@skybridge/test 0.0.0-dev.e3d3217 → 0.0.0-dev.ec70681
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/README.md +1 -1
- package/dist/chat.d.ts +9 -1
- package/dist/chat.d.ts.map +1 -1
- package/dist/chat.js +31 -4
- package/dist/chat.js.map +1 -1
- package/dist/index.d.ts +2 -2
- package/dist/index.d.ts.map +1 -1
- package/dist/index.js.map +1 -1
- package/dist/matchers/index.d.ts +1 -1
- package/dist/matchers/index.d.ts.map +1 -1
- package/dist/matchers/matchers.d.ts +18 -0
- package/dist/matchers/matchers.d.ts.map +1 -1
- package/dist/matchers/matchers.js +29 -19
- package/dist/matchers/matchers.js.map +1 -1
- package/dist/session-registry.d.ts +16 -0
- package/dist/session-registry.d.ts.map +1 -1
- package/dist/session-registry.js +6 -1
- package/dist/session-registry.js.map +1 -1
- package/package.json +4 -4
package/README.md
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
# @skybridge/test
|
|
2
2
|
|
|
3
|
-
Evals for [Skybridge](https://skybridge.tech) MCP
|
|
3
|
+
Evals for [Skybridge](https://skybridge.tech) MCP Apps: they answer whether a
|
|
4
4
|
prompt reaches the right tool with the right arguments, which no unit test can
|
|
5
5
|
check.
|
|
6
6
|
|
package/dist/chat.d.ts
CHANGED
|
@@ -6,6 +6,13 @@ interface HostConfig {
|
|
|
6
6
|
systemPrompt?: string;
|
|
7
7
|
maxSteps?: number;
|
|
8
8
|
}
|
|
9
|
+
/**
|
|
10
|
+
* Stubbed results keyed by tool name. The typed form scenarios write lives on
|
|
11
|
+
* `start`'s options.
|
|
12
|
+
*
|
|
13
|
+
* @internal
|
|
14
|
+
*/
|
|
15
|
+
export type StubMap = Record<string, (args: Record<string, unknown>) => unknown>;
|
|
9
16
|
/**
|
|
10
17
|
* A `fetch` replacement the transport dials instead of the network. Mirrors the
|
|
11
18
|
* MCP client's own `fetch` option so an in-process app can serve the session.
|
|
@@ -29,11 +36,12 @@ export declare class Chat<App = unknown> {
|
|
|
29
36
|
private readonly client;
|
|
30
37
|
private readonly transport;
|
|
31
38
|
private readonly host;
|
|
39
|
+
private readonly stubs;
|
|
32
40
|
private tools;
|
|
33
41
|
/** The model the conversation runs on, and the judge's default. */
|
|
34
42
|
get model(): LanguageModel;
|
|
35
43
|
private constructor();
|
|
36
|
-
static open<App>(host: HostConfig, fetchImpl: TransportFetch): Promise<Chat<App>>;
|
|
44
|
+
static open<App>(host: HostConfig, fetchImpl: TransportFetch, stubs: StubMap): Promise<Chat<App>>;
|
|
37
45
|
/** Takes a turn. Assert on the result with `expect.chat`. */
|
|
38
46
|
send(prompt: string): Promise<void>;
|
|
39
47
|
close(): Promise<void>;
|
package/dist/chat.d.ts.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAKA,OAAO,EAIL,KAAK,aAAa,EAGnB,MAAM,IAAI,CAAC;AACZ,OAAO,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,qBAAqB,CAAC;AAE1D,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;
|
|
1
|
+
{"version":3,"file":"chat.d.ts","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAKA,OAAO,EAIL,KAAK,aAAa,EAGnB,MAAM,IAAI,CAAC;AACZ,OAAO,KAAK,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,qBAAqB,CAAC;AAE1D,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;;;;;GAKG;AACH,MAAM,MAAM,OAAO,GAAG,MAAM,CAC1B,MAAM,EACN,CAAC,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,KAAK,OAAO,CAC3C,CAAC;AA4BF;;;GAGG;AACH,MAAM,MAAM,cAAc,GAAG,CAC3B,GAAG,EAAE,MAAM,GAAG,GAAG,EACjB,IAAI,CAAC,EAAE,WAAW,KACf,OAAO,CAAC,QAAQ,CAAC,CAAC;AAEvB;;;GAGG;AACH,qBAAa,IAAI,CAAC,GAAG,GAAG,OAAO;IAC7B,4EAA4E;IAC5E,SAAiB,IAAI,EAAE,GAAG,CAAC;IAE3B,0DAA0D;IAC1D,QAAQ,CAAC,KAAK,EAAE,IAAI,CAAC,GAAG,CAAC,EAAE,CAAM;IAEjC,mEAAmE;IACnE,IAAI,SAAS,IAAI,QAAQ,CAAC,GAAG,CAAC,EAAE,CAI/B;IAED,+EAA+E;IAC/E,IAAI,cAAc,IAAI,MAAM,EAAE,CAI7B;IAED,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,QAAQ,CAAC,KAAK,CAA+B;IACrD,OAAO,CAAC,KAAK,CAAc;IAE3B,mEAAmE;IACnE,IAAI,KAAK,IAAI,aAAa,CAEzB;IAED,OAAO;WAYM,IAAI,CAAC,GAAG,EACnB,IAAI,EAAE,UAAU,EAChB,SAAS,EAAE,cAAc,EACzB,KAAK,EAAE,OAAO,GACb,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC;IA0BrB,6DAA6D;IACvD,IAAI,CAAC,MAAM,EAAE,MAAM,GAAG,OAAO,CAAC,IAAI,CAAC;IAmCnC,KAAK,IAAI,OAAO,CAAC,IAAI,CAAC;IAI5B,OAAO,CAAC,UAAU;IAalB,OAAO,CAAC,OAAO;CA8ChB"}
|
package/dist/chat.js
CHANGED
|
@@ -2,6 +2,20 @@ import { Client, StreamableHTTPClientTransport, } from "@modelcontextprotocol/cl
|
|
|
2
2
|
import { dynamicTool, generateText, jsonSchema, stepCountIs, } from "ai";
|
|
3
3
|
const DEFAULT_MAX_STEPS = 8;
|
|
4
4
|
const IN_PROCESS_URL = "http://in-process.skybridge.test/mcp";
|
|
5
|
+
/**
|
|
6
|
+
* Dress a stubbed value the way the server dresses a real one, so the model
|
|
7
|
+
* reads the same envelope under a stub as it does live.
|
|
8
|
+
*/
|
|
9
|
+
function asToolResult(value) {
|
|
10
|
+
return {
|
|
11
|
+
content: [
|
|
12
|
+
{
|
|
13
|
+
type: "text",
|
|
14
|
+
text: typeof value === "string" ? value : JSON.stringify(value),
|
|
15
|
+
},
|
|
16
|
+
],
|
|
17
|
+
};
|
|
18
|
+
}
|
|
5
19
|
function outputOf(results, toolCallId) {
|
|
6
20
|
return results.find((result) => result.toolCallId === toolCallId)?.output;
|
|
7
21
|
}
|
|
@@ -29,24 +43,30 @@ export class Chat {
|
|
|
29
43
|
client;
|
|
30
44
|
transport;
|
|
31
45
|
host;
|
|
46
|
+
stubs;
|
|
32
47
|
tools = [];
|
|
33
48
|
/** The model the conversation runs on, and the judge's default. */
|
|
34
49
|
get model() {
|
|
35
50
|
return this.host.model;
|
|
36
51
|
}
|
|
37
|
-
constructor(client, transport, host) {
|
|
52
|
+
constructor(client, transport, host, stubs) {
|
|
38
53
|
this.client = client;
|
|
39
54
|
this.transport = transport;
|
|
40
55
|
this.host = host;
|
|
56
|
+
this.stubs = new Map(Object.entries(stubs));
|
|
41
57
|
}
|
|
42
|
-
static async open(host, fetchImpl) {
|
|
58
|
+
static async open(host, fetchImpl, stubs) {
|
|
43
59
|
const client = new Client({ name: "skybridge-eval", version: "0" });
|
|
44
60
|
const transport = new StreamableHTTPClientTransport(new URL(IN_PROCESS_URL), { fetch: fetchImpl });
|
|
45
61
|
await client.connect(transport);
|
|
46
|
-
const chat = new Chat(client, transport, host);
|
|
62
|
+
const chat = new Chat(client, transport, host, stubs);
|
|
47
63
|
try {
|
|
48
64
|
const { tools } = await client.listTools();
|
|
49
65
|
chat.tools = tools;
|
|
66
|
+
const unknownStubs = Object.keys(stubs).filter((name) => !tools.some((tool) => tool.name === name));
|
|
67
|
+
if (unknownStubs.length > 0) {
|
|
68
|
+
throw new Error(`the app exposes no tool named ${unknownStubs.map((name) => `"${name}"`).join(", ")}, so the stub would never run`);
|
|
69
|
+
}
|
|
50
70
|
}
|
|
51
71
|
catch (error) {
|
|
52
72
|
await transport.close();
|
|
@@ -108,6 +128,10 @@ export class Chat {
|
|
|
108
128
|
description: definition.description ?? "",
|
|
109
129
|
inputSchema: jsonSchema(parameters),
|
|
110
130
|
execute: async (input, { toolCallId }) => {
|
|
131
|
+
const stubbed = await this.stubs.get(definition.name)?.((input ?? {}));
|
|
132
|
+
if (stubbed !== undefined) {
|
|
133
|
+
return asToolResult(stubbed);
|
|
134
|
+
}
|
|
111
135
|
let result;
|
|
112
136
|
try {
|
|
113
137
|
result = await this.client.callTool({
|
|
@@ -122,7 +146,10 @@ export class Chat {
|
|
|
122
146
|
if (result.isError === true) {
|
|
123
147
|
this.failures.set(toolCallId, `the tool returned an error: ${JSON.stringify(result.content)}`);
|
|
124
148
|
}
|
|
125
|
-
return
|
|
149
|
+
return {
|
|
150
|
+
content: result.content,
|
|
151
|
+
structuredContent: result.structuredContent,
|
|
152
|
+
};
|
|
126
153
|
},
|
|
127
154
|
}),
|
|
128
155
|
];
|
package/dist/chat.js.map
CHANGED
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"chat.js","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,MAAM,EACN,6BAA6B,GAE9B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EAGV,WAAW,GACZ,MAAM,IAAI,CAAC;
|
|
1
|
+
{"version":3,"file":"chat.js","sourceRoot":"","sources":["../src/chat.ts"],"names":[],"mappings":"AAAA,OAAO,EACL,MAAM,EACN,6BAA6B,GAE9B,MAAM,8BAA8B,CAAC;AACtC,OAAO,EACL,WAAW,EACX,YAAY,EACZ,UAAU,EAGV,WAAW,GACZ,MAAM,IAAI,CAAC;AAqBZ,MAAM,iBAAiB,GAAG,CAAC,CAAC;AAE5B,MAAM,cAAc,GAAG,sCAAsC,CAAC;AAE9D;;;GAGG;AACH,SAAS,YAAY,CAAC,KAAc;IAClC,OAAO;QACL,OAAO,EAAE;YACP;gBACE,IAAI,EAAE,MAAM;gBACZ,IAAI,EAAE,OAAO,KAAK,KAAK,QAAQ,CAAC,CAAC,CAAC,KAAK,CAAC,CAAC,CAAC,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC;aAChE;SACF;KACF,CAAC;AACJ,CAAC;AAED,SAAS,QAAQ,CACf,OAAmD,EACnD,UAAkB;IAElB,OAAO,OAAO,CAAC,IAAI,CAAC,CAAC,MAAM,EAAE,EAAE,CAAC,MAAM,CAAC,UAAU,KAAK,UAAU,CAAC,EAAE,MAAM,CAAC;AAC5E,CAAC;AAWD;;;GAGG;AACH,MAAM,OAAO,IAAI;IAIf,0DAA0D;IACjD,KAAK,GAAgB,EAAE,CAAC;IAEjC,mEAAmE;IACnE,IAAI,SAAS;QACX,OAAO,IAAI,CAAC,KAAK;aACd,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,MAAM,CAAC;aACtC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC9B,CAAC;IAED,+EAA+E;IAC/E,IAAI,cAAc;QAChB,OAAO,IAAI,CAAC,KAAK;aACd,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,WAAW,CAAC;aAC3C,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;IAC9B,CAAC;IAEgB,QAAQ,GAAmB,EAAE,CAAC;IAC9B,QAAQ,GAAG,IAAI,GAAG,EAAkB,CAAC;IACrC,MAAM,CAAS;IACf,SAAS,CAAgC;IACzC,IAAI,CAAa;IACjB,KAAK,CAA+B;IAC7C,KAAK,GAAW,EAAE,CAAC;IAE3B,mEAAmE;IACnE,IAAI,KAAK;QACP,OAAO,IAAI,CAAC,IAAI,CAAC,KAAK,CAAC;IACzB,CAAC;IAED,YACE,MAAc,EACd,SAAwC,EACxC,IAAgB,EAChB,KAAc;QAEd,IAAI,CAAC,MAAM,GAAG,MAAM,CAAC;QACrB,IAAI,CAAC,SAAS,GAAG,SAAS,CAAC;QAC3B,IAAI,CAAC,IAAI,GAAG,IAAI,CAAC;QACjB,IAAI,CAAC,KAAK,GAAG,IAAI,GAAG,CAAC,MAAM,CAAC,OAAO,CAAC,KAAK,CAAC,CAAC,CAAC;IAC9C,CAAC;IAED,MAAM,CAAC,KAAK,CAAC,IAAI,CACf,IAAgB,EAChB,SAAyB,EACzB,KAAc;QAEd,MAAM,MAAM,GAAG,IAAI,MAAM,CAAC,EAAE,IAAI,EAAE,gBAAgB,EAAE,OAAO,EAAE,GAAG,EAAE,CAAC,CAAC;QACpE,MAAM,SAAS,GAAG,IAAI,6BAA6B,CACjD,IAAI,GAAG,CAAC,cAAc,CAAC,EACvB,EAAE,KAAK,EAAE,SAAS,EAAE,CACrB,CAAC;QACF,MAAM,MAAM,CAAC,OAAO,CAAC,SAAS,CAAC,CAAC;QAChC,MAAM,IAAI,GAAG,IAAI,IAAI,CAAM,MAAM,EAAE,SAAS,EAAE,IAAI,EAAE,KAAK,CAAC,CAAC;QAC3D,IAAI,CAAC;YACH,MAAM,EAAE,KAAK,EAAE,GAAG,MAAM,MAAM,CAAC,SAAS,EAAE,CAAC;YAC3C,IAAI,CAAC,KAAK,GAAG,KAAK,CAAC;YACnB,MAAM,YAAY,GAAG,MAAM,CAAC,IAAI,CAAC,KAAK,CAAC,CAAC,MAAM,CAC5C,CAAC,IAAI,EAAE,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,IAAI,CAAC,CACpD,CAAC;YACF,IAAI,YAAY,CAAC,MAAM,GAAG,CAAC,EAAE,CAAC;gBAC5B,MAAM,IAAI,KAAK,CACb,iCAAiC,YAAY,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,+BAA+B,CACnH,CAAC;YACJ,CAAC;QACH,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,6DAA6D;IAC7D,KAAK,CAAC,IAAI,CAAC,MAAc;QACvB,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,OAAO,EAAE,MAAM,EAAE,CAAC,CAAC;QACtD,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,MAAM,EAAE,IAAI,EAAE,MAAM,EAAE,CAAC,CAAC;QAEhD,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,IAAI,IAAI,CAAC,IAAI,KAAK,EAAE,EAAE,CAAC;gBACrB,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC,EAAE,IAAI,EAAE,WAAW,EAAE,IAAI,EAAE,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC;YAC1D,CAAC;YACD,KAAK,MAAM,IAAI,IAAI,IAAI,CAAC,SAAS,EAAE,CAAC;gBAClC,IAAI,CAAC,KAAK,CAAC,IAAI,CAAC;oBACd,IAAI,EAAE,MAAM;oBACZ,IAAI,EAAE;wBACJ,IAAI,EAAE,IAAI,CAAC,QAAQ;wBACnB,SAAS,EAAE,CAAC,IAAI,CAAC,KAAK,IAAI,EAAE,CAA4B;wBACxD,GAAG,IAAI,CAAC,UAAU,CAAC,IAAI,CAAC;qBACR;oBAClB,MAAM,EAAE,QAAQ,CAAC,IAAI,CAAC,WAAW,EAAE,IAAI,CAAC,UAAU,CAAC;iBACpD,CAAC,CAAC;YACL,CAAC;QACH,CAAC;QAED,IAAI,CAAC,QAAQ,CAAC,IAAI,CAAC,GAAG,MAAM,CAAC,QAAQ,CAAC,QAAQ,CAAC,CAAC;IAClD,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,CACrB,UAA8C,CAC/C;oBACD,OAAO,EAAE,KAAK,EAAE,KAAK,EAAE,EAAE,UAAU,EAAE,EAAE,EAAE;wBACvC,MAAM,OAAO,GAAG,MAAM,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,UAAU,CAAC,IAAI,CAAC,EAAE,CACrD,CAAC,KAAK,IAAI,EAAE,CAA4B,CACzC,CAAC;wBACF,IAAI,OAAO,KAAK,SAAS,EAAE,CAAC;4BAC1B,OAAO,YAAY,CAAC,OAAO,CAAC,CAAC;wBAC/B,CAAC;wBACD,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;4BACL,OAAO,EAAE,MAAM,CAAC,OAAO;4BACvB,iBAAiB,EAAE,MAAM,CAAC,iBAAiB;yBAC5C,CAAC;oBACJ,CAAC;iBACF,CAAC;aACM,CAAC;QACb,CAAC,CAAC,CAAC;QACH,OAAO,MAAM,CAAC,WAAW,CAAC,OAAO,CAAC,CAAC;IACrC,CAAC;CACF"}
|
package/dist/index.d.ts
CHANGED
|
@@ -1,5 +1,5 @@
|
|
|
1
1
|
export type { Chat } from "./chat.js";
|
|
2
|
-
export type { ChatAssertion, ChatMatchers, JudgmentOptions, } from "./matchers/index.js";
|
|
3
|
-
export type { EvalApp, EvalIdentity } from "./session-registry.js";
|
|
2
|
+
export type { ChatAssertion, ChatMatchers, Judgment, JudgmentOptions, Verdict, } from "./matchers/index.js";
|
|
3
|
+
export type { EvalApp, EvalIdentity, ToolStubs, } from "./session-registry.js";
|
|
4
4
|
export { start } from "./session-registry.js";
|
|
5
5
|
//# sourceMappingURL=index.d.ts.map
|
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,EACV,aAAa,EACb,YAAY,EACZ,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,EACV,aAAa,EACb,YAAY,EACZ,QAAQ,EACR,eAAe,EACf,OAAO,GACR,MAAM,qBAAqB,CAAC;AAC7B,YAAY,EACV,OAAO,EACP,YAAY,EACZ,SAAS,GACV,MAAM,uBAAuB,CAAC;AAC/B,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":"AAaA,OAAO,EAAE,KAAK,EAAE,MAAM,uBAAuB,CAAC"}
|
package/dist/matchers/index.d.ts
CHANGED
|
@@ -1,4 +1,4 @@
|
|
|
1
1
|
import "./matchers.js";
|
|
2
|
-
export type { ChatAssertion, ChatMatchers, JudgmentOptions, } from "./matchers.js";
|
|
2
|
+
export type { ChatAssertion, ChatMatchers, Judgment, JudgmentOptions, Verdict, } from "./matchers.js";
|
|
3
3
|
export type { ChatLike, ToolCall, Turn } from "./types.js";
|
|
4
4
|
//# sourceMappingURL=index.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/matchers/index.ts"],"names":[],"mappings":"AAAA,OAAO,eAAe,CAAC;AAEvB,YAAY,EACV,aAAa,EACb,YAAY,EACZ,eAAe,
|
|
1
|
+
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/matchers/index.ts"],"names":[],"mappings":"AAAA,OAAO,eAAe,CAAC;AAEvB,YAAY,EACV,aAAa,EACb,YAAY,EACZ,QAAQ,EACR,eAAe,EACf,OAAO,GACR,MAAM,eAAe,CAAC;AACvB,YAAY,EAAE,QAAQ,EAAE,QAAQ,EAAE,IAAI,EAAE,MAAM,YAAY,CAAC"}
|
|
@@ -2,10 +2,28 @@ import "vitest";
|
|
|
2
2
|
import type { LanguageModel } from "ai";
|
|
3
3
|
import type { ToolInput, ToolNames } from "skybridge/server";
|
|
4
4
|
import type { ChatLike } from "./types.js";
|
|
5
|
+
/** What a judge decided about a conversation. */
|
|
6
|
+
export interface Verdict {
|
|
7
|
+
pass: boolean;
|
|
8
|
+
/** Why, quoted in the failure message. Omit it when the judge gives no prose. */
|
|
9
|
+
reasoning?: string;
|
|
10
|
+
}
|
|
11
|
+
/** The conversation handed to a {@link JudgmentOptions.judge}. */
|
|
12
|
+
export interface Judgment {
|
|
13
|
+
criteria: string;
|
|
14
|
+
/** The rendered conversation, one turn per line, tool results included. */
|
|
15
|
+
transcript: string;
|
|
16
|
+
}
|
|
5
17
|
/** Options for {@link ChatMatchers.toPassJudgment}. */
|
|
6
18
|
export interface JudgmentOptions {
|
|
7
19
|
/** Overrides the judge for this assertion. Defaults to the chat's model. */
|
|
8
20
|
model?: LanguageModel;
|
|
21
|
+
/**
|
|
22
|
+
* Grades the conversation instead of the default model call, so an
|
|
23
|
+
* evaluation model or a scoring service can stand in for the judge. Takes
|
|
24
|
+
* precedence over {@link JudgmentOptions.model}.
|
|
25
|
+
*/
|
|
26
|
+
judge?: (judgment: Judgment) => Verdict | Promise<Verdict>;
|
|
9
27
|
}
|
|
10
28
|
/**
|
|
11
29
|
* Assertions about the tool calls a conversation produced. Kept off vitest's
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"matchers.d.ts","sourceRoot":"","sources":["../../src/matchers/matchers.ts"],"names":[],"mappings":"AAAA,OAAO,QAAQ,CAAC;AAChB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AACxC,OAAO,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAE7D,OAAO,KAAK,EAAE,QAAQ,EAAkB,MAAM,YAAY,CAAC;AAE3D,uDAAuD;AACvD,MAAM,WAAW,eAAe;IAC9B,4EAA4E;IAC5E,KAAK,CAAC,EAAE,aAAa,CAAC;
|
|
1
|
+
{"version":3,"file":"matchers.d.ts","sourceRoot":"","sources":["../../src/matchers/matchers.ts"],"names":[],"mappings":"AAAA,OAAO,QAAQ,CAAC;AAChB,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AACxC,OAAO,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAE7D,OAAO,KAAK,EAAE,QAAQ,EAAkB,MAAM,YAAY,CAAC;AAE3D,iDAAiD;AACjD,MAAM,WAAW,OAAO;IACtB,IAAI,EAAE,OAAO,CAAC;IACd,iFAAiF;IACjF,SAAS,CAAC,EAAE,MAAM,CAAC;CACpB;AAED,kEAAkE;AAClE,MAAM,WAAW,QAAQ;IACvB,QAAQ,EAAE,MAAM,CAAC;IACjB,2EAA2E;IAC3E,UAAU,EAAE,MAAM,CAAC;CACpB;AAED,uDAAuD;AACvD,MAAM,WAAW,eAAe;IAC9B,4EAA4E;IAC5E,KAAK,CAAC,EAAE,aAAa,CAAC;IACtB;;;;OAIG;IACH,KAAK,CAAC,EAAE,CAAC,QAAQ,EAAE,QAAQ,KAAK,OAAO,GAAG,OAAO,CAAC,OAAO,CAAC,CAAC;CAC5D;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY,CAAC,GAAG;IAC/B,oBAAoB,CAAC,IAAI,SAAS,SAAS,CAAC,GAAG,CAAC,EAC9C,IAAI,EAAE,IAAI,EACV,IAAI,CAAC,EAAE,OAAO,CAAC,SAAS,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,GACnC,IAAI,CAAC;IACR,oBAAoB,CAAC,IAAI,SAAS,SAAS,CAAC,GAAG,CAAC,EAC9C,IAAI,EAAE,IAAI,EACV,IAAI,EAAE,OAAO,CAAC,SAAS,CAAC,GAAG,EAAE,IAAI,CAAC,CAAC,GAClC,IAAI,CAAC;IACR,qBAAqB,CAAC,IAAI,SAAS,SAAS,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,IAAI,GAAG,IAAI,CAAC;IACrE,oBAAoB,CAAC,IAAI,SAAS,SAAS,CAAC,GAAG,CAAC,EAAE,IAAI,EAAE,IAAI,GAAG,IAAI,CAAC;IACpE,UAAU,CAAC,IAAI,EAAE,MAAM,GAAG,MAAM,GAAG,IAAI,CAAC;IACxC,wBAAwB,CACtB,GAAG,KAAK,EAAE,CAAC,SAAS,CAAC,GAAG,CAAC,EAAE,GAAG,SAAS,CAAC,GAAG,CAAC,EAAE,CAAC,GAC9C,IAAI,CAAC;IACR,mBAAmB,IAAI,IAAI,CAAC;IAC5B,cAAc,CAAC,QAAQ,EAAE,MAAM,EAAE,OAAO,CAAC,EAAE,eAAe,GAAG,OAAO,CAAC,IAAI,CAAC,CAAC;CAC5E;AAED,MAAM,WAAW,aAAa,CAAC,GAAG,CAAE,SAAQ,YAAY,CAAC,GAAG,CAAC;IAC3D,GAAG,EAAE,YAAY,CAAC,GAAG,CAAC,CAAC;CACxB;AAED,OAAO,QAAQ,QAAQ,CAAC;IACtB,UAAU,YAAY;QACpB,IAAI,CAAC,GAAG,EAAE,IAAI,EAAE,QAAQ,CAAC,GAAG,CAAC,GAAG,aAAa,CAAC,GAAG,CAAC,CAAC;KACpD;CACF"}
|
|
@@ -83,6 +83,25 @@ function report(chat, summary) {
|
|
|
83
83
|
Calls the model actually made:
|
|
84
84
|
${observed(chat)}`;
|
|
85
85
|
}
|
|
86
|
+
async function modelVerdict(chat, criteria, model) {
|
|
87
|
+
const { generateObject, jsonSchema } = await import("ai");
|
|
88
|
+
const { object } = await generateObject({
|
|
89
|
+
model: model ?? chat.model,
|
|
90
|
+
temperature: 0,
|
|
91
|
+
system: JUDGE_SYSTEM,
|
|
92
|
+
prompt: `Criteria:\n${criteria}\n\n<conversation>\n${transcriptFor(chat)}\n</conversation>`,
|
|
93
|
+
schema: jsonSchema({
|
|
94
|
+
type: "object",
|
|
95
|
+
properties: {
|
|
96
|
+
pass: { type: "boolean" },
|
|
97
|
+
reasoning: { type: "string" },
|
|
98
|
+
},
|
|
99
|
+
required: ["pass", "reasoning"],
|
|
100
|
+
additionalProperties: false,
|
|
101
|
+
}),
|
|
102
|
+
});
|
|
103
|
+
return object;
|
|
104
|
+
}
|
|
86
105
|
expect.extend({
|
|
87
106
|
toHaveCalledToolOnce(received, name, args) {
|
|
88
107
|
const matching = acceptedCalls(received, name);
|
|
@@ -164,23 +183,15 @@ ${spoken(received)}`,
|
|
|
164
183
|
async toPassJudgment(received, criteria, options) {
|
|
165
184
|
let verdict;
|
|
166
185
|
try {
|
|
167
|
-
|
|
168
|
-
|
|
169
|
-
|
|
170
|
-
|
|
171
|
-
|
|
172
|
-
|
|
173
|
-
|
|
174
|
-
|
|
175
|
-
|
|
176
|
-
pass: { type: "boolean" },
|
|
177
|
-
reasoning: { type: "string" },
|
|
178
|
-
},
|
|
179
|
-
required: ["pass", "reasoning"],
|
|
180
|
-
additionalProperties: false,
|
|
181
|
-
}),
|
|
182
|
-
});
|
|
183
|
-
verdict = object;
|
|
186
|
+
verdict = options?.judge
|
|
187
|
+
? await options.judge({
|
|
188
|
+
criteria,
|
|
189
|
+
transcript: transcriptFor(received),
|
|
190
|
+
})
|
|
191
|
+
: await modelVerdict(received, criteria, options?.model);
|
|
192
|
+
if (typeof verdict?.pass !== "boolean") {
|
|
193
|
+
throw new Error(`the judge returned ${JSON.stringify(verdict)} instead of a verdict with a boolean "pass"`);
|
|
194
|
+
}
|
|
184
195
|
}
|
|
185
196
|
catch (error) {
|
|
186
197
|
throw new Error(`judge unavailable: ${error instanceof Error ? error.message : String(error)}`);
|
|
@@ -190,8 +201,7 @@ ${spoken(received)}`,
|
|
|
190
201
|
message: () => `expected the conversation ${verdict.pass ? "not " : ""}to pass judgment:
|
|
191
202
|
"${criteria}"
|
|
192
203
|
|
|
193
|
-
judge: ${verdict.pass ? "PASS" : "FAIL"}
|
|
194
|
-
${verdict.reasoning}`,
|
|
204
|
+
judge: ${verdict.pass ? "PASS" : "FAIL"}${verdict.reasoning === undefined ? "" : `\n ${verdict.reasoning}`}`,
|
|
195
205
|
};
|
|
196
206
|
},
|
|
197
207
|
toNeverHaveCalledTool(received, name) {
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"matchers.js","sourceRoot":"","sources":["../../src/matchers/matchers.ts"],"names":[],"mappings":"AAAA,OAAO,QAAQ,CAAC;AAGhB,OAAO,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;
|
|
1
|
+
{"version":3,"file":"matchers.js","sourceRoot":"","sources":["../../src/matchers/matchers.ts"],"names":[],"mappings":"AAAA,OAAO,QAAQ,CAAC;AAGhB,OAAO,EAAE,MAAM,EAAE,MAAM,QAAQ,CAAC;AA+DhC,SAAS,KAAK,CAAC,IAAuB;IACpC,OAAO,IAAI,CAAC,KAAK;SACd,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,MAAM,CAAC;SACtC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC9B,CAAC;AAED,SAAS,WAAW,CAAC,IAAuB;IAC1C,OAAO,IAAI,CAAC,KAAK;SACd,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,KAAK,WAAW,CAAC;SAC3C,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC9B,CAAC;AAED,SAAS,cAAc,CACrB,IAAuB,EACvB,IAAY;IAEZ,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;AAClE,CAAC;AAED,SAAS,WAAW,CAAC,IAAuB;IAC1C,OAAO,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;AACjE,CAAC;AAED,SAAS,aAAa,CACpB,IAAuB,EACvB,IAAY;IAEZ,OAAO,WAAW,CAAC,IAAI,CAAC,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,IAAI,CAAC,CAAC;AACxE,CAAC;AAED,SAAS,QAAQ,CAAC,IAAuB;IACvC,MAAM,IAAI,GAAG,KAAK,CAAC,IAAI,CAAC,CAAC;IACzB,IAAI,IAAI,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACtB,OAAO,wBAAwB,CAAC;IAClC,CAAC;IACD,OAAO,IAAI;SACR,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,QAAQ,CAAC,IAAY;IAC5B,OAAO,IAAI,CAAC,OAAO,CAAC,MAAM,EAAE,GAAG,CAAC,CAAC,IAAI,EAAE,CAAC;AAC1C,CAAC;AAED,SAAS,IAAI,CAAC,IAAuB,EAAE,IAAqB;IAC1D,OAAO,WAAW,CAAC,IAAI,CAAC,CAAC,IAAI,CAAC,CAAC,IAAI,EAAE,EAAE,CACrC,OAAO,IAAI,KAAK,QAAQ;QACtB,CAAC,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,CAAC,QAAQ,CAAC,QAAQ,CAAC,IAAI,CAAC,CAAC,WAAW,EAAE,CAAC;QACrE,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CACpB,CAAC;AACJ,CAAC;AAED,SAAS,MAAM,CAAC,IAAuB;IACrC,MAAM,KAAK,GAAG,WAAW,CAAC,IAAI,CAAC,CAAC;IAChC,IAAI,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QACvB,OAAO,gCAAgC,CAAC;IAC1C,CAAC;IACD,OAAO,KAAK;SACT,GAAG,CAAC,CAAC,IAAI,EAAE,KAAK,EAAE,EAAE,CAAC,KAAK,KAAK,GAAG,CAAC,KAAK,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;SACzD,IAAI,CAAC,IAAI,CAAC,CAAC;AAChB,CAAC;AAED;;;GAGG;AACH,SAAS,KAAK,CAAC,IAAY;IACzB,OAAO,IAAI,CAAC,OAAO,CAAC,uBAAuB,EAAE,QAAQ,CAAC,CAAC;AACzD,CAAC;AAED,MAAM,YAAY,GAAG;;;6KAGwJ,CAAC;AAE9K,MAAM,gBAAgB,GAAG,IAAI,CAAC;AAE9B,SAAS,QAAQ,CAAC,IAAmB;IACnC,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACzB,OAAO,GAAG,IAAI,CAAC,IAAI,KAAK,KAAK,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;IAC7C,CAAC;IACD,MAAM,OAAO,GACX,IAAI,CAAC,IAAI,CAAC,MAAM,KAAK,SAAS;QAC5B,CAAC,CAAC,OAAO,QAAQ,CAAC,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,MAAM,IAAI,IAAI,CAAC,CAAC,EAAE;QACxD,CAAC,CAAC,aAAa,IAAI,CAAC,IAAI,CAAC,MAAM,GAAG,CAAC;IACvC,OAAO,SAAS,KAAK,CAAC,GAAG,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,IAAI,CAAC,SAAS,CAAC,IAAI,CAAC,IAAI,CAAC,SAAS,CAAC,GAAG,OAAO,EAAE,CAAC,EAAE,CAAC;AACxG,CAAC;AAED,SAAS,QAAQ,CAAC,IAAY;IAC5B,OAAO,IAAI,CAAC,MAAM,GAAG,gBAAgB;QACnC,CAAC,CAAC,GAAG,IAAI,CAAC,KAAK,CAAC,CAAC,EAAE,gBAAgB,CAAC,eAAe;QACnD,CAAC,CAAC,IAAI,CAAC;AACX,CAAC;AAED,SAAS,aAAa,CAAC,IAAuB;IAC5C,IAAI,IAAI,CAAC,KAAK,CAAC,MAAM,KAAK,CAAC,EAAE,CAAC;QAC5B,OAAO,6BAA6B,CAAC;IACvC,CAAC;IACD,OAAO,IAAI,CAAC,KAAK,CAAC,GAAG,CAAC,QAAQ,CAAC,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;AAC7C,CAAC;AAED,SAAS,MAAM,CAAC,IAAuB,EAAE,OAAe;IACtD,OAAO,GAAG,OAAO;;;EAGjB,QAAQ,CAAC,IAAI,CAAC,EAAE,CAAC;AACnB,CAAC;AAED,KAAK,UAAU,YAAY,CACzB,IAAuB,EACvB,QAAgB,EAChB,KAAqB;IAErB,MAAM,EAAE,cAAc,EAAE,UAAU,EAAE,GAAG,MAAM,MAAM,CAAC,IAAI,CAAC,CAAC;IAC1D,MAAM,EAAE,MAAM,EAAE,GAAG,MAAM,cAAc,CAAC;QACtC,KAAK,EAAE,KAAK,IAAI,IAAI,CAAC,KAAK;QAC1B,WAAW,EAAE,CAAC;QACd,MAAM,EAAE,YAAY;QACpB,MAAM,EAAE,cAAc,QAAQ,uBAAuB,aAAa,CAAC,IAAI,CAAC,mBAAmB;QAC3F,MAAM,EAAE,UAAU,CAAkC;YAClD,IAAI,EAAE,QAAQ;YACd,UAAU,EAAE;gBACV,IAAI,EAAE,EAAE,IAAI,EAAE,SAAS,EAAE;gBACzB,SAAS,EAAE,EAAE,IAAI,EAAE,QAAQ,EAAE;aAC9B;YACD,QAAQ,EAAE,CAAC,MAAM,EAAE,WAAW,CAAC;YAC/B,oBAAoB,EAAE,KAAK;SAC5B,CAAC;KACH,CAAC,CAAC;IACH,OAAO,MAAM,CAAC;AAChB,CAAC;AAED,MAAM,CAAC,MAAM,CAAC;IACZ,oBAAoB,CAClB,QAA2B,EAC3B,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,QAA2B,EAC3B,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,oBAAoB,CAAC,QAA2B,EAAE,IAAY;QAC5D,MAAM,SAAS,GAAG,cAAc,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QACjD,MAAM,MAAM,GAAG,SAAS,CAAC,MAAM,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,CAAC,MAAM,KAAK,SAAS,CAAC,CAAC;QAErE,OAAO;YACL,IAAI,EAAE,MAAM,CAAC,MAAM,GAAG,CAAC;YACvB,OAAO,EAAE,GAAG,EAAE,CACZ,MAAM,CACJ,QAAQ,EACR,MAAM,CAAC,MAAM,GAAG,CAAC;gBACf,CAAC,CAAC,2BAA2B,IAAI,eAAe;gBAChD,CAAC,CAAC,SAAS,CAAC,MAAM,KAAK,CAAC;oBACtB,CAAC,CAAC,aAAa,IAAI,mDAAmD;oBACtE,CAAC,CAAC,aAAa,IAAI,sBAAsB,SAAS,CAAC,MAAM,mBAAmB,CACjF;SACJ,CAAC;IACJ,CAAC;IAED,UAAU,CAAC,QAA2B,EAAE,IAAqB;QAC3D,MAAM,IAAI,GAAG,IAAI,CAAC,QAAQ,EAAE,IAAI,CAAC,CAAC;QAClC,MAAM,MAAM,GAAG,OAAO,IAAI,KAAK,QAAQ,CAAC,CAAC,CAAC,IAAI,IAAI,GAAG,CAAC,CAAC,CAAC,MAAM,CAAC,IAAI,CAAC,CAAC;QAErE,OAAO;YACL,IAAI;YACJ,OAAO,EAAE,GAAG,EAAE,CACZ,GACE,IAAI;gBACF,CAAC,CAAC,qCAAqC,MAAM,GAAG;gBAChD,CAAC,CAAC,iCAAiC,MAAM,GAC7C;;;EAGN,MAAM,CAAC,QAAQ,CAAC,EAAE;SACf,CAAC;IACJ,CAAC;IAED,mBAAmB,CAAC,QAA2B;QAC7C,MAAM,KAAK,GAAG,KAAK,CAAC,QAAQ,CAAC,CAAC,MAAM,CAAC;QAErC,OAAO;YACL,IAAI,EAAE,KAAK,KAAK,CAAC;YACjB,OAAO,EAAE,GAAG,EAAE,CACZ,MAAM,CACJ,QAAQ,EACR,KAAK,KAAK,CAAC;gBACT,CAAC,CAAC,+CAA+C;gBACjD,CAAC,CAAC,wDAAwD,KAAK,IAAI,KAAK,KAAK,CAAC,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,OAAO,GAAG,CACvG;SACJ,CAAC;IACJ,CAAC;IAED,wBAAwB,CAAC,QAA2B,EAAE,GAAG,KAAe;QACtE,IAAI,OAAO,GAAG,CAAC,CAAC;QAChB,KAAK,MAAM,IAAI,IAAI,WAAW,CAAC,QAAQ,CAAC,EAAE,CAAC;YACzC,IAAI,MAAM,CAAC,IAAI,CAAC,IAAI,CAAC,KAAK,KAAK,CAAC,OAAO,CAAC,EAAE,CAAC;gBACzC,OAAO,IAAI,CAAC,CAAC;YACf,CAAC;YACD,IAAI,OAAO,KAAK,KAAK,CAAC,MAAM,EAAE,CAAC;gBAC7B,MAAM;YACR,CAAC;QACH,CAAC;QACD,MAAM,IAAI,GAAG,OAAO,KAAK,KAAK,CAAC,MAAM,CAAC;QACtC,MAAM,MAAM,GAAG,KAAK,CAAC,GAAG,CAAC,CAAC,IAAI,EAAE,EAAE,CAAC,IAAI,IAAI,GAAG,CAAC,CAAC,IAAI,CAAC,QAAQ,CAAC,CAAC;QAE/D,OAAO;YACL,IAAI;YACJ,OAAO,EAAE,GAAG,EAAE,CACZ,MAAM,CACJ,QAAQ,EACR,IAAI;gBACF,CAAC,CAAC,YAAY,MAAM,kCAAkC;gBACtD,CAAC,CAAC,YAAY,MAAM,wBAAwB,KAAK,CAAC,OAAO,CAAC,eAAe,OAAO,KAAK,CAAC,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,WAAW,KAAK,CAAC,OAAO,GAAG,CAAC,CAAC,GAAG,GAAG,CACpI;SACJ,CAAC;IACJ,CAAC;IAED,KAAK,CAAC,cAAc,CAClB,QAA2B,EAC3B,QAAgB,EAChB,OAAyB;QAEzB,IAAI,OAAgB,CAAC;QACrB,IAAI,CAAC;YACH,OAAO,GAAG,OAAO,EAAE,KAAK;gBACtB,CAAC,CAAC,MAAM,OAAO,CAAC,KAAK,CAAC;oBAClB,QAAQ;oBACR,UAAU,EAAE,aAAa,CAAC,QAAQ,CAAC;iBACpC,CAAC;gBACJ,CAAC,CAAC,MAAM,YAAY,CAAC,QAAQ,EAAE,QAAQ,EAAE,OAAO,EAAE,KAAK,CAAC,CAAC;YAE3D,IAAI,OAAO,OAAO,EAAE,IAAI,KAAK,SAAS,EAAE,CAAC;gBACvC,MAAM,IAAI,KAAK,CACb,sBAAsB,IAAI,CAAC,SAAS,CAAC,OAAO,CAAC,6CAA6C,CAC3F,CAAC;YACJ,CAAC;QACH,CAAC;QAAC,OAAO,KAAK,EAAE,CAAC;YACf,MAAM,IAAI,KAAK,CACb,sBAAsB,KAAK,YAAY,KAAK,CAAC,CAAC,CAAC,KAAK,CAAC,OAAO,CAAC,CAAC,CAAC,MAAM,CAAC,KAAK,CAAC,EAAE,CAC/E,CAAC;QACJ,CAAC;QAED,OAAO;YACL,IAAI,EAAE,OAAO,CAAC,IAAI;YAClB,OAAO,EAAE,GAAG,EAAE,CACZ,6BAA6B,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,EAAE;KAC1D,QAAQ;;SAEJ,OAAO,CAAC,IAAI,CAAC,CAAC,CAAC,MAAM,CAAC,CAAC,CAAC,MAAM,GAAG,OAAO,CAAC,SAAS,KAAK,SAAS,CAAC,CAAC,CAAC,EAAE,CAAC,CAAC,CAAC,OAAO,OAAO,CAAC,SAAS,EAAE,EAAE;SACxG,CAAC;IACJ,CAAC;IAED,qBAAqB,CAAC,QAA2B,EAAE,IAAY;QAC7D,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,CAAM,IAAmB,EAAsB,EAAE,CAC7D,MAAM,CAAC,IAAI,CAAkC,CAAC"}
|
|
@@ -1,5 +1,6 @@
|
|
|
1
1
|
import { type Server } from "@modelcontextprotocol/server";
|
|
2
2
|
import type { LanguageModel } from "ai";
|
|
3
|
+
import type { ToolInput, ToolNames } from "skybridge/server";
|
|
3
4
|
import { Chat } from "./chat.js";
|
|
4
5
|
/**
|
|
5
6
|
* The minimum a Skybridge app has to expose to be served in-process: the
|
|
@@ -26,6 +27,15 @@ export interface EvalIdentity {
|
|
|
26
27
|
resource?: URL;
|
|
27
28
|
extra?: Record<string, unknown>;
|
|
28
29
|
}
|
|
30
|
+
/**
|
|
31
|
+
* Results the runner answers with instead of calling the app, keyed by tool
|
|
32
|
+
* name and typed against the project's own registry. A stub may be async, and
|
|
33
|
+
* one that returns `undefined` falls through to the real server, so a scenario
|
|
34
|
+
* can pin one set of arguments and leave the rest live.
|
|
35
|
+
*/
|
|
36
|
+
export type ToolStubs<App> = {
|
|
37
|
+
[Name in ToolNames<App>]?: (args: ToolInput<App, Name>) => unknown;
|
|
38
|
+
};
|
|
29
39
|
interface StartOptions {
|
|
30
40
|
/** Any AI SDK model instance, built by the project. */
|
|
31
41
|
model: LanguageModel;
|
|
@@ -43,6 +53,11 @@ interface StartOptions {
|
|
|
43
53
|
* The assertions are inferred from the app value itself, so `expect.chat` gets
|
|
44
54
|
* the project's tool names and argument shapes with no type parameter.
|
|
45
55
|
*
|
|
56
|
+
* `stubs` answers a tool from the scenario instead of the app, so a case that
|
|
57
|
+
* depends on today's date or on a live catalogue still reads the same weeks
|
|
58
|
+
* later. A stubbed call never reaches the server, and still shows up in
|
|
59
|
+
* `chat.toolCalls` like any other. See {@link ToolStubs}.
|
|
60
|
+
*
|
|
46
61
|
* `authInfo` claims an identity for the session: the app's per-tool scheme and
|
|
47
62
|
* scope enforcement runs against it for real, only token verification is
|
|
48
63
|
* skipped. Omit it to exercise the anonymous path, challenges included.
|
|
@@ -50,6 +65,7 @@ interface StartOptions {
|
|
|
50
65
|
export declare function start<App extends EvalApp>(options: StartOptions & {
|
|
51
66
|
app: App;
|
|
52
67
|
authInfo?: EvalIdentity;
|
|
68
|
+
stubs?: ToolStubs<App>;
|
|
53
69
|
}): Promise<Chat<App>>;
|
|
54
70
|
export {};
|
|
55
71
|
//# sourceMappingURL=session-registry.d.ts.map
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAoB,KAAK,MAAM,EAAE,MAAM,8BAA8B,CAAC;AAC7E,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;
|
|
1
|
+
{"version":3,"file":"session-registry.d.ts","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAoB,KAAK,MAAM,EAAE,MAAM,8BAA8B,CAAC;AAC7E,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,IAAI,CAAC;AACxC,OAAO,KAAK,EAAE,SAAS,EAAE,SAAS,EAAE,MAAM,kBAAkB,CAAC;AAE7D,OAAO,EAAE,IAAI,EAAgB,MAAM,WAAW,CAAC;AAE/C;;;;;GAKG;AACH,MAAM,WAAW,OAAO;IACtB,QAAQ,CAAC,MAAM,EAAE;QAAE,QAAQ,CAAC,KAAK,EAAE,MAAM,CAAA;KAAE,CAAC;IAC5C,oBAAoB,IAAI,OAAO,CAAC,MAAM,CAAC,CAAC;CACzC;AAED;;;;GAIG;AACH,MAAM,WAAW,YAAY;IAC3B,KAAK,EAAE,MAAM,CAAC;IACd,QAAQ,EAAE,MAAM,CAAC;IACjB,MAAM,EAAE,MAAM,EAAE,CAAC;IACjB,SAAS,CAAC,EAAE,MAAM,CAAC;IACnB,QAAQ,CAAC,EAAE,GAAG,CAAC;IACf,KAAK,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAC;CACjC;AAED;;;;;GAKG;AACH,MAAM,MAAM,SAAS,CAAC,GAAG,IAAI;KAC1B,IAAI,IAAI,SAAS,CAAC,GAAG,CAAC,CAAC,CAAC,EAAE,CAAC,IAAI,EAAE,SAAS,CAAC,GAAG,EAAE,IAAI,CAAC,KAAK,OAAO;CACnE,CAAC;AAEF,UAAU,YAAY;IACpB,uDAAuD;IACvD,KAAK,EAAE,aAAa,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,QAAQ,CAAC,EAAE,MAAM,CAAC;CACnB;AAUD;;;;;;;;;;;;;;;;;;GAkBG;AACH,wBAAsB,KAAK,CAAC,GAAG,SAAS,OAAO,EAC7C,OAAO,EAAE,YAAY,GAAG;IACtB,GAAG,EAAE,GAAG,CAAC;IACT,QAAQ,CAAC,EAAE,YAAY,CAAC;IACxB,KAAK,CAAC,EAAE,SAAS,CAAC,GAAG,CAAC,CAAC;CACxB,GACA,OAAO,CAAC,IAAI,CAAC,GAAG,CAAC,CAAC,CAyBpB"}
|
package/dist/session-registry.js
CHANGED
|
@@ -19,6 +19,11 @@ function sharedDefaults() {
|
|
|
19
19
|
* The assertions are inferred from the app value itself, so `expect.chat` gets
|
|
20
20
|
* the project's tool names and argument shapes with no type parameter.
|
|
21
21
|
*
|
|
22
|
+
* `stubs` answers a tool from the scenario instead of the app, so a case that
|
|
23
|
+
* depends on today's date or on a live catalogue still reads the same weeks
|
|
24
|
+
* later. A stubbed call never reaches the server, and still shows up in
|
|
25
|
+
* `chat.toolCalls` like any other. See {@link ToolStubs}.
|
|
26
|
+
*
|
|
22
27
|
* `authInfo` claims an identity for the session: the app's per-tool scheme and
|
|
23
28
|
* scope enforcement runs against it for real, only token verification is
|
|
24
29
|
* skipped. Omit it to exercise the anonymous path, challenges included.
|
|
@@ -32,7 +37,7 @@ export async function start(options) {
|
|
|
32
37
|
temperature: options.temperature ?? config?.temperature,
|
|
33
38
|
systemPrompt: options.systemPrompt ?? config?.systemPrompt,
|
|
34
39
|
maxSteps: options.maxSteps ?? config?.maxSteps,
|
|
35
|
-
}, (url, init) => handler.fetch(new Request(url, init), authInfo === undefined ? undefined : { authInfo }));
|
|
40
|
+
}, (url, init) => handler.fetch(new Request(url, init), authInfo === undefined ? undefined : { authInfo }), (options.stubs ?? {}));
|
|
36
41
|
onTestFinished(async () => {
|
|
37
42
|
await chat.close();
|
|
38
43
|
await handler.close();
|
|
@@ -1 +1 @@
|
|
|
1
|
-
{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAe,MAAM,8BAA8B,CAAC;
|
|
1
|
+
{"version":3,"file":"session-registry.js","sourceRoot":"","sources":["../src/session-registry.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,gBAAgB,EAAe,MAAM,8BAA8B,CAAC;AAG7E,OAAO,EAAE,MAAM,EAAE,cAAc,EAAE,MAAM,QAAQ,CAAC;AAChD,OAAO,EAAE,IAAI,EAAgB,MAAM,WAAW,CAAC;AA6C/C,SAAS,cAAc;IACrB,IAAI,CAAC;QACH,OAAO,MAAM,CAAC,gBAAgB,CAAC,CAAC;IAClC,CAAC;IAAC,MAAM,CAAC;QACP,OAAO,SAAS,CAAC;IACnB,CAAC;AACH,CAAC;AAED;;;;;;;;;;;;;;;;;;GAkBG;AACH,MAAM,CAAC,KAAK,UAAU,KAAK,CACzB,OAIC;IAED,MAAM,MAAM,GAAG,cAAc,EAAE,CAAC;IAChC,MAAM,EAAE,GAAG,EAAE,QAAQ,EAAE,GAAG,OAAO,CAAC;IAClC,MAAM,OAAO,GAAG,gBAAgB,CAAC,GAAG,EAAE,CAAC,GAAG,CAAC,oBAAoB,EAAE,CAAC,CAAC;IAEnE,MAAM,IAAI,GAAG,MAAM,IAAI,CAAC,IAAI,CAC1B;QACE,KAAK,EAAE,OAAO,CAAC,KAAK;QACpB,WAAW,EAAE,OAAO,CAAC,WAAW,IAAI,MAAM,EAAE,WAAW;QACvD,YAAY,EAAE,OAAO,CAAC,YAAY,IAAI,MAAM,EAAE,YAAY;QAC1D,QAAQ,EAAE,OAAO,CAAC,QAAQ,IAAI,MAAM,EAAE,QAAQ;KAC/C,EACD,CAAC,GAAG,EAAE,IAAI,EAAE,EAAE,CACZ,OAAO,CAAC,KAAK,CACX,IAAI,OAAO,CAAC,GAAG,EAAE,IAAI,CAAC,EACtB,QAAQ,KAAK,SAAS,CAAC,CAAC,CAAC,SAAS,CAAC,CAAC,CAAC,EAAE,QAAQ,EAAE,CAClD,EACH,CAAC,OAAO,CAAC,KAAK,IAAI,EAAE,CAAY,CACjC,CAAC;IAEF,cAAc,CAAC,KAAK,IAAI,EAAE;QACxB,MAAM,IAAI,CAAC,KAAK,EAAE,CAAC;QACnB,MAAM,OAAO,CAAC,KAAK,EAAE,CAAC;IACxB,CAAC,CAAC,CAAC;IACH,OAAO,IAAI,CAAC;AACd,CAAC"}
|
package/package.json
CHANGED
|
@@ -1,7 +1,7 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "@skybridge/test",
|
|
3
|
-
"version": "0.0.0-dev.
|
|
4
|
-
"description": "Eval runner for Skybridge MCP
|
|
3
|
+
"version": "0.0.0-dev.ec70681",
|
|
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",
|
|
7
7
|
"repository": {
|
|
@@ -37,8 +37,8 @@
|
|
|
37
37
|
"typescript": "^5.9.3",
|
|
38
38
|
"vitest": "^4.1.10",
|
|
39
39
|
"zod": "^4.4.3",
|
|
40
|
-
"@skybridge/vite-plugin": "0.0.0-dev.
|
|
41
|
-
"skybridge": "0.0.0-dev.
|
|
40
|
+
"@skybridge/vite-plugin": "0.0.0-dev.ec70681",
|
|
41
|
+
"skybridge": "0.0.0-dev.ec70681"
|
|
42
42
|
},
|
|
43
43
|
"engines": {
|
|
44
44
|
"node": ">=22.23.1"
|