Skip to content

Commit 48e23c8

Browse files
committed
Update prepared grading artifact expectations
1 parent 55641dd commit 48e23c8

1 file changed

Lines changed: 6 additions & 7 deletions

File tree

apps/cli/test/commands/grade/grade-prepared.test.ts

Lines changed: 6 additions & 7 deletions
Original file line numberDiff line numberDiff line change
@@ -209,8 +209,8 @@ describe('agentv grade prepared attempts', () => {
209209
);
210210

211211
const grading = JSON.parse(await readFile(path.join(runDir, row.grading_path), 'utf8'));
212-
expect(grading.workspace_changes).not.toHaveProperty('diff_summary');
213-
expect(grading.workspace_changes.files_modified).toBeGreaterThanOrEqual(1);
212+
expect(grading.metadata.workspace_changes).not.toHaveProperty('diff_summary');
213+
expect(grading.metadata.workspace_changes.files_modified).toBeGreaterThanOrEqual(1);
214214
}, 20_000);
215215

216216
it('fails clearly when the prepared manifest is missing', async () => {
@@ -289,10 +289,8 @@ describe('agentv grade prepared attempts', () => {
289289
name: 'expected-tool-sequence',
290290
type: 'tool-trajectory',
291291
score: 0,
292-
});
293-
expect(row.scores[0].assertions[0]).toMatchObject({
294-
text: 'No trace available for evaluation',
295-
passed: false,
292+
pass: false,
293+
reason: 'No trace available for evaluation',
296294
});
297295
});
298296

@@ -401,7 +399,8 @@ describe('agentv grade prepared attempts', () => {
401399
name: 'expected-tool-sequence',
402400
type: 'tool-trajectory',
403401
score: 1,
404-
assertions: [{ text: 'Found Read at position 0', passed: true }],
402+
pass: true,
403+
reason: 'Grader passed.',
405404
});
406405
expect(row.metadata.prepared_attempt.trace_path).toBe(tracePath);
407406
});

0 commit comments

Comments
 (0)