chore: import upstream snapshot with attribution
CI / Shell Format Check (push) Has been cancelled
CI / Check Ruby (3.4) (push) Has been cancelled
CI / CI Config (push) Has been cancelled
CI / Test on Node ${{ matrix.node }} and ${{ matrix.os }}${{ matrix.shard && format(' (shard {0}/3)', matrix.shard) || '' }} (push) Has been cancelled
CI / Build on Node ${{ matrix.node }} (push) Has been cancelled
CI / Style Check (push) Has been cancelled
CI / Generate Assets (push) Has been cancelled
CI / Check Python (3.14) (push) Has been cancelled
CI / Check Python (3.9) (push) Has been cancelled
CI / Build Docs (push) Has been cancelled
CI / Code Scan Action (push) Has been cancelled
CI / Site tests (push) Has been cancelled
CI / webui tests (push) Has been cancelled
CI / Run Integration Tests (push) Has been cancelled
CI / Run Smoke Tests (push) Has been cancelled
CI / Go Tests (push) Has been cancelled
CI / Share Test (push) Has been cancelled
CI / Redteam (Production API) (push) Has been cancelled
CI / Redteam (Staging API) (push) Has been cancelled
CI / GitHub Actions Lint (push) Has been cancelled
CI / Check Ruby (3.0) (push) Has been cancelled
release-please / release-please (push) Has been cancelled
release-please / build (push) Has been cancelled
release-please / publish-npm (push) Has been cancelled
release-please / publish-npm-backfill (push) Has been cancelled
release-please / docker (push) Has been cancelled
release-please / publish-code-scan-action (push) Has been cancelled
release-please / attest-code-scan-action (push) Has been cancelled
Deploy local.promptfoo.app / Deploy to Cloudflare Pages (push) Has been cancelled
Test and Publish Multi-arch Docker Image / test (push) Has been cancelled
Test and Publish Multi-arch Docker Image / build-docker-and-push-digests (map[digest-suffix:linux-amd64 platform:linux/amd64 runner:ubuntu-latest]) (push) Has been cancelled
Test and Publish Multi-arch Docker Image / build-docker-and-push-digests (map[digest-suffix:linux-arm64 platform:linux/arm64 runner:ubuntu-24.04-arm]) (push) Has been cancelled
Test and Publish Multi-arch Docker Image / merge-docker-digests (push) Has been cancelled
Test and Publish Multi-arch Docker Image / Attest Multi-arch Image (push) Has been cancelled
Validate Renovate Config / Validate Renovate Configuration (push) Has been cancelled
CI / Shell Format Check (push) Has been cancelled
CI / Check Ruby (3.4) (push) Has been cancelled
CI / CI Config (push) Has been cancelled
CI / Test on Node ${{ matrix.node }} and ${{ matrix.os }}${{ matrix.shard && format(' (shard {0}/3)', matrix.shard) || '' }} (push) Has been cancelled
CI / Build on Node ${{ matrix.node }} (push) Has been cancelled
CI / Style Check (push) Has been cancelled
CI / Generate Assets (push) Has been cancelled
CI / Check Python (3.14) (push) Has been cancelled
CI / Check Python (3.9) (push) Has been cancelled
CI / Build Docs (push) Has been cancelled
CI / Code Scan Action (push) Has been cancelled
CI / Site tests (push) Has been cancelled
CI / webui tests (push) Has been cancelled
CI / Run Integration Tests (push) Has been cancelled
CI / Run Smoke Tests (push) Has been cancelled
CI / Go Tests (push) Has been cancelled
CI / Share Test (push) Has been cancelled
CI / Redteam (Production API) (push) Has been cancelled
CI / Redteam (Staging API) (push) Has been cancelled
CI / GitHub Actions Lint (push) Has been cancelled
CI / Check Ruby (3.0) (push) Has been cancelled
release-please / release-please (push) Has been cancelled
release-please / build (push) Has been cancelled
release-please / publish-npm (push) Has been cancelled
release-please / publish-npm-backfill (push) Has been cancelled
release-please / docker (push) Has been cancelled
release-please / publish-code-scan-action (push) Has been cancelled
release-please / attest-code-scan-action (push) Has been cancelled
Deploy local.promptfoo.app / Deploy to Cloudflare Pages (push) Has been cancelled
Test and Publish Multi-arch Docker Image / test (push) Has been cancelled
Test and Publish Multi-arch Docker Image / build-docker-and-push-digests (map[digest-suffix:linux-amd64 platform:linux/amd64 runner:ubuntu-latest]) (push) Has been cancelled
Test and Publish Multi-arch Docker Image / build-docker-and-push-digests (map[digest-suffix:linux-arm64 platform:linux/arm64 runner:ubuntu-24.04-arm]) (push) Has been cancelled
Test and Publish Multi-arch Docker Image / merge-docker-digests (push) Has been cancelled
Test and Publish Multi-arch Docker Image / Attest Multi-arch Image (push) Has been cancelled
Validate Renovate Config / Validate Renovate Configuration (push) Has been cancelled
This commit is contained in:
@@ -0,0 +1,162 @@
|
||||
import { describe, expect, it } from 'vitest';
|
||||
import {
|
||||
ATTACK_PROVIDER_IDS,
|
||||
getAttackProviderFullId,
|
||||
isAttackProvider,
|
||||
isMultiTurnStrategy,
|
||||
} from '../../../src/redteam/shared/attackProviders';
|
||||
|
||||
describe('attackProviders', () => {
|
||||
describe('ATTACK_PROVIDER_IDS', () => {
|
||||
it('should contain all expected multi-turn attack providers', () => {
|
||||
expect(ATTACK_PROVIDER_IDS).toContain('hydra');
|
||||
expect(ATTACK_PROVIDER_IDS).toContain('crescendo');
|
||||
expect(ATTACK_PROVIDER_IDS).toContain('goat');
|
||||
expect(ATTACK_PROVIDER_IDS).toContain('custom');
|
||||
});
|
||||
|
||||
it('should contain all expected single-turn iterative attack providers', () => {
|
||||
expect(ATTACK_PROVIDER_IDS).toContain('iterative');
|
||||
expect(ATTACK_PROVIDER_IDS).toContain('iterative:meta');
|
||||
expect(ATTACK_PROVIDER_IDS).toContain('iterative:tree');
|
||||
});
|
||||
|
||||
it('should not contain unsupported strategies', () => {
|
||||
expect(ATTACK_PROVIDER_IDS).not.toContain('mischievous-user');
|
||||
});
|
||||
});
|
||||
|
||||
describe('isAttackProvider', () => {
|
||||
describe('multi-turn conversation providers', () => {
|
||||
it('should return true for hydra variants', () => {
|
||||
expect(isAttackProvider('hydra')).toBe(true);
|
||||
expect(isAttackProvider('jailbreak:hydra')).toBe(true);
|
||||
expect(isAttackProvider('promptfoo:redteam:hydra')).toBe(true);
|
||||
});
|
||||
|
||||
it('should return true for crescendo variants', () => {
|
||||
expect(isAttackProvider('crescendo')).toBe(true);
|
||||
expect(isAttackProvider('promptfoo:redteam:crescendo')).toBe(true);
|
||||
});
|
||||
|
||||
it('should return true for goat variants', () => {
|
||||
expect(isAttackProvider('goat')).toBe(true);
|
||||
expect(isAttackProvider('promptfoo:redteam:goat')).toBe(true);
|
||||
});
|
||||
|
||||
it('should return true for custom variants', () => {
|
||||
expect(isAttackProvider('custom')).toBe(true);
|
||||
expect(isAttackProvider('custom:my-custom')).toBe(true);
|
||||
expect(isAttackProvider('promptfoo:redteam:custom')).toBe(true);
|
||||
});
|
||||
});
|
||||
|
||||
describe('multi-attempt single-turn providers', () => {
|
||||
it('should return true for iterative (base jailbreak)', () => {
|
||||
expect(isAttackProvider('iterative')).toBe(true);
|
||||
expect(isAttackProvider('jailbreak')).toBe(true);
|
||||
expect(isAttackProvider('promptfoo:redteam:iterative')).toBe(true);
|
||||
});
|
||||
|
||||
it('should return true for iterative:meta (jailbreak:meta)', () => {
|
||||
expect(isAttackProvider('iterative:meta')).toBe(true);
|
||||
expect(isAttackProvider('jailbreak:meta')).toBe(true);
|
||||
expect(isAttackProvider('promptfoo:redteam:iterative:meta')).toBe(true);
|
||||
});
|
||||
|
||||
it('should return true for iterative:tree (jailbreak:tree)', () => {
|
||||
expect(isAttackProvider('iterative:tree')).toBe(true);
|
||||
expect(isAttackProvider('jailbreak:tree')).toBe(true);
|
||||
expect(isAttackProvider('promptfoo:redteam:iterative:tree')).toBe(true);
|
||||
});
|
||||
});
|
||||
|
||||
describe('non-attack providers', () => {
|
||||
it('should return false for transform strategies', () => {
|
||||
expect(isAttackProvider('base64')).toBe(false);
|
||||
expect(isAttackProvider('rot13')).toBe(false);
|
||||
expect(isAttackProvider('hex')).toBe(false);
|
||||
expect(isAttackProvider('audio')).toBe(false);
|
||||
expect(isAttackProvider('image')).toBe(false);
|
||||
});
|
||||
|
||||
it('should return false for unsupported attack strategies', () => {
|
||||
expect(isAttackProvider('mischievous-user')).toBe(false);
|
||||
});
|
||||
|
||||
it('should return false for basic and other non-attack strategies', () => {
|
||||
expect(isAttackProvider('basic')).toBe(false);
|
||||
expect(isAttackProvider('default')).toBe(false);
|
||||
expect(isAttackProvider('unknown')).toBe(false);
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe('getAttackProviderFullId', () => {
|
||||
describe('already full IDs', () => {
|
||||
it('should return the same ID if already full', () => {
|
||||
expect(getAttackProviderFullId('promptfoo:redteam:hydra')).toBe('promptfoo:redteam:hydra');
|
||||
expect(getAttackProviderFullId('promptfoo:redteam:iterative')).toBe(
|
||||
'promptfoo:redteam:iterative',
|
||||
);
|
||||
expect(getAttackProviderFullId('promptfoo:redteam:iterative:meta')).toBe(
|
||||
'promptfoo:redteam:iterative:meta',
|
||||
);
|
||||
});
|
||||
});
|
||||
|
||||
describe('jailbreak variants', () => {
|
||||
it('should convert jailbreak (base) to promptfoo:redteam:iterative', () => {
|
||||
expect(getAttackProviderFullId('jailbreak')).toBe('promptfoo:redteam:iterative');
|
||||
});
|
||||
|
||||
it('should convert jailbreak:meta to promptfoo:redteam:iterative:meta', () => {
|
||||
expect(getAttackProviderFullId('jailbreak:meta')).toBe('promptfoo:redteam:iterative:meta');
|
||||
});
|
||||
|
||||
it('should convert jailbreak:tree to promptfoo:redteam:iterative:tree', () => {
|
||||
expect(getAttackProviderFullId('jailbreak:tree')).toBe('promptfoo:redteam:iterative:tree');
|
||||
});
|
||||
|
||||
it('should convert jailbreak:hydra to promptfoo:redteam:hydra', () => {
|
||||
expect(getAttackProviderFullId('jailbreak:hydra')).toBe('promptfoo:redteam:hydra');
|
||||
});
|
||||
});
|
||||
|
||||
describe('custom variants', () => {
|
||||
it('should convert custom to promptfoo:redteam:custom', () => {
|
||||
expect(getAttackProviderFullId('custom')).toBe('promptfoo:redteam:custom');
|
||||
});
|
||||
|
||||
it('should convert custom:variant to promptfoo:redteam:custom', () => {
|
||||
expect(getAttackProviderFullId('custom:my-custom')).toBe('promptfoo:redteam:custom');
|
||||
});
|
||||
});
|
||||
|
||||
describe('short IDs', () => {
|
||||
it('should convert short IDs to full IDs', () => {
|
||||
expect(getAttackProviderFullId('hydra')).toBe('promptfoo:redteam:hydra');
|
||||
expect(getAttackProviderFullId('crescendo')).toBe('promptfoo:redteam:crescendo');
|
||||
expect(getAttackProviderFullId('goat')).toBe('promptfoo:redteam:goat');
|
||||
});
|
||||
});
|
||||
});
|
||||
|
||||
describe('isMultiTurnStrategy', () => {
|
||||
it('should return true for multi-turn strategies', () => {
|
||||
expect(isMultiTurnStrategy('hydra')).toBe(true);
|
||||
expect(isMultiTurnStrategy('crescendo')).toBe(true);
|
||||
expect(isMultiTurnStrategy('goat')).toBe(true);
|
||||
});
|
||||
|
||||
it('should handle various ID formats', () => {
|
||||
expect(isMultiTurnStrategy('jailbreak:hydra')).toBe(true);
|
||||
expect(isMultiTurnStrategy('promptfoo:redteam:hydra')).toBe(true);
|
||||
});
|
||||
|
||||
it('should return false for non-multi-turn strategies', () => {
|
||||
expect(isMultiTurnStrategy('base64')).toBe(false);
|
||||
expect(isMultiTurnStrategy('basic')).toBe(false);
|
||||
});
|
||||
});
|
||||
});
|
||||
@@ -0,0 +1,250 @@
|
||||
import { beforeEach, describe, expect, it, vi } from 'vitest';
|
||||
import { applyRuntimeTransforms } from '../../../src/redteam/shared/runtimeTransform';
|
||||
|
||||
import type { Strategy } from '../../../src/redteam/strategies/types';
|
||||
import type { TestCaseWithPlugin } from '../../../src/types';
|
||||
|
||||
describe('runtimeTransform', () => {
|
||||
const mockBase64Strategy: Strategy = {
|
||||
id: 'base64',
|
||||
action: vi.fn(async (testCases: TestCaseWithPlugin[]) =>
|
||||
testCases.map((tc) => ({
|
||||
...tc,
|
||||
vars: {
|
||||
...tc.vars,
|
||||
input: Buffer.from(String(tc.vars?.input || '')).toString('base64'),
|
||||
},
|
||||
})),
|
||||
),
|
||||
};
|
||||
|
||||
const mockAudioStrategy: Strategy = {
|
||||
id: 'audio',
|
||||
action: vi.fn(async (testCases: TestCaseWithPlugin[]) =>
|
||||
testCases.map((tc) => ({
|
||||
...tc,
|
||||
vars: {
|
||||
...tc.vars,
|
||||
input: 'data:audio/mp3;base64,SGVsbG9BdWRpbw==', // Mock audio data URL
|
||||
},
|
||||
})),
|
||||
),
|
||||
};
|
||||
|
||||
const mockImageStrategy: Strategy = {
|
||||
id: 'image',
|
||||
action: vi.fn(async (testCases: TestCaseWithPlugin[]) =>
|
||||
testCases.map((tc) => ({
|
||||
...tc,
|
||||
vars: {
|
||||
...tc.vars,
|
||||
input: 'data:image/png;base64,SGVsbG9JbWFnZQ==', // Mock image data URL
|
||||
},
|
||||
})),
|
||||
),
|
||||
};
|
||||
|
||||
const mockRawAudioStrategy: Strategy = {
|
||||
id: 'audio-raw',
|
||||
action: vi.fn(async (testCases: TestCaseWithPlugin[]) =>
|
||||
testCases.map((tc) => ({
|
||||
...tc,
|
||||
vars: {
|
||||
...tc.vars,
|
||||
input: 'SGVsbG9SYXdBdWRpbw==', // Raw base64 without data URL
|
||||
},
|
||||
})),
|
||||
),
|
||||
};
|
||||
|
||||
const mockStrategies: Strategy[] = [
|
||||
mockBase64Strategy,
|
||||
mockAudioStrategy,
|
||||
mockImageStrategy,
|
||||
mockRawAudioStrategy,
|
||||
];
|
||||
|
||||
beforeEach(() => {
|
||||
vi.clearAllMocks();
|
||||
});
|
||||
|
||||
describe('applyRuntimeTransforms', () => {
|
||||
it('should return original prompt when no layers provided', async () => {
|
||||
const result = await applyRuntimeTransforms('test prompt', 'input', [], mockStrategies);
|
||||
|
||||
expect(result).toEqual({
|
||||
prompt: 'test prompt',
|
||||
originalPrompt: 'test prompt',
|
||||
});
|
||||
});
|
||||
|
||||
it('should return original prompt when layers is empty array', async () => {
|
||||
const result = await applyRuntimeTransforms('test prompt', 'input', [], mockStrategies);
|
||||
|
||||
expect(result.prompt).toBe('test prompt');
|
||||
expect(result.originalPrompt).toBe('test prompt');
|
||||
expect(result.audio).toBeUndefined();
|
||||
expect(result.image).toBeUndefined();
|
||||
});
|
||||
|
||||
it('should apply single transform layer', async () => {
|
||||
const result = await applyRuntimeTransforms('hello', 'input', ['base64'], mockStrategies);
|
||||
|
||||
expect(result.prompt).toBe('aGVsbG8='); // base64('hello')
|
||||
expect(result.originalPrompt).toBe('hello');
|
||||
expect(mockBase64Strategy.action).toHaveBeenCalledTimes(1);
|
||||
});
|
||||
|
||||
it('should apply multiple transform layers in order', async () => {
|
||||
// First base64, then the second strategy would transform the base64 result
|
||||
// But for this test, let's just verify both are called
|
||||
const mockSecondStrategy: Strategy = {
|
||||
id: 'rot13',
|
||||
action: vi.fn(async (testCases) => testCases),
|
||||
};
|
||||
|
||||
const strategies = [...mockStrategies, mockSecondStrategy];
|
||||
|
||||
await applyRuntimeTransforms('hello', 'input', ['base64', 'rot13'], strategies);
|
||||
|
||||
expect(mockBase64Strategy.action).toHaveBeenCalledTimes(1);
|
||||
expect(mockSecondStrategy.action).toHaveBeenCalledTimes(1);
|
||||
});
|
||||
|
||||
it('should extract audio data from data URL', async () => {
|
||||
const result = await applyRuntimeTransforms('hello', 'input', ['audio'], mockStrategies);
|
||||
|
||||
expect(result.audio).toEqual({
|
||||
data: 'SGVsbG9BdWRpbw==',
|
||||
format: 'mp3',
|
||||
});
|
||||
expect(result.originalPrompt).toBe('hello');
|
||||
});
|
||||
|
||||
it('should extract image data from data URL', async () => {
|
||||
const result = await applyRuntimeTransforms('hello', 'input', ['image'], mockStrategies);
|
||||
|
||||
expect(result.image).toEqual({
|
||||
data: 'SGVsbG9JbWFnZQ==',
|
||||
format: 'png',
|
||||
});
|
||||
expect(result.originalPrompt).toBe('hello');
|
||||
});
|
||||
|
||||
it('should handle raw base64 audio (assume mp3 format)', async () => {
|
||||
// Create a strategy that mimics returning raw base64 audio
|
||||
const rawAudioStrategy: Strategy = {
|
||||
id: 'audio',
|
||||
action: vi.fn(async (testCases: TestCaseWithPlugin[]) =>
|
||||
testCases.map((tc) => ({
|
||||
...tc,
|
||||
vars: {
|
||||
...tc.vars,
|
||||
input: 'SGVsbG9SYXdBdWRpbw==', // Raw base64 without data URL prefix
|
||||
},
|
||||
})),
|
||||
),
|
||||
};
|
||||
|
||||
const strategies = [rawAudioStrategy];
|
||||
const result = await applyRuntimeTransforms('hello', 'input', ['audio'], strategies);
|
||||
|
||||
expect(result.audio).toEqual({
|
||||
data: 'SGVsbG9SYXdBdWRpbw==',
|
||||
format: 'mp3', // Default format for raw base64
|
||||
});
|
||||
});
|
||||
|
||||
it('should handle layer config objects', async () => {
|
||||
const result = await applyRuntimeTransforms(
|
||||
'hello',
|
||||
'input',
|
||||
[{ id: 'base64', config: { someOption: true } }],
|
||||
mockStrategies,
|
||||
);
|
||||
|
||||
expect(result.prompt).toBe('aGVsbG8=');
|
||||
expect(mockBase64Strategy.action).toHaveBeenCalledWith(
|
||||
expect.any(Array),
|
||||
'input',
|
||||
expect.objectContaining({ someOption: true }),
|
||||
);
|
||||
});
|
||||
|
||||
it('should pass target context to per-turn layer strategies', async () => {
|
||||
const result = await applyRuntimeTransforms('hello', 'input', ['base64'], mockStrategies, {
|
||||
targetId: 'cloud-target-123',
|
||||
});
|
||||
|
||||
expect(mockBase64Strategy.action).toHaveBeenCalledWith(
|
||||
expect.any(Array),
|
||||
'input',
|
||||
expect.objectContaining({ targetId: 'cloud-target-123' }),
|
||||
);
|
||||
|
||||
expect(result.prompt).toBe('aGVsbG8=');
|
||||
expect(result.originalPrompt).toBe('hello');
|
||||
});
|
||||
|
||||
it('should skip unknown strategies with warning', async () => {
|
||||
const result = await applyRuntimeTransforms(
|
||||
'hello',
|
||||
'input',
|
||||
['unknown-strategy', 'base64'],
|
||||
mockStrategies,
|
||||
);
|
||||
|
||||
// Should still apply base64
|
||||
expect(result.prompt).toBe('aGVsbG8=');
|
||||
expect(mockBase64Strategy.action).toHaveBeenCalledTimes(1);
|
||||
});
|
||||
|
||||
it('should return error when transform fails', async () => {
|
||||
const failingStrategy: Strategy = {
|
||||
id: 'failing',
|
||||
action: vi.fn(async () => {
|
||||
throw new Error('Transform failed');
|
||||
}),
|
||||
};
|
||||
|
||||
const strategies = [...mockStrategies, failingStrategy];
|
||||
const result = await applyRuntimeTransforms('hello', 'input', ['failing'], strategies);
|
||||
|
||||
expect(result.error).toContain('Transform failing failed');
|
||||
expect(result.prompt).toBe('hello'); // Returns original prompt on error
|
||||
expect(result.originalPrompt).toBe('hello');
|
||||
});
|
||||
|
||||
it('should preserve pluginId in metadata during transforms', async () => {
|
||||
let capturedTestCase: TestCaseWithPlugin | undefined;
|
||||
|
||||
const inspectingStrategy: Strategy = {
|
||||
id: 'inspect',
|
||||
action: vi.fn(async (testCases: TestCaseWithPlugin[]) => {
|
||||
capturedTestCase = testCases[0];
|
||||
return testCases;
|
||||
}),
|
||||
};
|
||||
|
||||
const strategies = [...mockStrategies, inspectingStrategy];
|
||||
await applyRuntimeTransforms('hello', 'input', ['inspect'], strategies);
|
||||
|
||||
expect(capturedTestCase?.metadata?.pluginId).toBe('runtime-transform');
|
||||
});
|
||||
|
||||
it('should handle different inject variable names', async () => {
|
||||
await applyRuntimeTransforms('test', 'query', ['base64'], mockStrategies);
|
||||
|
||||
// The strategy should receive the correct inject var name
|
||||
expect(mockBase64Strategy.action).toHaveBeenCalledWith(
|
||||
expect.arrayContaining([
|
||||
expect.objectContaining({
|
||||
vars: expect.objectContaining({ query: 'test' }),
|
||||
}),
|
||||
]),
|
||||
'query',
|
||||
expect.any(Object),
|
||||
);
|
||||
});
|
||||
});
|
||||
});
|
||||
Reference in New Issue
Block a user