC,UAAU;IAalB,OAAO,CAAC,OAAO;CAmChB"}
|
package/dist/chat.js
CHANGED
|
@@ -10,6 +10,7 @@ export class Chat {
|
|
|
10
10
|
/** Every call the model made, in order, as it crossed the wire. */
|
|
11
11
|
toolCalls = [];
|
|
12
12
|
messages = [];
|
|
13
|
+
failures = new Map();
|
|
13
14
|
client;
|
|
14
15
|
transport;
|
|
15
16
|
host;
|
|
@@ -56,7 +57,7 @@ export class Chat {
|
|
|
56
57
|
this.toolCalls.push({
|
|
57
58
|
name: call.toolName,
|
|
58
59
|
arguments: (call.input ?? {}),
|
|
59
|
-
...(call
|
|
60
|
+
...this.failureFor(call),
|
|
60
61
|
});
|
|
61
62
|
}
|
|
62
63
|
}
|
|
@@ -66,6 +67,16 @@ export class Chat {
|
|
|
66
67
|
async close() {
|
|
67
68
|
await this.transport.close();
|
|
68
69
|
}
|
|
70
|
+
failureFor(call) {
|
|
71
|
+
const reported = this.failures.get(call.toolCallId);
|
|
72
|
+
if (reported !== undefined) {
|
|
73
|
+
return { failed: reported };
|
|
74
|
+
}
|
|
75
|
+
if (call.invalid === true) {
|
|
76
|
+
return { failed: "the model's arguments could not be parsed" };
|
|
77
|
+
}
|
|
78
|
+
return {};
|
|
79
|
+
}
|
|
69
80
|
toolSet() {
|
|
70
81
|
const entries = this.tools.map((definition) => {
|
|
71
82
|
const { $schema: _ignored, ...parameters } = definition.inputSchema;
|
|
@@ -74,11 +85,21 @@ export class Chat {
|
|
|
74
85
|
dynamicTool({
|
|
75
86
|
description: definition.description ?? "",
|
|
76
87
|
inputSchema: jsonSchema(parameters),
|
|
77
|
-
execute: async (input) => {
|
|
78
|
-
|
|
79
|
-
|
|
80
|
-
|
|
81
|
-
|
|
88
|
+
execute: async (input, { toolCallId }) => {
|
|
89
|
+
let result;
|
|
90
|
+
try {
|
|
91
|
+
result = await this.client.callTool({
|
|
92
|
+
name: definition.name,
|
|
93
|
+
arguments: (input ?? {}),
|
|
94
|
+
});
|
|
95
|
+
}
|
|
96
|
+
catch (error) {
|
|
97
|
+
this.failures.set(toolCallId, `the call never reached the server: ${error instanceof Error ? error.message : String(error)}`);
|
|
98
|
+
throw error;
|
|
99
|
+
}
|
|
100
|
+
if (result.isError === true) {
|
|
101
|
+
this.failures.set(toolCallId, `the tool returned an error: ${JSON.stringify(result.content)}`);
|
|
102
|
+
}
|
|
82
103
|
return result.content ?? result;
|
|
83
104
|
},
|
|
84
105
|
}),
|
package/dist/chat.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"chat.js","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,6BAA6B,EAAE,MAAM,oDAAoD,CAAC;AACnG,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EAGV,WAAW,GACZ,MAAM,IAAI,CAAC;AAiCZ,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAE5B;;;GAGG;AACH,MAAM,OAAO,IAAI;IACf,mEAAmE;IAC1D,SAAS,GAAoB,EAAE,CAAC;IAExB,QAAQ,GAAmB,EAAE,CAAC;IAC9B,MAAM,CAAS;IACf,SAAS,CAAgC;IACzC,IAAI,CAAa;IAC1B,KAAK,GAAwB,EAAE,CAAC;IAExC,YACE,MAAc,EACd,SAAwC,EACxC,IAAgB;QAEhB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,CAAC;IAED,MAAM,CAAC,KAAK,CAAC,IAAI,CACf,SAAiB,EACjB,IAAgB;QAEhB,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;QACpE,MAAM,SAAS,GAAG,IAAI,6BAA6B,CAAC,IAAI,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC;QACxE,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,IAAI,CAAM,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;QACpD,IAAI,CAAC;YACH,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,EAAE,CAAC;YAC3C,IAAI,CAAC,KAAK,GAAG,KAAuC,CAAC;QACvD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,SAAS,CAAC,KAAK,EAAE,CAAC;YACxB,MAAM,KAAK,CAAC;QACd,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,0EAA0E;IAC1E,IAAI,eAAe;QACjB,OAAO,IAAI,CAAC,KAAK,CAAC;IACpB,CAAC;IAED,mDAAmD;IACnD,KAAK,CAAC,IAAI,CAAC,MAAc;QACvB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAEtD,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC;YAChC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK;YACtB,WAAW,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC;YACvC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,KAAK,SAAS;gBACtC,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,IAAI,CAAC,OAAO,EAAE;YACrB,QAAQ,EAAE,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,iBAAiB,CAAC;SAC/D,CAAC,CAAC;QAEH,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;YAChC,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;gBAClC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;oBAClB,IAAI,EAAE,IAAI,CAAC,QAAQ;oBACnB,SAAS,EAAE,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,CAA4B;oBACxD,GAAG,
|
|
1
|
+
{"version":3,"file":"chat.js","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AACnE,OAAO,EAAE,6BAA6B,EAAE,MAAM,oDAAoD,CAAC;AACnG,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EAGV,WAAW,GACZ,MAAM,IAAI,CAAC;AAiCZ,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAE5B;;;GAGG;AACH,MAAM,OAAO,IAAI;IACf,mEAAmE;IAC1D,SAAS,GAAoB,EAAE,CAAC;IAExB,QAAQ,GAAmB,EAAE,CAAC;IAC9B,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IACrC,MAAM,CAAS;IACf,SAAS,CAAgC;IACzC,IAAI,CAAa;IAC1B,KAAK,GAAwB,EAAE,CAAC;IAExC,YACE,MAAc,EACd,SAAwC,EACxC,IAAgB;QAEhB,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;IACnB,CAAC;IAED,MAAM,CAAC,KAAK,CAAC,IAAI,CACf,SAAiB,EACjB,IAAgB;QAEhB,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;QACpE,MAAM,SAAS,GAAG,IAAI,6BAA6B,CAAC,IAAI,GAAG,CAAC,SAAS,CAAC,CAAC,CAAC;QACxE,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,IAAI,CAAM,MAAM,EAAE,SAAS,EAAE,IAAI,CAAC,CAAC;QACpD,IAAI,CAAC;YACH,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,EAAE,CAAC;YAC3C,IAAI,CAAC,KAAK,GAAG,KAAuC,CAAC;QACvD,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,SAAS,CAAC,KAAK,EAAE,CAAC;YACxB,MAAM,KAAK,CAAC;QACd,CAAC;QACD,OAAO,IAAI,CAAC;IACd,CAAC;IAED,0EAA0E;IAC1E,IAAI,eAAe;QACjB,OAAO,IAAI,CAAC,KAAK,CAAC;IACpB,CAAC;IAED,mDAAmD;IACnD,KAAK,CAAC,IAAI,CAAC,MAAc;QACvB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QAEtD,MAAM,MAAM,GAAG,MAAM,YAAY,CAAC;YAChC,KAAK,EAAE,IAAI,CAAC,IAAI,CAAC,KAAK;YACtB,WAAW,EAAE,IAAI,CAAC,IAAI,CAAC,WAAW,IAAI,CAAC;YACvC,GAAG,CAAC,IAAI,CAAC,IAAI,CAAC,YAAY,KAAK,SAAS;gBACtC,CAAC,CAAC,EAAE;gBACJ,CAAC,CAAC,EAAE,MAAM,EAAE,IAAI,CAAC,IAAI,CAAC,YAAY,EAAE,CAAC;YACvC,QAAQ,EAAE,IAAI,CAAC,QAAQ;YACvB,KAAK,EAAE,IAAI,CAAC,OAAO,EAAE;YACrB,QAAQ,EAAE,WAAW,CAAC,IAAI,CAAC,IAAI,CAAC,QAAQ,IAAI,iBAAiB,CAAC;SAC/D,CAAC,CAAC;QAEH,KAAK,MAAM,IAAI,IAAI,MAAM,CAAC,KAAK,EAAE,CAAC;YAChC,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;gBAClC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC;oBAClB,IAAI,EAAE,IAAI,CAAC,QAAQ;oBACnB,SAAS,EAAE,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,CAA4B;oBACxD,GAAG,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;iBACR,CAAC,CAAC;YACtB,CAAC;QACH,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;QAChD,OAAO,MAAM,CAAC,IAAI,CAAC;IACrB,CAAC;IAED,KAAK,CAAC,KAAK;QACT,MAAM,IAAI,CAAC,SAAS,CAAC,KAAK,EAAE,CAAC;IAC/B,CAAC;IAEO,UAAU,CAAC,IAA+C;QAGhE,MAAM,QAAQ,GAAG,IAAI,CAAC,QAAQ,CAAC,GAAG,CAAC,IAAI,CAAC,UAAU,CAAC,CAAC;QACpD,IAAI,QAAQ,KAAK,SAAS,EAAE,CAAC;YAC3B,OAAO,EAAE,MAAM,EAAE,QAAQ,EAAE,CAAC;QAC9B,CAAC;QACD,IAAI,IAAI,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;YAC1B,OAAO,EAAE,MAAM,EAAE,2CAA2C,EAAE,CAAC;QACjE,CAAC;QACD,OAAO,EAAE,CAAC;IACZ,CAAC;IAEO,OAAO;QACb,MAAM,OAAO,GAAG,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,CAAC,UAAU,EAAE,EAAE;YAC5C,MAAM,EAAE,OAAO,EAAE,QAAQ,EAAE,GAAG,UAAU,EAAE,GAAG,UAAU,CAAC,WAAW,CAAC;YACpE,OAAO;gBACL,UAAU,CAAC,IAAI;gBACf,WAAW,CAAC;oBACV,WAAW,EAAE,UAAU,CAAC,WAAW,IAAI,EAAE;oBACzC,WAAW,EAAE,UAAU,CAAC,UAAU,CAAC;oBACnC,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE;wBACvC,IAAI,MAA+C,CAAC;wBACpD,IAAI,CAAC;4BACH,MAAM,GAAG,MAAM,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC;gCAClC,IAAI,EAAE,UAAU,CAAC,IAAI;gCACrB,SAAS,EAAE,CAAC,KAAK,IAAI,EAAE,CAA4B;6BACpD,CAAC,CAAC;wBACL,CAAC;wBAAC,OAAO,KAAK,EAAE,CAAC;4BACf,IAAI,CAAC,QAAQ,CAAC,GAAG,CACf,UAAU,EACV,sCAAsC,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAC/F,CAAC;4BACF,MAAM,KAAK,CAAC;wBACd,CAAC;wBACD,IAAI,MAAM,CAAC,OAAO,KAAK,IAAI,EAAE,CAAC;4BAC5B,IAAI,CAAC,QAAQ,CAAC,GAAG,CACf,UAAU,EACV,+BAA+B,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,OAAO,CAAC,EAAE,CAChE,CAAC;wBACJ,CAAC;wBACD,OAAO,MAAM,CAAC,OAAO,IAAI,MAAM,CAAC;oBAClC,CAAC;iBACF,CAAC;aACM,CAAC;QACb,CAAC,CAAC,CAAC;QACH,OAAO,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;IACrC,CAAC;CACF"}
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"global-setup.d.ts","sourceRoot":"","sources":["../src/global-setup.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAI/C,wBAA8B,KAAK,CAAC,OAAO,EAAE,WAAW,
|
|
1
|
+
{"version":3,"file":"global-setup.d.ts","sourceRoot":"","sources":["../src/global-setup.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,WAAW,EAAE,MAAM,aAAa,CAAC;AAI/C,wBAA8B,KAAK,CAAC,OAAO,EAAE,WAAW,8CA4BvD"}
|
package/dist/global-setup.js
CHANGED
|
@@ -1,7 +1,9 @@
|
|
|
1
1
|
import { startServer } from "./server-harness.js";
|
|
2
2
|
export default async function setup(project) {
|
|
3
|
-
const options = project.config.provide
|
|
4
|
-
|
|
3
|
+
const options = project.config.provide?.skybridgeEvals;
|
|
4
|
+
if (options === undefined) {
|
|
5
|
+
throw new Error("The evals plugin is not registered. Add `skybridge({ evals: {...} })` to the vitest config.");
|
|
6
|
+
}
|
|
5
7
|
if (options.server !== undefined) {
|
|
6
8
|
project.provide("skybridgeEvalsUrl", options.server);
|
|
7
9
|
return;
|
package/dist/global-setup.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"global-setup.js","sourceRoot":"","sources":["../src/global-setup.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AAElD,MAAM,CAAC,OAAO,CAAC,KAAK,UAAU,KAAK,CAAC,OAAoB;IACtD,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,OAAO;
|
|
1
|
+
{"version":3,"file":"global-setup.js","sourceRoot":"","sources":["../src/global-setup.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,WAAW,EAAE,MAAM,qBAAqB,CAAC;AAElD,MAAM,CAAC,OAAO,CAAC,KAAK,UAAU,KAAK,CAAC,OAAoB;IACtD,MAAM,OAAO,GAAG,OAAO,CAAC,MAAM,CAAC,OAAO,EAAE,cAE3B,CAAC;IACd,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;QAC1B,MAAM,IAAI,KAAK,CACb,6FAA6F,CAC9F,CAAC;IACJ,CAAC;IAED,IAAI,OAAO,CAAC,MAAM,KAAK,SAAS,EAAE,CAAC;QACjC,OAAO,CAAC,OAAO,CAAC,mBAAmB,EAAE,OAAO,CAAC,MAAM,CAAC,CAAC;QACrD,OAAO;IACT,CAAC;IACD,IAAI,OAAO,CAAC,OAAO,KAAK,SAAS,EAAE,CAAC;QAClC,MAAM,IAAI,KAAK,CAAC,qDAAqD,CAAC,CAAC;IACzE,CAAC;IAED,MAAM,MAAM,GAAG,MAAM,WAAW,CAAC,OAAO,CAAC,OAAO,CAAC,CAAC;IAClD,OAAO,CAAC,OAAO,CAAC,mBAAmB,EAAE,MAAM,CAAC,GAAG,CAAC,CAAC;IAEjD,OAAO,KAAK,IAAI,EAAE;QAChB,MAAM,MAAM,GAAG,MAAM,CAAC,MAAM,EAAE,CAAC,IAAI,EAAE,CAAC;QACtC,MAAM,MAAM,CAAC,IAAI,EAAE,CAAC;QACpB,IAAI,MAAM,KAAK,EAAE,EAAE,CAAC;YAClB,OAAO,CAAC,GAAG,CAAC,4BAA4B,MAAM,OAAO,CAAC,CAAC;QACzD,CAAC;IACH,CAAC,CAAC;AACJ,CAAC"}
|
package/dist/index.d.ts
CHANGED
package/dist/index.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACtC,YAAY,EAAE,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAAA,YAAY,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACtC,YAAY,EAAE,aAAa,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AACjE,OAAO,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAC7C,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC"}
|
package/dist/index.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"index.js","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AAEA,OAAO,EAAE,eAAe,EAAE,MAAM,YAAY,CAAC;AAC7C,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC"}
|
package/dist/matchers.js
CHANGED
|
@@ -3,14 +3,14 @@ function attemptedCalls(chat, name) {
|
|
|
3
3
|
return chat.toolCalls.filter((call) => String(call.name) === name);
|
|
4
4
|
}
|
|
5
5
|
function acceptedCalls(chat, name) {
|
|
6
|
-
return attemptedCalls(chat, name).filter((call) => call.
|
|
6
|
+
return attemptedCalls(chat, name).filter((call) => call.failed === undefined);
|
|
7
7
|
}
|
|
8
8
|
function observed(chat) {
|
|
9
9
|
if (chat.toolCalls.length === 0) {
|
|
10
10
|
return " (no tool was called)";
|
|
11
11
|
}
|
|
12
12
|
return chat.toolCalls
|
|
13
|
-
.map((call, index) => ` ${index + 1}. ${
|
|
13
|
+
.map((call, index) => ` ${index + 1}. ${call.name} ${JSON.stringify(call.arguments)}${call.failed === undefined ? "" : ` (failed: ${call.failed})`}`)
|
|
14
14
|
.join("\n");
|
|
15
15
|
}
|
|
16
16
|
function definitions(chat) {
|
|
@@ -40,7 +40,7 @@ expect.extend({
|
|
|
40
40
|
? `Expected "${name}" not to be called exactly once.`
|
|
41
41
|
: matching.length === 1
|
|
42
42
|
? `Expected "${name}" to be called once with ${JSON.stringify(args)}, but it was called with ${JSON.stringify(matching[0]?.arguments)}.`
|
|
43
|
-
: `Expected "${name}" to be called exactly once by a call
|
|
43
|
+
: `Expected "${name}" to be called exactly once by a call that succeeded, but ${matching.length} such calls were made.`),
|
|
44
44
|
};
|
|
45
45
|
},
|
|
46
46
|
toHaveCalledToolWith(received, name, args) {
|
package/dist/matchers.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"matchers.js","sourceRoot":"","sources":["../src/matchers.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;AA8BhC,SAAS,cAAc,CACrB,IAAmB,EACnB,IAAY;IAEZ,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;AACrE,CAAC;AAED,SAAS,aAAa,CAAC,IAAmB,EAAE,IAAY;IACtD,OAAO,cAAc,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,
|
|
1
|
+
{"version":3,"file":"matchers.js","sourceRoot":"","sources":["../src/matchers.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;AA8BhC,SAAS,cAAc,CACrB,IAAmB,EACnB,IAAY;IAEZ,OAAO,IAAI,CAAC,SAAS,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;AACrE,CAAC;AAED,SAAS,aAAa,CAAC,IAAmB,EAAE,IAAY;IACtD,OAAO,cAAc,CAAC,IAAI,EAAE,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;AAChF,CAAC;AAED,SAAS,QAAQ,CAAC,IAAmB;IACnC,IAAI,IAAI,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAChC,OAAO,wBAAwB,CAAC;IAClC,CAAC;IACD,OAAO,IAAI,CAAC,SAAS;SAClB,GAAG,CACF,CAAC,IAAuB,EAAE,KAAa,EAAE,EAAE,CACzC,KAAK,KAAK,GAAG,CAAC,KAAK,IAAI,CAAC,IAAI,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,IAAI,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,cAAc,IAAI,CAAC,MAAM,GAAG,EAAE,CACnI;SACA,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,SAAS,WAAW,CAAC,IAAmB;IACtC,OAAO,IAAI,CAAC,eAAe;SACxB,GAAG,CACF,CAAC,IAAI,EAAE,EAAE,CACP,OAAO,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,WAAW,IAAI,EAAE,oBAAoB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,WAAW,CAAC,EAAE,CACpG;SACA,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED,SAAS,MAAM,CAAC,IAAmB,EAAE,OAAe;IAClD,OAAO,GAAG,OAAO;;;EAGjB,QAAQ,CAAC,IAAI,CAAC;;;EAGd,WAAW,CAAC,IAAI,CAAC,EAAE,CAAC;AACtB,CAAC;AAED,MAAM,CAAC,MAAM,CAAC;IACZ,oBAAoB,CAClB,QAAuB,EACvB,IAAY,EACZ,IAA8B;QAE9B,MAAM,QAAQ,GAAG,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QAC/C,MAAM,SAAS,GACb,IAAI,KAAK,SAAS;YAClB,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC;gBACpB,IAAI,CAAC,MAAM,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,SAAS,EAAE,MAAM,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC,CAAC,CAAC;QACxE,MAAM,IAAI,GAAG,QAAQ,CAAC,MAAM,KAAK,CAAC,IAAI,SAAS,CAAC;QAEhD,OAAO;YACL,IAAI;YACJ,OAAO,EAAE,GAAG,EAAE,CACZ,MAAM,CACJ,QAAQ,EACR,IAAI;gBACF,CAAC,CAAC,aAAa,IAAI,kCAAkC;gBACrD,CAAC,CAAC,QAAQ,CAAC,MAAM,KAAK,CAAC;oBACrB,CAAC,CAAC,aAAa,IAAI,4BAA4B,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,4BAA4B,IAAI,CAAC,SAAS,CAAC,QAAQ,CAAC,CAAC,CAAC,EAAE,SAAS,CAAC,GAAG;oBACxI,CAAC,CAAC,aAAa,IAAI,6DAA6D,QAAQ,CAAC,MAAM,wBAAwB,CAC5H;SACJ,CAAC;IACJ,CAAC;IAED,oBAAoB,CAClB,QAAuB,EACvB,IAAY,EACZ,IAA6B;QAE7B,MAAM,IAAI,GAAG,aAAa,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CACvD,IAAI,CAAC,MAAM,CAAC,IAAI,CAAC,SAAS,EAAE,MAAM,CAAC,gBAAgB,CAAC,IAAI,CAAC,CAAC,CAC3D,CAAC;QAEF,OAAO;YACL,IAAI;YACJ,OAAO,EAAE,GAAG,EAAE,CACZ,MAAM,CACJ,QAAQ,EACR,IAAI;gBACF,CAAC,CAAC,wBAAwB,IAAI,UAAU,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG;gBAC/D,CAAC,CAAC,aAAa,IAAI,uBAAuB,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,GAAG,CACpE;SACJ,CAAC;IACJ,CAAC;IAED,qBAAqB,CAAC,QAAuB,EAAE,IAAY;QACzD,MAAM,QAAQ,GAAG,cAAc,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QAEhD,OAAO;YACL,IAAI,EAAE,QAAQ,CAAC,MAAM,KAAK,CAAC;YAC3B,OAAO,EAAE,GAAG,EAAE,CACZ,MAAM,CACJ,QAAQ,EACR,QAAQ,CAAC,MAAM,KAAK,CAAC;gBACnB,CAAC,CAAC,aAAa,IAAI,+BAA+B;gBAClD,CAAC,CAAC,aAAa,IAAI,2CAA2C,QAAQ,CAAC,MAAM,SAAS,CACzF;SACJ,CAAC;IACJ,CAAC;CACF,CAAC,CAAC;AAEH,MAAM,CAAC,IAAI,GAAG,CAAC,CAAM,IAAe,EAAE,EAAE,CACtC,MAAM,CAAC,IAAI,CAAC,CAAkC,CAAC"}
|
package/dist/model.d.ts
CHANGED
|
@@ -5,5 +5,5 @@ import type { ModelDescriptor } from "./config.js";
|
|
|
5
5
|
* the descriptor cannot construct, such as a company LLM gateway.
|
|
6
6
|
*/
|
|
7
7
|
export declare function defineEvalModel(model: LanguageModel): void;
|
|
8
|
-
export declare function resolveModel(descriptor: ModelDescriptor | undefined):
|
|
8
|
+
export declare function resolveModel(descriptor: ModelDescriptor | undefined): LanguageModel;
|
|
9
9
|
//# sourceMappingURL=model.d.ts.map
|
package/dist/model.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../src/model.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"model.d.ts","sourceRoot":"","sources":["../src/model.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AACxC,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,aAAa,CAAC;AAInD;;;GAGG;AACH,wBAAgB,eAAe,CAAC,KAAK,EAAE,aAAa,GAAG,IAAI,CAE1D;AAED,wBAAgB,YAAY,CAC1B,UAAU,EAAE,eAAe,GAAG,SAAS,GACtC,aAAa,CAef"}
|
package/dist/model.js
CHANGED
|
@@ -1,3 +1,4 @@
|
|
|
1
|
+
import { createAnthropic } from "@ai-sdk/anthropic";
|
|
1
2
|
let registered;
|
|
2
3
|
/**
|
|
3
4
|
* Registers a model instance from a setup file. The escape hatch for providers
|
|
@@ -6,23 +7,18 @@ let registered;
|
|
|
6
7
|
export function defineEvalModel(model) {
|
|
7
8
|
registered = model;
|
|
8
9
|
}
|
|
9
|
-
|
|
10
|
-
function requireKey(descriptor) {
|
|
11
|
-
const name = descriptor.apiKeyEnv ?? DEFAULT_KEY_ENV[descriptor.provider];
|
|
12
|
-
const key = process.env[name];
|
|
13
|
-
if (key === undefined || key === "") {
|
|
14
|
-
throw new Error(`${name} is not set`);
|
|
15
|
-
}
|
|
16
|
-
return key;
|
|
17
|
-
}
|
|
18
|
-
export async function resolveModel(descriptor) {
|
|
10
|
+
export function resolveModel(descriptor) {
|
|
19
11
|
if (registered !== undefined) {
|
|
20
12
|
return registered;
|
|
21
13
|
}
|
|
22
14
|
if (descriptor === undefined) {
|
|
23
15
|
throw new Error("No model configured. Pass `model` to the evals plugin, or call defineEvalModel() from a setup file.");
|
|
24
16
|
}
|
|
25
|
-
const
|
|
26
|
-
|
|
17
|
+
const name = descriptor.apiKeyEnv ?? "ANTHROPIC_API_KEY";
|
|
18
|
+
const apiKey = process.env[name];
|
|
19
|
+
if (apiKey === undefined || apiKey === "") {
|
|
20
|
+
throw new Error(`${name} is not set`);
|
|
21
|
+
}
|
|
22
|
+
return createAnthropic({ apiKey })(descriptor.name);
|
|
27
23
|
}
|
|
28
24
|
//# sourceMappingURL=model.js.map
|
package/dist/model.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"model.js","sourceRoot":"","sources":["../src/model.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"model.js","sourceRoot":"","sources":["../src/model.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AAIpD,IAAI,UAAqC,CAAC;AAE1C;;;GAGG;AACH,MAAM,UAAU,eAAe,CAAC,KAAoB;IAClD,UAAU,GAAG,KAAK,CAAC;AACrB,CAAC;AAED,MAAM,UAAU,YAAY,CAC1B,UAAuC;IAEvC,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;QAC7B,OAAO,UAAU,CAAC;IACpB,CAAC;IACD,IAAI,UAAU,KAAK,SAAS,EAAE,CAAC;QAC7B,MAAM,IAAI,KAAK,CACb,qGAAqG,CACtG,CAAC;IACJ,CAAC;IACD,MAAM,IAAI,GAAG,UAAU,CAAC,SAAS,IAAI,mBAAmB,CAAC;IACzD,MAAM,MAAM,GAAG,OAAO,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC;IACjC,IAAI,MAAM,KAAK,SAAS,IAAI,MAAM,KAAK,EAAE,EAAE,CAAC;QAC1C,MAAM,IAAI,KAAK,CAAC,GAAG,IAAI,aAAa,CAAC,CAAC;IACxC,CAAC;IACD,OAAO,eAAe,CAAC,EAAE,MAAM,EAAE,CAAC,CAAC,UAAU,CAAC,IAAI,CAAC,CAAC;AACtD,CAAC"}
|
package/dist/server-harness.d.ts
CHANGED
|
@@ -10,10 +10,9 @@ export interface RunningServer {
|
|
|
10
10
|
* Binding and reporting both happen in the child, so there is no window in
|
|
11
11
|
* which another process can steal the port.
|
|
12
12
|
*/
|
|
13
|
-
export declare function startServer({ cwd, command, env,
|
|
13
|
+
export declare function startServer({ cwd, command, env, }: {
|
|
14
14
|
cwd: string;
|
|
15
15
|
command: string[];
|
|
16
16
|
env?: Record<string, string>;
|
|
17
|
-
timeoutMs?: number;
|
|
18
17
|
}): Promise<RunningServer>;
|
|
19
18
|
//# sourceMappingURL=server-harness.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"server-harness.d.ts","sourceRoot":"","sources":["../src/server-harness.ts"],"names":[],"mappings":"AAGA,MAAM,WAAW,aAAa;IAC5B,GAAG,EAAE,MAAM,CAAC;IACZ,yEAAyE;IACzE,MAAM,EAAE,MAAM,MAAM,CAAC;IACrB,IAAI,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;CAC3B;AAiBD;;;;;GAKG;AACH,wBAAsB,WAAW,CAAC,EAChC,GAAG,EACH,OAAO,EACP,GAAQ,
|
|
1
|
+
{"version":3,"file":"server-harness.d.ts","sourceRoot":"","sources":["../src/server-harness.ts"],"names":[],"mappings":"AAGA,MAAM,WAAW,aAAa;IAC5B,GAAG,EAAE,MAAM,CAAC;IACZ,yEAAyE;IACzE,MAAM,EAAE,MAAM,MAAM,CAAC;IACrB,IAAI,EAAE,MAAM,OAAO,CAAC,IAAI,CAAC,CAAC;CAC3B;AAiBD;;;;;GAKG;AACH,wBAAsB,WAAW,CAAC,EAChC,GAAG,EACH,OAAO,EACP,GAAQ,GACT,EAAE;IACD,GAAG,EAAE,MAAM,CAAC;IACZ,OAAO,EAAE,MAAM,EAAE,CAAC;IAClB,GAAG,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,CAAC;CAC9B,GAAG,OAAO,CAAC,aAAa,CAAC,CA4EzB"}
|
package/dist/server-harness.js
CHANGED
|
@@ -15,7 +15,8 @@ function isListeningMessage(message) {
|
|
|
15
15
|
* Binding and reporting both happen in the child, so there is no window in
|
|
16
16
|
* which another process can steal the port.
|
|
17
17
|
*/
|
|
18
|
-
export async function startServer({ cwd, command, env = {},
|
|
18
|
+
export async function startServer({ cwd, command, env = {}, }) {
|
|
19
|
+
const timeoutMs = 60_000;
|
|
19
20
|
const [bin, ...args] = command;
|
|
20
21
|
if (bin === undefined) {
|
|
21
22
|
throw new Error("Empty server command");
|
|
@@ -53,7 +54,7 @@ export async function startServer({ cwd, command, env = {}, timeoutMs = 60_000,
|
|
|
53
54
|
};
|
|
54
55
|
const port = await new Promise((resolvePort, reject) => {
|
|
55
56
|
const timer = setTimeout(() => {
|
|
56
|
-
reject(new Error(`Server did not report
|
|
57
|
+
reject(new Error(`Server did not report a port in ${timeoutMs}ms. \`command\` must start the server process directly (node, tsx): wrappers such as \`pnpm run\` do not forward the IPC channel the readiness signal uses. It must also run a skybridge version that sends skybridge:listening.\n${output()}`));
|
|
57
58
|
}, timeoutMs);
|
|
58
59
|
child.on("message", (message) => {
|
|
59
60
|
if (isListeningMessage(message)) {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"server-harness.js","sourceRoot":"","sources":["../src/server-harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAqB,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC9D,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AASpC,SAAS,SAAS,CAAC,KAAmB;IACpC,OAAO,KAAK,CAAC,QAAQ,KAAK,IAAI,IAAI,KAAK,CAAC,UAAU,KAAK,IAAI,CAAC;AAC9D,CAAC;AAED,SAAS,kBAAkB,CACzB,OAAgB;IAEhB,OAAO,CACL,OAAO,OAAO,KAAK,QAAQ;QAC3B,OAAO,KAAK,IAAI;QACf,OAA8B,CAAC,IAAI,KAAK,qBAAqB;QAC9D,OAAQ,OAA8B,CAAC,IAAI,KAAK,QAAQ,CACzD,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,EAChC,GAAG,EACH,OAAO,EACP,GAAG,GAAG,EAAE,
|
|
1
|
+
{"version":3,"file":"server-harness.js","sourceRoot":"","sources":["../src/server-harness.ts"],"names":[],"mappings":"AAAA,OAAO,EAAqB,KAAK,EAAE,MAAM,oBAAoB,CAAC;AAC9D,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AASpC,SAAS,SAAS,CAAC,KAAmB;IACpC,OAAO,KAAK,CAAC,QAAQ,KAAK,IAAI,IAAI,KAAK,CAAC,UAAU,KAAK,IAAI,CAAC;AAC9D,CAAC;AAED,SAAS,kBAAkB,CACzB,OAAgB;IAEhB,OAAO,CACL,OAAO,OAAO,KAAK,QAAQ;QAC3B,OAAO,KAAK,IAAI;QACf,OAA8B,CAAC,IAAI,KAAK,qBAAqB;QAC9D,OAAQ,OAA8B,CAAC,IAAI,KAAK,QAAQ,CACzD,CAAC;AACJ,CAAC;AAED;;;;;GAKG;AACH,MAAM,CAAC,KAAK,UAAU,WAAW,CAAC,EAChC,GAAG,EACH,OAAO,EACP,GAAG,GAAG,EAAE,GAKT;IACC,MAAM,SAAS,GAAG,MAAM,CAAC;IACzB,MAAM,CAAC,GAAG,EAAE,GAAG,IAAI,CAAC,GAAG,OAAO,CAAC;IAC/B,IAAI,GAAG,KAAK,SAAS,EAAE,CAAC;QACtB,MAAM,IAAI,KAAK,CAAC,sBAAsB,CAAC,CAAC;IAC1C,CAAC;IAED,MAAM,UAAU,GAAG,GAAG,CAAC,UAAU,CAAC,GAAG,CAAC,CAAC,CAAC,CAAC,OAAO,CAAC,GAAG,EAAE,GAAG,CAAC,CAAC,CAAC,CAAC,GAAG,CAAC;IAEjE,MAAM,KAAK,GAAG,KAAK,CAAC,UAAU,EAAE,IAAI,EAAE;QACpC,GAAG;QACH,GAAG,EAAE;YACH,GAAG,OAAO,CAAC,GAAG;YACd,QAAQ,EAAE,YAAY;YACtB,GAAG,GAAG;YACN,MAAM,EAAE,GAAG;SACZ;QACD,KAAK,EAAE,CAAC,QAAQ,EAAE,MAAM,EAAE,MAAM,EAAE,KAAK,CAAC;KACzC,CAAC,CAAC;IAEH,IAAI,QAAQ,GAAG,EAAE,CAAC;IAClB,MAAM,MAAM,GAAG,GAAG,EAAE,CAAC,QAAQ,CAAC;IAC9B,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE;QACjC,QAAQ,IAAI,KAAK,CAAC;IACpB,CAAC,CAAC,CAAC;IACH,KAAK,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,EAAE,CAAC,KAAK,EAAE,EAAE;QACjC,QAAQ,IAAI,KAAK,CAAC;IACpB,CAAC,CAAC,CAAC;IAEH,MAAM,IAAI,GAAG,KAAK,IAAI,EAAE;QACtB,IAAI,SAAS,CAAC,KAAK,CAAC,EAAE,CAAC;YACrB,OAAO;QACT,CAAC;QACD,MAAM,MAAM,GAAG,IAAI,OAAO,CAAO,CAAC,WAAW,EAAE,EAAE;YAC/C,KAAK,CAAC,IAAI,CAAC,MAAM,EAAE,GAAG,EAAE,CAAC,WAAW,EAAE,CAAC,CAAC;QAC1C,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,CAAC;QACtB,MAAM,MAAM,GAAG,UAAU,CAAC,GAAG,EAAE,CAAC,KAAK,CAAC,IAAI,CAAC,SAAS,CAAC,EAAE,KAAK,CAAC,CAAC;QAC9D,MAAM,MAAM,CAAC;QACb,YAAY,CAAC,MAAM,CAAC,CAAC;IACvB,CAAC,CAAC;IAEF,MAAM,IAAI,GAAG,MAAM,IAAI,OAAO,CAAS,CAAC,WAAW,EAAE,MAAM,EAAE,EAAE;QAC7D,MAAM,KAAK,GAAG,UAAU,CAAC,GAAG,EAAE;YAC5B,MAAM,CACJ,IAAI,KAAK,CACP,mCAAmC,SAAS,qOAAqO,MAAM,EAAE,EAAE,CAC5R,CACF,CAAC;QACJ,CAAC,EAAE,SAAS,CAAC,CAAC;QACd,KAAK,CAAC,EAAE,CAAC,SAAS,EAAE,CAAC,OAAO,EAAE,EAAE;YAC9B,IAAI,kBAAkB,CAAC,OAAO,CAAC,EAAE,CAAC;gBAChC,YAAY,CAAC,KAAK,CAAC,CAAC;gBACpB,WAAW,CAAC,OAAO,CAAC,IAAI,CAAC,CAAC;YAC5B,CAAC;QACH,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,OAAO,EAAE,CAAC,KAAK,EAAE,EAAE;YAC1B,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,MAAM,CACJ,IAAI,KAAK,CAAC,mBAAmB,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,KAAK,KAAK,CAAC,OAAO,EAAE,CAAC,CACpE,CAAC;QACJ,CAAC,CAAC,CAAC;QACH,KAAK,CAAC,EAAE,CAAC,MAAM,EAAE,GAAG,EAAE;YACpB,YAAY,CAAC,KAAK,CAAC,CAAC;YACpB,MAAM,CACJ,IAAI,KAAK,CACP,2BAA2B,KAAK,CAAC,QAAQ,gCAAgC,MAAM,EAAE,EAAE,CACpF,CACF,CAAC;QACJ,CAAC,CAAC,CAAC;IACL,CAAC,CAAC,CAAC,KAAK,CAAC,KAAK,EAAE,KAAc,EAAE,EAAE;QAChC,MAAM,IAAI,EAAE,CAAC;QACb,MAAM,KAAK,CAAC;IACd,CAAC,CAAC,CAAC;IAEH,OAAO,EAAE,GAAG,EAAE,oBAAoB,IAAI,MAAM,EAAE,MAAM,EAAE,IAAI,EAAE,CAAC;AAC/D,CAAC"}
|
|
@@ -1,8 +1,9 @@
|
|
|
1
1
|
import { Chat } from "./chat.js";
|
|
2
2
|
/**
|
|
3
3
|
* Opens a fresh MCP session and conversation against the server the plugin
|
|
4
|
-
* started.
|
|
5
|
-
*
|
|
4
|
+
* started. The session is closed when the current test finishes, so scenarios
|
|
5
|
+
* never leak sessions and never write teardown, and tests that run
|
|
6
|
+
* concurrently cannot close each other's sessions.
|
|
6
7
|
*
|
|
7
8
|
* The type parameter pins the assertions to the project's registry:
|
|
8
9
|
* `start<AppType>()` returns a `Chat<AppType>`, and `expect.chat` infers the
|
|
@@ -12,5 +13,4 @@ export declare function start<App>(overrides?: {
|
|
|
12
13
|
systemPrompt?: string;
|
|
13
14
|
temperature?: number;
|
|
14
15
|
}): Promise<Chat<App>>;
|
|
15
|
-
export declare function closeOpenedSessions(): Promise<void>;
|
|
16
16
|
//# sourceMappingURL=session-registry.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;
|
|
1
|
+
{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AACA,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AAGjC;;;;;;;;;GASG;AACH,wBAAsB,KAAK,CAAC,GAAG,EAAE,SAAS,CAAC,EAAE;IAC3C,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;CACtB,GAAG,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAUrB"}
|
package/dist/session-registry.js
CHANGED
|
@@ -1,11 +1,11 @@
|
|
|
1
|
-
import { inject } from "vitest";
|
|
1
|
+
import { inject, onTestFinished } from "vitest";
|
|
2
2
|
import { Chat } from "./chat.js";
|
|
3
3
|
import { resolveModel } from "./model.js";
|
|
4
|
-
const opened = [];
|
|
5
4
|
/**
|
|
6
5
|
* Opens a fresh MCP session and conversation against the server the plugin
|
|
7
|
-
* started.
|
|
8
|
-
*
|
|
6
|
+
* started. The session is closed when the current test finishes, so scenarios
|
|
7
|
+
* never leak sessions and never write teardown, and tests that run
|
|
8
|
+
* concurrently cannot close each other's sessions.
|
|
9
9
|
*
|
|
10
10
|
* The type parameter pins the assertions to the project's registry:
|
|
11
11
|
* `start<AppType>()` returns a `Chat<AppType>`, and `expect.chat` infers the
|
|
@@ -14,15 +14,12 @@ const opened = [];
|
|
|
14
14
|
export async function start(overrides) {
|
|
15
15
|
const config = inject("skybridgeEvals");
|
|
16
16
|
const chat = await Chat.open(inject("skybridgeEvalsUrl"), {
|
|
17
|
-
model:
|
|
17
|
+
model: resolveModel(config.model),
|
|
18
18
|
temperature: overrides?.temperature ?? config.temperature,
|
|
19
19
|
systemPrompt: overrides?.systemPrompt ?? config.systemPrompt,
|
|
20
20
|
maxSteps: config.maxSteps,
|
|
21
21
|
});
|
|
22
|
-
|
|
22
|
+
onTestFinished(() => chat.close());
|
|
23
23
|
return chat;
|
|
24
24
|
}
|
|
25
|
-
export async function closeOpenedSessions() {
|
|
26
|
-
await Promise.all(opened.splice(0).map((chat) => chat.close()));
|
|
27
|
-
}
|
|
28
25
|
//# sourceMappingURL=session-registry.js.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;
|
|
1
|
+
{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,IAAI,EAAE,MAAM,WAAW,CAAC;AACjC,OAAO,EAAE,YAAY,EAAE,MAAM,YAAY,CAAC;AAE1C;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CAAM,SAGhC;IACC,MAAM,MAAM,GAAG,MAAM,CAAC,gBAAgB,CAAC,CAAC;IACxC,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAAM,MAAM,CAAC,mBAAmB,CAAC,EAAE;QAC7D,KAAK,EAAE,YAAY,CAAC,MAAM,CAAC,KAAK,CAAC;QACjC,WAAW,EAAE,SAAS,EAAE,WAAW,IAAI,MAAM,CAAC,WAAW;QACzD,YAAY,EAAE,SAAS,EAAE,YAAY,IAAI,MAAM,CAAC,YAAY;QAC5D,QAAQ,EAAE,MAAM,CAAC,QAAQ;KAC1B,CAAC,CAAC;IACH,cAAc,CAAC,GAAG,EAAE,CAAC,IAAI,CAAC,KAAK,EAAE,CAAC,CAAC;IACnC,OAAO,IAAI,CAAC;AACd,CAAC"}
|
package/dist/setup.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"setup.d.ts","sourceRoot":"","sources":["../src/setup.ts"],"names":[],"mappings":"
|
|
1
|
+
{"version":3,"file":"setup.d.ts","sourceRoot":"","sources":["../src/setup.ts"],"names":[],"mappings":"AAAA,OAAO,eAAe,CAAC"}
|
package/dist/setup.js
CHANGED
package/dist/setup.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"setup.js","sourceRoot":"","sources":["../src/setup.ts"],"names":[],"mappings":"AAAA,OAAO,
|
|
1
|
+
{"version":3,"file":"setup.js","sourceRoot":"","sources":["../src/setup.ts"],"names":[],"mappings":"AAAA,OAAO,eAAe,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@skybridge/test",
|
|
3
|
-
"version": "0.0.0-dev.
|
|
3
|
+
"version": "0.0.0-dev.df0cfdd",
|
|
4
4
|
"description": "Eval runner for Skybridge MCP apps: does a prompt reach the right tool with the right arguments",
|
|
5
5
|
"license": "ISC",
|
|
6
6
|
"type": "module",
|
|
@@ -25,17 +25,17 @@
|
|
|
25
25
|
"@ai-sdk/anthropic": "^2.0.0",
|
|
26
26
|
"@modelcontextprotocol/sdk": "^1.29.0",
|
|
27
27
|
"ai": "^5.0.0",
|
|
28
|
-
"@skybridge/vite-plugin": "0.0.0-dev.
|
|
28
|
+
"@skybridge/vite-plugin": "0.0.0-dev.df0cfdd"
|
|
29
29
|
},
|
|
30
30
|
"peerDependencies": {
|
|
31
|
-
"skybridge": "^
|
|
31
|
+
"skybridge": "^2.0.0",
|
|
32
32
|
"vitest": "^4.1.0"
|
|
33
33
|
},
|
|
34
34
|
"devDependencies": {
|
|
35
35
|
"@types/node": "^24.13.3",
|
|
36
36
|
"typescript": "^5.9.3",
|
|
37
37
|
"vitest": "^4.1.10",
|
|
38
|
-
"skybridge": "0.0.0-dev.
|
|
38
|
+
"skybridge": "0.0.0-dev.df0cfdd"
|
|
39
39
|
},
|
|
40
40
|
"engines": {
|
|
41
41
|
"node": ">=22.23.1"
|