@hasna/evals 0.1.22 → 0.1.23
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/cli/index.js +1775 -324
- package/dist/index.js +3764 -898
- package/dist/mcp/index.js +3450 -1437
- package/dist/server/index.js +3764 -898
- package/package.json +1 -1
- package/dist/adapters/adapters.test.d.ts +0 -2
- package/dist/adapters/adapters.test.d.ts.map +0 -1
- package/dist/adapters/anthropic-openai.test.d.ts +0 -2
- package/dist/adapters/anthropic-openai.test.d.ts.map +0 -1
- package/dist/adapters/anthropic.d.ts +0 -4
- package/dist/adapters/anthropic.d.ts.map +0 -1
- package/dist/adapters/cli.d.ts +0 -4
- package/dist/adapters/cli.d.ts.map +0 -1
- package/dist/adapters/function.d.ts +0 -4
- package/dist/adapters/function.d.ts.map +0 -1
- package/dist/adapters/http-cli-coverage.test.d.ts +0 -2
- package/dist/adapters/http-cli-coverage.test.d.ts.map +0 -1
- package/dist/adapters/http.d.ts +0 -16
- package/dist/adapters/http.d.ts.map +0 -1
- package/dist/adapters/mcp-adapter.test.d.ts +0 -2
- package/dist/adapters/mcp-adapter.test.d.ts.map +0 -1
- package/dist/adapters/mcp.d.ts +0 -4
- package/dist/adapters/mcp.d.ts.map +0 -1
- package/dist/adapters/openai.d.ts +0 -4
- package/dist/adapters/openai.d.ts.map +0 -1
- package/dist/cli/adapter-parser.d.ts +0 -3
- package/dist/cli/adapter-parser.d.ts.map +0 -1
- package/dist/cli/cli.test.d.ts +0 -2
- package/dist/cli/cli.test.d.ts.map +0 -1
- package/dist/cli/commands/calibrate.d.ts +0 -3
- package/dist/cli/commands/calibrate.d.ts.map +0 -1
- package/dist/cli/commands/capture.d.ts +0 -3
- package/dist/cli/commands/capture.d.ts.map +0 -1
- package/dist/cli/commands/ci.d.ts +0 -3
- package/dist/cli/commands/ci.d.ts.map +0 -1
- package/dist/cli/commands/compare.d.ts +0 -5
- package/dist/cli/commands/compare.d.ts.map +0 -1
- package/dist/cli/commands/compare.test.d.ts +0 -2
- package/dist/cli/commands/compare.test.d.ts.map +0 -1
- package/dist/cli/commands/completion.d.ts +0 -3
- package/dist/cli/commands/completion.d.ts.map +0 -1
- package/dist/cli/commands/doctor.d.ts +0 -3
- package/dist/cli/commands/doctor.d.ts.map +0 -1
- package/dist/cli/commands/estimate.d.ts +0 -3
- package/dist/cli/commands/estimate.d.ts.map +0 -1
- package/dist/cli/commands/generate.d.ts +0 -3
- package/dist/cli/commands/generate.d.ts.map +0 -1
- package/dist/cli/commands/judge.d.ts +0 -3
- package/dist/cli/commands/judge.d.ts.map +0 -1
- package/dist/cli/commands/mcp.d.ts +0 -3
- package/dist/cli/commands/mcp.d.ts.map +0 -1
- package/dist/cli/commands/run.d.ts +0 -3
- package/dist/cli/commands/run.d.ts.map +0 -1
- package/dist/cli/commands/sync.d.ts +0 -3
- package/dist/cli/commands/sync.d.ts.map +0 -1
- package/dist/cli/index.d.ts +0 -3
- package/dist/cli/index.d.ts.map +0 -1
- package/dist/core/assertions-coverage.test.d.ts +0 -2
- package/dist/core/assertions-coverage.test.d.ts.map +0 -1
- package/dist/core/assertions.d.ts +0 -18
- package/dist/core/assertions.d.ts.map +0 -1
- package/dist/core/assertions.test.d.ts +0 -2
- package/dist/core/assertions.test.d.ts.map +0 -1
- package/dist/core/e2e.test.d.ts +0 -2
- package/dist/core/e2e.test.d.ts.map +0 -1
- package/dist/core/judge.d.ts +0 -13
- package/dist/core/judge.d.ts.map +0 -1
- package/dist/core/judge.test.d.ts +0 -2
- package/dist/core/judge.test.d.ts.map +0 -1
- package/dist/core/reporter.d.ts +0 -21
- package/dist/core/reporter.d.ts.map +0 -1
- package/dist/core/reporter.test.d.ts +0 -2
- package/dist/core/reporter.test.d.ts.map +0 -1
- package/dist/core/runner.d.ts +0 -4
- package/dist/core/runner.d.ts.map +0 -1
- package/dist/core/runner.test.d.ts +0 -2
- package/dist/core/runner.test.d.ts.map +0 -1
- package/dist/datasets/loader.d.ts +0 -18
- package/dist/datasets/loader.d.ts.map +0 -1
- package/dist/datasets/loader.test.d.ts +0 -2
- package/dist/datasets/loader.test.d.ts.map +0 -1
- package/dist/db/store.d.ts +0 -17
- package/dist/db/store.d.ts.map +0 -1
- package/dist/db/store.test.d.ts +0 -2
- package/dist/db/store.test.d.ts.map +0 -1
- package/dist/index.d.ts +0 -8
- package/dist/index.d.ts.map +0 -1
- package/dist/mcp/http.d.ts +0 -12
- package/dist/mcp/http.d.ts.map +0 -1
- package/dist/mcp/http.test.d.ts +0 -2
- package/dist/mcp/http.test.d.ts.map +0 -1
- package/dist/mcp/index.d.ts +0 -3
- package/dist/mcp/index.d.ts.map +0 -1
- package/dist/mcp/mcp.test.d.ts +0 -2
- package/dist/mcp/mcp.test.d.ts.map +0 -1
- package/dist/mcp/server.d.ts +0 -5
- package/dist/mcp/server.d.ts.map +0 -1
- package/dist/server/index.d.ts +0 -3
- package/dist/server/index.d.ts.map +0 -1
- package/dist/server/server.test.d.ts +0 -2
- package/dist/server/server.test.d.ts.map +0 -1
- package/dist/types/index.d.ts +0 -171
- package/dist/types/index.d.ts.map +0 -1
package/package.json
CHANGED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"adapters.test.d.ts","sourceRoot":"","sources":["../../src/adapters/adapters.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"anthropic-openai.test.d.ts","sourceRoot":"","sources":["../../src/adapters/anthropic-openai.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,4 +0,0 @@
|
|
|
1
|
-
import type { AnthropicAdapterConfig, ConversationTurn } from "../types/index.js";
|
|
2
|
-
import type { AdapterResponse } from "./http.js";
|
|
3
|
-
export declare function callAnthropicAdapter(config: AnthropicAdapterConfig, input: string, turns?: ConversationTurn[]): Promise<AdapterResponse>;
|
|
4
|
-
//# sourceMappingURL=anthropic.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"anthropic.d.ts","sourceRoot":"","sources":["../../src/adapters/anthropic.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,sBAAsB,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAClF,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAEjD,wBAAsB,oBAAoB,CACxC,MAAM,EAAE,sBAAsB,EAC9B,KAAK,EAAE,MAAM,EACb,KAAK,CAAC,EAAE,gBAAgB,EAAE,GACzB,OAAO,CAAC,eAAe,CAAC,CA+C1B"}
|
package/dist/adapters/cli.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"cli.d.ts","sourceRoot":"","sources":["../../src/adapters/cli.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAC1D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAEjD,wBAAsB,cAAc,CAClC,MAAM,EAAE,gBAAgB,EACxB,KAAK,EAAE,MAAM,GACZ,OAAO,CAAC,eAAe,CAAC,CA4C1B"}
|
|
@@ -1,4 +0,0 @@
|
|
|
1
|
-
import type { FunctionAdapterConfig } from "../types/index.js";
|
|
2
|
-
import type { AdapterResponse } from "./http.js";
|
|
3
|
-
export declare function callFunctionAdapter(config: FunctionAdapterConfig, input: string): Promise<AdapterResponse>;
|
|
4
|
-
//# sourceMappingURL=function.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"function.d.ts","sourceRoot":"","sources":["../../src/adapters/function.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,qBAAqB,EAAE,MAAM,mBAAmB,CAAC;AAC/D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAEjD,wBAAsB,mBAAmB,CACvC,MAAM,EAAE,qBAAqB,EAC7B,KAAK,EAAE,MAAM,GACZ,OAAO,CAAC,eAAe,CAAC,CA2B1B"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"http-cli-coverage.test.d.ts","sourceRoot":"","sources":["../../src/adapters/http-cli-coverage.test.ts"],"names":[],"mappings":""}
|
package/dist/adapters/http.d.ts
DELETED
|
@@ -1,16 +0,0 @@
|
|
|
1
|
-
import type { HttpAdapterConfig, ConversationTurn } from "../types/index.js";
|
|
2
|
-
export interface AdapterResponse {
|
|
3
|
-
output: string;
|
|
4
|
-
durationMs: number;
|
|
5
|
-
inputTokens?: number;
|
|
6
|
-
outputTokens?: number;
|
|
7
|
-
costUsd?: number;
|
|
8
|
-
toolCalls?: Array<{
|
|
9
|
-
name: string;
|
|
10
|
-
arguments?: Record<string, unknown>;
|
|
11
|
-
}>;
|
|
12
|
-
statusCode?: number;
|
|
13
|
-
error?: string;
|
|
14
|
-
}
|
|
15
|
-
export declare function callHttpAdapter(config: HttpAdapterConfig, input: string, turns?: ConversationTurn[]): Promise<AdapterResponse>;
|
|
16
|
-
//# sourceMappingURL=http.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"http.d.ts","sourceRoot":"","sources":["../../src/adapters/http.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,iBAAiB,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAE7E,MAAM,WAAW,eAAe;IAC9B,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,EAAE,MAAM,CAAC;IACnB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAC,CAAC;IACzE,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,KAAK,CAAC,EAAE,MAAM,CAAC;CAChB;AA+BD,wBAAsB,eAAe,CACnC,MAAM,EAAE,iBAAiB,EACzB,KAAK,EAAE,MAAM,EACb,KAAK,CAAC,EAAE,gBAAgB,EAAE,GACzB,OAAO,CAAC,eAAe,CAAC,CAqE1B"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"mcp-adapter.test.d.ts","sourceRoot":"","sources":["../../src/adapters/mcp-adapter.test.ts"],"names":[],"mappings":""}
|
package/dist/adapters/mcp.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"mcp.d.ts","sourceRoot":"","sources":["../../src/adapters/mcp.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAC1D,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAEjD,wBAAsB,cAAc,CAClC,MAAM,EAAE,gBAAgB,EACxB,KAAK,EAAE,MAAM,GACZ,OAAO,CAAC,eAAe,CAAC,CAqD1B"}
|
|
@@ -1,4 +0,0 @@
|
|
|
1
|
-
import type { OpenAIAdapterConfig, ConversationTurn } from "../types/index.js";
|
|
2
|
-
import type { AdapterResponse } from "./http.js";
|
|
3
|
-
export declare function callOpenAIAdapter(config: OpenAIAdapterConfig, input: string, turns?: ConversationTurn[]): Promise<AdapterResponse>;
|
|
4
|
-
//# sourceMappingURL=openai.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"openai.d.ts","sourceRoot":"","sources":["../../src/adapters/openai.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,mBAAmB,EAAE,gBAAgB,EAAE,MAAM,mBAAmB,CAAC;AAC/E,OAAO,KAAK,EAAE,eAAe,EAAE,MAAM,WAAW,CAAC;AAEjD,wBAAsB,iBAAiB,CACrC,MAAM,EAAE,mBAAmB,EAC3B,KAAK,EAAE,MAAM,EACb,KAAK,CAAC,EAAE,gBAAgB,EAAE,GACzB,OAAO,CAAC,eAAe,CAAC,CAuD1B"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"adapter-parser.d.ts","sourceRoot":"","sources":["../../src/cli/adapter-parser.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,aAAa,EAAE,MAAM,mBAAmB,CAAC;AAEvD,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,MAAM,CAAC,MAAM,EAAE,MAAM,CAAC,GAAG,aAAa,CA8B9E"}
|
package/dist/cli/cli.test.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"cli.test.d.ts","sourceRoot":"","sources":["../../src/cli/cli.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"calibrate.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/calibrate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAKpC,wBAAgB,gBAAgB,IAAI,OAAO,CA+C1C"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"capture.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/capture.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAKpC,wBAAgB,cAAc,IAAI,OAAO,CAyExC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"ci.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/ci.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAOpC,wBAAgB,SAAS,IAAI,OAAO,CAiEnC"}
|
|
@@ -1,5 +0,0 @@
|
|
|
1
|
-
import { Command } from "commander";
|
|
2
|
-
import { compareRuns } from "../../core/reporter.js";
|
|
3
|
-
export declare function renderMarkdownDiff(diff: ReturnType<typeof compareRuns>): string;
|
|
4
|
-
export declare function compareCommand(): Command;
|
|
5
|
-
//# sourceMappingURL=compare.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"compare.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/compare.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AACpC,OAAO,EAAE,WAAW,EAA+B,MAAM,wBAAwB,CAAC;AAGlF,wBAAgB,kBAAkB,CAAC,IAAI,EAAE,UAAU,CAAC,OAAO,WAAW,CAAC,GAAG,MAAM,CA2B/E;AAED,wBAAgB,cAAc,IAAI,OAAO,CAgCxC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"compare.test.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/compare.test.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"completion.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/completion.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAiFpC,wBAAgB,iBAAiB,IAAI,OAAO,CAgB3C"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"doctor.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/doctor.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAqCpC,wBAAgB,aAAa,IAAI,OAAO,CAqFvC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"estimate.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/estimate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAapC,wBAAgB,eAAe,IAAI,OAAO,CA2CzC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"generate.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/generate.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAiBpC,wBAAgB,eAAe,IAAI,OAAO,CAyDzC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/judge.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAGpC,wBAAgB,YAAY,IAAI,OAAO,CA+BtC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"mcp.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/mcp.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAKpC,wBAAgB,UAAU,IAAI,OAAO,CAgCpC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"run.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/run.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAOpC,wBAAgB,UAAU,IAAI,OAAO,CAyDpC"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"sync.d.ts","sourceRoot":"","sources":["../../../src/cli/commands/sync.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,OAAO,EAAE,MAAM,WAAW,CAAC;AAEpC,wBAAgB,WAAW,IAAI,OAAO,CA2FrC"}
|
package/dist/cli/index.d.ts
DELETED
package/dist/cli/index.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/cli/index.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"assertions-coverage.test.d.ts","sourceRoot":"","sources":["../../src/core/assertions-coverage.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,18 +0,0 @@
|
|
|
1
|
-
import type { Assertion, AssertionResult } from "../types/index.js";
|
|
2
|
-
export declare function sortAssertionsCheapestFirst(assertions: Assertion[]): Assertion[];
|
|
3
|
-
export interface AssertionContext {
|
|
4
|
-
output: string;
|
|
5
|
-
durationMs?: number;
|
|
6
|
-
inputTokens?: number;
|
|
7
|
-
outputTokens?: number;
|
|
8
|
-
costUsd?: number;
|
|
9
|
-
toolCalls?: Array<{
|
|
10
|
-
name: string;
|
|
11
|
-
arguments?: Record<string, unknown>;
|
|
12
|
-
}>;
|
|
13
|
-
}
|
|
14
|
-
export declare function runAssertion(assertion: Assertion, ctx: AssertionContext): Promise<AssertionResult>;
|
|
15
|
-
export declare function runAssertions(assertions: Assertion[], ctx: AssertionContext): Promise<AssertionResult[]>;
|
|
16
|
-
export declare function assertionsPassed(results: AssertionResult[]): boolean;
|
|
17
|
-
export declare function allAssertionsPassed(results: AssertionResult[]): boolean;
|
|
18
|
-
//# sourceMappingURL=assertions.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"assertions.d.ts","sourceRoot":"","sources":["../../src/core/assertions.ts"],"names":[],"mappings":"AACA,OAAO,KAAK,EAAE,SAAS,EAAE,eAAe,EAAE,MAAM,mBAAmB,CAAC;AA6BpE,wBAAgB,2BAA2B,CAAC,UAAU,EAAE,SAAS,EAAE,GAAG,SAAS,EAAE,CAMhF;AAED,MAAM,WAAW,gBAAgB;IAC/B,MAAM,EAAE,MAAM,CAAC;IACf,UAAU,CAAC,EAAE,MAAM,CAAC;IACpB,WAAW,CAAC,EAAE,MAAM,CAAC;IACrB,YAAY,CAAC,EAAE,MAAM,CAAC;IACtB,OAAO,CAAC,EAAE,MAAM,CAAC;IACjB,SAAS,CAAC,EAAE,KAAK,CAAC;QAAE,IAAI,EAAE,MAAM,CAAC;QAAC,SAAS,CAAC,EAAE,MAAM,CAAC,MAAM,EAAE,OAAO,CAAC,CAAA;KAAE,CAAC,CAAC;CAC1E;AAED,wBAAsB,YAAY,CAChC,SAAS,EAAE,SAAS,EACpB,GAAG,EAAE,gBAAgB,GACpB,OAAO,CAAC,eAAe,CAAC,CAqB1B;AAED,wBAAsB,aAAa,CACjC,UAAU,EAAE,SAAS,EAAE,EACvB,GAAG,EAAE,gBAAgB,GACpB,OAAO,CAAC,eAAe,EAAE,CAAC,CAyB5B;AAED,wBAAgB,gBAAgB,CAAC,OAAO,EAAE,eAAe,EAAE,GAAG,OAAO,CAKpE;AAED,wBAAgB,mBAAmB,CAAC,OAAO,EAAE,eAAe,EAAE,GAAG,OAAO,CAEvE"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"assertions.test.d.ts","sourceRoot":"","sources":["../../src/core/assertions.test.ts"],"names":[],"mappings":""}
|
package/dist/core/e2e.test.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"e2e.test.d.ts","sourceRoot":"","sources":["../../src/core/e2e.test.ts"],"names":[],"mappings":"AAyBA,wBAAsB,SAAS,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAE9D"}
|
package/dist/core/judge.d.ts
DELETED
|
@@ -1,13 +0,0 @@
|
|
|
1
|
-
import type { JudgeConfig, JudgeResult } from "../types/index.js";
|
|
2
|
-
export declare function runJudge(input: string, output: string, config: JudgeConfig, expected?: string): Promise<JudgeResult>;
|
|
3
|
-
/** One-shot judge: no full eval case, just input/output/rubric */
|
|
4
|
-
export declare function judgeOnce(params: {
|
|
5
|
-
input: string;
|
|
6
|
-
output: string;
|
|
7
|
-
rubric: string;
|
|
8
|
-
expected?: string;
|
|
9
|
-
model?: string;
|
|
10
|
-
provider?: "anthropic" | "openai";
|
|
11
|
-
apiKey?: string;
|
|
12
|
-
}): Promise<JudgeResult>;
|
|
13
|
-
//# sourceMappingURL=judge.d.ts.map
|
package/dist/core/judge.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"judge.d.ts","sourceRoot":"","sources":["../../src/core/judge.ts"],"names":[],"mappings":"AAKA,OAAO,KAAK,EAAE,WAAW,EAAE,WAAW,EAAW,MAAM,mBAAmB,CAAC;AA4E3E,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,MAAM,EACb,MAAM,EAAE,MAAM,EACd,MAAM,EAAE,WAAW,EACnB,QAAQ,CAAC,EAAE,MAAM,GAChB,OAAO,CAAC,WAAW,CAAC,CAmBtB;AA4ED,kEAAkE;AAClE,wBAAsB,SAAS,CAAC,MAAM,EAAE;IACtC,KAAK,EAAE,MAAM,CAAC;IACd,MAAM,EAAE,MAAM,CAAC;IACf,MAAM,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,KAAK,CAAC,EAAE,MAAM,CAAC;IACf,QAAQ,CAAC,EAAE,WAAW,GAAG,QAAQ,CAAC;IAClC,MAAM,CAAC,EAAE,MAAM,CAAC;CACjB,GAAG,OAAO,CAAC,WAAW,CAAC,CAOvB"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"judge.test.d.ts","sourceRoot":"","sources":["../../src/core/judge.test.ts"],"names":[],"mappings":""}
|
package/dist/core/reporter.d.ts
DELETED
|
@@ -1,21 +0,0 @@
|
|
|
1
|
-
import type { EvalRun, Verdict } from "../types/index.js";
|
|
2
|
-
export declare function printTerminalReport(run: EvalRun): void;
|
|
3
|
-
export declare function toJson(run: EvalRun): string;
|
|
4
|
-
export declare function toMarkdown(run: EvalRun): string;
|
|
5
|
-
export interface RunDiff {
|
|
6
|
-
regressions: Array<{
|
|
7
|
-
caseId: string;
|
|
8
|
-
before: Verdict;
|
|
9
|
-
after: Verdict;
|
|
10
|
-
}>;
|
|
11
|
-
improvements: Array<{
|
|
12
|
-
caseId: string;
|
|
13
|
-
before: Verdict;
|
|
14
|
-
after: Verdict;
|
|
15
|
-
}>;
|
|
16
|
-
scoreDelta: number;
|
|
17
|
-
passRateDelta: number;
|
|
18
|
-
}
|
|
19
|
-
export declare function compareRuns(before: EvalRun, after: EvalRun): RunDiff;
|
|
20
|
-
export declare function printDiffReport(diff: RunDiff): void;
|
|
21
|
-
//# sourceMappingURL=reporter.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"reporter.d.ts","sourceRoot":"","sources":["../../src/core/reporter.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,OAAO,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAgB1D,wBAAgB,mBAAmB,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,CAoCtD;AAID,wBAAgB,MAAM,CAAC,GAAG,EAAE,OAAO,GAAG,MAAM,CAE3C;AAID,wBAAgB,UAAU,CAAC,GAAG,EAAE,OAAO,GAAG,MAAM,CAkD/C;AAID,MAAM,WAAW,OAAO;IACtB,WAAW,EAAE,KAAK,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,OAAO,CAAA;KAAE,CAAC,CAAC;IACxE,YAAY,EAAE,KAAK,CAAC;QAAE,MAAM,EAAE,MAAM,CAAC;QAAC,MAAM,EAAE,OAAO,CAAC;QAAC,KAAK,EAAE,OAAO,CAAA;KAAE,CAAC,CAAC;IACzE,UAAU,EAAE,MAAM,CAAC;IACnB,aAAa,EAAE,MAAM,CAAC;CACvB;AAED,wBAAgB,WAAW,CAAC,MAAM,EAAE,OAAO,EAAE,KAAK,EAAE,OAAO,GAAG,OAAO,CAuBpE;AAED,wBAAgB,eAAe,CAAC,IAAI,EAAE,OAAO,GAAG,IAAI,CAcnD"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"reporter.test.d.ts","sourceRoot":"","sources":["../../src/core/reporter.test.ts"],"names":[],"mappings":""}
|
package/dist/core/runner.d.ts
DELETED
|
@@ -1,4 +0,0 @@
|
|
|
1
|
-
import type { AdapterConfig, EvalCase, EvalResult, EvalRun, RunOptions } from "../types/index.js";
|
|
2
|
-
export declare function runEvals(cases: EvalCase[], options: RunOptions): Promise<EvalRun>;
|
|
3
|
-
export declare function runSingleCase(evalCase: EvalCase, adapterConfig: AdapterConfig, skipJudge?: boolean): Promise<EvalResult>;
|
|
4
|
-
//# sourceMappingURL=runner.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"runner.d.ts","sourceRoot":"","sources":["../../src/core/runner.ts"],"names":[],"mappings":"AASA,OAAO,KAAK,EACV,aAAa,EACb,QAAQ,EACR,UAAU,EACV,OAAO,EAEP,UAAU,EAGX,MAAM,mBAAmB,CAAC;AAqI3B,wBAAsB,QAAQ,CAC5B,KAAK,EAAE,QAAQ,EAAE,EACjB,OAAO,EAAE,UAAU,GAClB,OAAO,CAAC,OAAO,CAAC,CAqClB;AAID,wBAAsB,aAAa,CACjC,QAAQ,EAAE,QAAQ,EAClB,aAAa,EAAE,aAAa,EAC5B,SAAS,UAAQ,GAChB,OAAO,CAAC,UAAU,CAAC,CAKrB"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"runner.test.d.ts","sourceRoot":"","sources":["../../src/core/runner.test.ts"],"names":[],"mappings":""}
|
|
@@ -1,18 +0,0 @@
|
|
|
1
|
-
import type { EvalCase } from "../types/index.js";
|
|
2
|
-
export interface LoadOptions {
|
|
3
|
-
/** Fail hard on malformed lines instead of warning and skipping */
|
|
4
|
-
strict?: boolean;
|
|
5
|
-
/** Filter by tags */
|
|
6
|
-
tags?: string[];
|
|
7
|
-
}
|
|
8
|
-
export interface LoadResult {
|
|
9
|
-
cases: EvalCase[];
|
|
10
|
-
warnings: string[];
|
|
11
|
-
totalLines: number;
|
|
12
|
-
skipped: number;
|
|
13
|
-
}
|
|
14
|
-
/** Load eval cases from a JSONL or JSON file (or glob pattern) */
|
|
15
|
-
export declare function loadDataset(pathOrGlob: string, opts?: LoadOptions): Promise<LoadResult>;
|
|
16
|
-
/** Stream large JSONL files case-by-case (for very large datasets) */
|
|
17
|
-
export declare function streamDataset(path: string, opts?: LoadOptions): AsyncGenerator<EvalCase>;
|
|
18
|
-
//# sourceMappingURL=loader.d.ts.map
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"loader.d.ts","sourceRoot":"","sources":["../../src/datasets/loader.ts"],"names":[],"mappings":"AAEA,OAAO,KAAK,EAAE,QAAQ,EAAE,MAAM,mBAAmB,CAAC;AAElD,MAAM,WAAW,WAAW;IAC1B,mEAAmE;IACnE,MAAM,CAAC,EAAE,OAAO,CAAC;IACjB,qBAAqB;IACrB,IAAI,CAAC,EAAE,MAAM,EAAE,CAAC;CACjB;AAED,MAAM,WAAW,UAAU;IACzB,KAAK,EAAE,QAAQ,EAAE,CAAC;IAClB,QAAQ,EAAE,MAAM,EAAE,CAAC;IACnB,UAAU,EAAE,MAAM,CAAC;IACnB,OAAO,EAAE,MAAM,CAAC;CACjB;AAED,kEAAkE;AAClE,wBAAsB,WAAW,CAAC,UAAU,EAAE,MAAM,EAAE,IAAI,GAAE,WAAgB,GAAG,OAAO,CAAC,UAAU,CAAC,CAiCjG;AAoFD,sEAAsE;AACtE,wBAAuB,aAAa,CAAC,IAAI,EAAE,MAAM,EAAE,IAAI,GAAE,WAAgB,GAAG,cAAc,CAAC,QAAQ,CAAC,CAyBnG"}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"loader.test.d.ts","sourceRoot":"","sources":["../../src/datasets/loader.test.ts"],"names":[],"mappings":""}
|
package/dist/db/store.d.ts
DELETED
|
@@ -1,17 +0,0 @@
|
|
|
1
|
-
import { Database } from "bun:sqlite";
|
|
2
|
-
import type { EvalRun } from "../types/index.js";
|
|
3
|
-
export declare function getDatabase(): Database;
|
|
4
|
-
export declare function closeDatabase(): void;
|
|
5
|
-
export declare function saveRun(run: EvalRun): void;
|
|
6
|
-
export declare function getRun(id: string): EvalRun | null;
|
|
7
|
-
export declare function listRuns(limit?: number, dataset?: string): EvalRun[];
|
|
8
|
-
export declare function deleteRun(id: string): void;
|
|
9
|
-
export declare function setBaseline(name: string, runId: string): void;
|
|
10
|
-
export declare function getBaseline(name: string): EvalRun | null;
|
|
11
|
-
export declare function listBaselines(): Array<{
|
|
12
|
-
name: string;
|
|
13
|
-
runId: string;
|
|
14
|
-
createdAt: string;
|
|
15
|
-
}>;
|
|
16
|
-
export declare function clearBaseline(name: string): void;
|
|
17
|
-
//# sourceMappingURL=store.d.ts.map
|
package/dist/db/store.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"store.d.ts","sourceRoot":"","sources":["../../src/db/store.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,QAAQ,EAAE,MAAM,YAAY,CAAC;AAItC,OAAO,KAAK,EAAE,OAAO,EAAE,MAAM,mBAAmB,CAAC;AAQjD,wBAAgB,WAAW,IAAI,QAAQ,CAUtC;AAED,wBAAgB,aAAa,IAAI,IAAI,CAGpC;AA6BD,wBAAgB,OAAO,CAAC,GAAG,EAAE,OAAO,GAAG,IAAI,CAa1C;AAED,wBAAgB,MAAM,CAAC,EAAE,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI,CAIjD;AAED,wBAAgB,QAAQ,CAAC,KAAK,SAAK,EAAE,OAAO,CAAC,EAAE,MAAM,GAAG,OAAO,EAAE,CAMhE;AAED,wBAAgB,SAAS,CAAC,EAAE,EAAE,MAAM,GAAG,IAAI,CAE1C;AAID,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,EAAE,KAAK,EAAE,MAAM,GAAG,IAAI,CAI7D;AAED,wBAAgB,WAAW,CAAC,IAAI,EAAE,MAAM,GAAG,OAAO,GAAG,IAAI,CAKxD;AAED,wBAAgB,aAAa,IAAI,KAAK,CAAC;IAAE,IAAI,EAAE,MAAM,CAAC;IAAC,KAAK,EAAE,MAAM,CAAC;IAAC,SAAS,EAAE,MAAM,CAAA;CAAE,CAAC,CAKzF;AAED,wBAAgB,aAAa,CAAC,IAAI,EAAE,MAAM,GAAG,IAAI,CAEhD"}
|
package/dist/db/store.test.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"store.test.d.ts","sourceRoot":"","sources":["../../src/db/store.test.ts"],"names":[],"mappings":""}
|
package/dist/index.d.ts
DELETED
|
@@ -1,8 +0,0 @@
|
|
|
1
|
-
export * from "./types/index.js";
|
|
2
|
-
export * from "./core/runner.js";
|
|
3
|
-
export * from "./core/assertions.js";
|
|
4
|
-
export * from "./core/judge.js";
|
|
5
|
-
export * from "./core/reporter.js";
|
|
6
|
-
export * from "./datasets/loader.js";
|
|
7
|
-
export * from "./db/store.js";
|
|
8
|
-
//# sourceMappingURL=index.d.ts.map
|
package/dist/index.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../src/index.ts"],"names":[],"mappings":"AACA,cAAc,kBAAkB,CAAC;AACjC,cAAc,kBAAkB,CAAC;AACjC,cAAc,sBAAsB,CAAC;AACrC,cAAc,iBAAiB,CAAC;AAChC,cAAc,oBAAoB,CAAC;AACnC,cAAc,sBAAsB,CAAC;AACrC,cAAc,eAAe,CAAC"}
|
package/dist/mcp/http.d.ts
DELETED
|
@@ -1,12 +0,0 @@
|
|
|
1
|
-
import { DEFAULT_MCP_HTTP_PORT, MCP_NAME } from "./server.js";
|
|
2
|
-
export { DEFAULT_MCP_HTTP_PORT, MCP_NAME };
|
|
3
|
-
export declare function isHttpMode(argv?: string[]): boolean;
|
|
4
|
-
export declare function resolveHttpPort(argv?: string[]): number;
|
|
5
|
-
export declare function handleMcpHttpRequest(req: Request): Promise<Response>;
|
|
6
|
-
export interface StartHttpServerOptions {
|
|
7
|
-
port?: number;
|
|
8
|
-
hostname?: string;
|
|
9
|
-
log?: (message: string) => void;
|
|
10
|
-
}
|
|
11
|
-
export declare function startHttpServer(options?: StartHttpServerOptions): ReturnType<typeof Bun.serve>;
|
|
12
|
-
//# sourceMappingURL=http.d.ts.map
|
package/dist/mcp/http.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"http.d.ts","sourceRoot":"","sources":["../../src/mcp/http.ts"],"names":[],"mappings":"AACA,OAAO,EAAe,qBAAqB,EAAE,QAAQ,EAAE,MAAM,aAAa,CAAC;AAE3E,OAAO,EAAE,qBAAqB,EAAE,QAAQ,EAAE,CAAC;AAE3C,wBAAgB,UAAU,CAAC,IAAI,GAAE,MAAM,EAA0B,GAAG,OAAO,CAE1E;AAED,wBAAgB,eAAe,CAAC,IAAI,GAAE,MAAM,EAA0B,GAAG,MAAM,CAa9E;AAUD,wBAAsB,oBAAoB,CAAC,GAAG,EAAE,OAAO,GAAG,OAAO,CAAC,QAAQ,CAAC,CAiB1E;AAED,MAAM,WAAW,sBAAsB;IACrC,IAAI,CAAC,EAAE,MAAM,CAAC;IACd,QAAQ,CAAC,EAAE,MAAM,CAAC;IAClB,GAAG,CAAC,EAAE,CAAC,OAAO,EAAE,MAAM,KAAK,IAAI,CAAC;CACjC;AAED,wBAAgB,eAAe,CAAC,OAAO,GAAE,sBAA2B,GAAG,UAAU,CAAC,OAAO,GAAG,CAAC,KAAK,CAAC,CAclG"}
|
package/dist/mcp/http.test.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"http.test.d.ts","sourceRoot":"","sources":["../../src/mcp/http.test.ts"],"names":[],"mappings":""}
|
package/dist/mcp/index.d.ts
DELETED
package/dist/mcp/index.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/mcp/index.ts"],"names":[],"mappings":""}
|
package/dist/mcp/mcp.test.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"mcp.test.d.ts","sourceRoot":"","sources":["../../src/mcp/mcp.test.ts"],"names":[],"mappings":"AAsDA,wBAAsB,MAAM,CAAC,KAAK,EAAE,MAAM,GAAG,OAAO,CAAC,MAAM,CAAC,CAE3D"}
|
package/dist/mcp/server.d.ts
DELETED
package/dist/mcp/server.d.ts.map
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"server.d.ts","sourceRoot":"","sources":["../../src/mcp/server.ts"],"names":[],"mappings":"AAAA,OAAO,EAAE,MAAM,EAAE,MAAM,2CAA2C,CAAC;AAanE,eAAO,MAAM,QAAQ,UAAU,CAAC;AAChC,eAAO,MAAM,qBAAqB,OAAO,CAAC;AAE1C,wBAAgB,WAAW,IAAI,MAAM,CAmSpC"}
|
package/dist/server/index.d.ts
DELETED
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"index.d.ts","sourceRoot":"","sources":["../../src/server/index.ts"],"names":[],"mappings":""}
|
|
@@ -1 +0,0 @@
|
|
|
1
|
-
{"version":3,"file":"server.test.d.ts","sourceRoot":"","sources":["../../src/server/server.test.ts"],"names":[],"mappings":""}
|