agentgrader 1.0.0 → 1.0.2
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/index.js +3 -3
- package/package.json +5 -5
package/dist/index.js
CHANGED
|
@@ -5,7 +5,7 @@ import { render, Box, Text } from 'ink';
|
|
|
5
5
|
import { initDb, saveTestCase, saveAgentConfig } from '@agentgrader/store';
|
|
6
6
|
import { runSingle, runBenchmark, validateTestCase, TestCaseSchema, AgentConfigSchema } from '@agentgrader/core';
|
|
7
7
|
import { DockerSandboxProvider } from '@agentgrader/sandbox-docker';
|
|
8
|
-
import {
|
|
8
|
+
import { AiSdkAgentAdapter } from '@agentgrader/agent-openrouter';
|
|
9
9
|
import { jsx, jsxs } from 'react/jsx-runtime';
|
|
10
10
|
import { mkdirSync, writeFileSync, readFileSync, readdirSync, statSync } from 'fs';
|
|
11
11
|
import { stringify, parse } from 'yaml';
|
|
@@ -257,7 +257,7 @@ async function runBenchCommand(opts) {
|
|
|
257
257
|
});
|
|
258
258
|
}
|
|
259
259
|
const sandboxProvider = new DockerSandboxProvider();
|
|
260
|
-
const adapter = new
|
|
260
|
+
const adapter = new AiSdkAgentAdapter();
|
|
261
261
|
const runStates = {};
|
|
262
262
|
const testCaseIds = testCases.map((tc) => tc.id || tc.name);
|
|
263
263
|
const configIds = agentConfigs.map((ac) => ac.id || ac.name);
|
|
@@ -456,7 +456,7 @@ async function runSingleCommand(testCasePath, opts) {
|
|
|
456
456
|
}
|
|
457
457
|
console.log(`Starting run for "${testCase.name}" using model "${agentConfig.model}"...`);
|
|
458
458
|
const sandboxProvider = new DockerSandboxProvider();
|
|
459
|
-
const adapter = new
|
|
459
|
+
const adapter = new AiSdkAgentAdapter();
|
|
460
460
|
const db = initDb();
|
|
461
461
|
await saveTestCase(db, testCaseToDbRow(testCase));
|
|
462
462
|
const runId = randomUUID();
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "agentgrader",
|
|
3
|
-
"version": "1.0.
|
|
3
|
+
"version": "1.0.2",
|
|
4
4
|
"description": "CLI for the Agentgrader benchmarking framework — run and bench coding agents",
|
|
5
5
|
"license": "MIT",
|
|
6
6
|
"type": "module",
|
|
@@ -19,10 +19,10 @@
|
|
|
19
19
|
"dev": "bun run src/index.ts"
|
|
20
20
|
},
|
|
21
21
|
"dependencies": {
|
|
22
|
-
"@agentgrader/agent-openrouter": "
|
|
23
|
-
"@agentgrader/core": "
|
|
24
|
-
"@agentgrader/sandbox-docker": "
|
|
25
|
-
"@agentgrader/store": "
|
|
22
|
+
"@agentgrader/agent-openrouter": "^2.0.0",
|
|
23
|
+
"@agentgrader/core": "^1.1.0",
|
|
24
|
+
"@agentgrader/sandbox-docker": "^2.0.0",
|
|
25
|
+
"@agentgrader/store": "^1.0.2",
|
|
26
26
|
"cac": "^6.7.14",
|
|
27
27
|
"ink": "^4.4.1",
|
|
28
28
|
"react": "^18.2.0",
|