6bf8bebf51
CI / Test and Build (push) Failing after 1s
CI / Migrate Dev DB (push) Has been skipped
CI / Migrate DB (push) Has been skipped
CodeQL / Analyze actions (push) Has been cancelled
CodeQL / Analyze javascript-typescript (push) Has been cancelled
CI / Detect Version (push) Has been cancelled
CI / Detect Desktop Changes (push) Has been cancelled
CI / Build AMD64 (blacksmith-2vcpu-ubuntu-2404, ./docker/cron.Dockerfile, ubuntu-latest, ghcr.io/simstudioai/cron) (push) Has been cancelled
CI / Build AMD64 (blacksmith-2vcpu-ubuntu-2404, ./docker/db.Dockerfile, ECR_MIGRATIONS, ubuntu-latest, ghcr.io/simstudioai/migrations) (push) Has been cancelled
CI / Build AMD64 (blacksmith-4vcpu-ubuntu-2404, ./docker/pii.Dockerfile, ECR_PII, ubuntu-latest, ghcr.io/simstudioai/pii) (push) Has been cancelled
CI / Build AMD64 (blacksmith-4vcpu-ubuntu-2404, ./docker/realtime.Dockerfile, ECR_REALTIME, ubuntu-latest, ghcr.io/simstudioai/realtime) (push) Has been cancelled
CI / Build AMD64 (blacksmith-8vcpu-ubuntu-2404, ./docker/app.Dockerfile, ECR_APP, linux-x64-8-core, ghcr.io/simstudioai/simstudio) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (blacksmith-4vcpu-ubuntu-2404-arm, ./docker/cron.Dockerfile, ubuntu-24.04-arm, ghcr.io/simstudioai/cron) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (blacksmith-4vcpu-ubuntu-2404-arm, ./docker/db.Dockerfile, ubuntu-24.04-arm, ghcr.io/simstudioai/migrations) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (blacksmith-4vcpu-ubuntu-2404-arm, ./docker/pii.Dockerfile, ubuntu-24.04-arm, ghcr.io/simstudioai/pii) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (blacksmith-4vcpu-ubuntu-2404-arm, ./docker/realtime.Dockerfile, ubuntu-24.04-arm, ghcr.io/simstudioai/realtime) (push) Has been cancelled
CI / Build ARM64 (GHCR Only) (blacksmith-8vcpu-ubuntu-2404-arm, ./docker/app.Dockerfile, linux-arm64-8-core, ghcr.io/simstudioai/simstudio) (push) Has been cancelled
CI / Check Docs Changes (push) Has been cancelled
Publish CLI Package / publish-npm (push) Has been cancelled
Publish Python SDK / publish-pypi (push) Has been cancelled
CI / Deploy Trigger.dev (Dev) (push) Has been cancelled
Helm Chart / Lint, test, and validate chart (push) Has been cancelled
Helm Chart / Chart version bumped (push) Has been cancelled
Publish TypeScript SDK / publish-npm (push) Has been cancelled
CI / Build Dev ECR (blacksmith-8vcpu-ubuntu-2404, ./docker/app.Dockerfile, ECR_APP, linux-x64-8-core) (push) Has been cancelled
CI / Promote Images (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/cron) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/migrations) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/pii) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/realtime) (push) Has been cancelled
CI / Build Dev ECR (blacksmith-2vcpu-ubuntu-2404, ./docker/db.Dockerfile, ECR_MIGRATIONS, ubuntu-latest) (push) Has been cancelled
CI / Build Dev ECR (blacksmith-4vcpu-ubuntu-2404, ./docker/pii.Dockerfile, ECR_PII, ubuntu-latest) (push) Has been cancelled
CI / Build Dev ECR (blacksmith-4vcpu-ubuntu-2404, ./docker/realtime.Dockerfile, ECR_REALTIME, ubuntu-latest) (push) Has been cancelled
CI / Create GHCR Manifests (ghcr.io/simstudioai/simstudio) (push) Has been cancelled
CI / Process Docs (push) Has been cancelled
CI / Create GitHub Release (push) Has been cancelled
CI / Check Desktop Signing Secrets (push) Has been cancelled
CI / Desktop Release (push) Has been cancelled
CI / Create Desktop Prerelease (push) Has been cancelled
CI / Desktop Prerelease Build (push) Has been cancelled
CI / Publish Desktop Prerelease (push) Has been cancelled
CI / Prune Desktop Prereleases (push) Has been cancelled
Helm Chart / Install on kind and run helm test (push) Has been cancelled
126 lines
4.4 KiB
TypeScript
126 lines
4.4 KiB
TypeScript
/**
|
|
* @vitest-environment node
|
|
*/
|
|
import { describe, expect, it } from 'vitest'
|
|
import { runTaskTool } from '@/tools/browser_use/run_task'
|
|
import { managedAgentCreateSessionTool } from '@/tools/managed_agent/create_session'
|
|
import { managedAgentRespondCustomToolTool } from '@/tools/managed_agent/respond_custom_tool'
|
|
import { managedAgentRespondToolConfirmationTool } from '@/tools/managed_agent/respond_tool_confirmation'
|
|
import { managedAgentRunSessionTool } from '@/tools/managed_agent/run_session'
|
|
import { managedAgentSendMessageTool } from '@/tools/managed_agent/send_message'
|
|
import type { ToolConfig } from '@/tools/types'
|
|
|
|
function selectModelInput(
|
|
tool: ToolConfig,
|
|
params: Record<string, unknown>
|
|
): Record<string, unknown> {
|
|
const modelInput = tool.request.modelInput
|
|
expect(modelInput?.mode).toBe('project')
|
|
if (modelInput?.mode !== 'project') throw new Error(`Expected ${tool.id} to project model input`)
|
|
return modelInput.select(params)
|
|
}
|
|
|
|
describe('direct-execution model-input selectors', () => {
|
|
it('selects only Browser Use fields that are supplied to its agent model', () => {
|
|
expect(
|
|
selectModelInput(runTaskTool, {
|
|
task: 'book a flight',
|
|
systemPromptExtension: 'prefer direct flights',
|
|
structuredOutput: '{"type":"object"}',
|
|
apiKey: 'secret-api-key',
|
|
variables: { password: 'secret-password' },
|
|
profile_id: 'profile-1',
|
|
startUrl: 'https://example.com',
|
|
allowedDomains: 'example.com',
|
|
metadata: { requestId: 'request-1' },
|
|
model: 'browser-use-2.0',
|
|
})
|
|
).toEqual({
|
|
task: 'book a flight',
|
|
systemPromptExtension: 'prefer direct flights',
|
|
structuredOutput: '{"type":"object"}',
|
|
})
|
|
})
|
|
|
|
it('selects only Managed Agent run-session text supplied to the model', () => {
|
|
expect(
|
|
selectModelInput(managedAgentRunSessionTool, {
|
|
userMessage: 'investigate this issue',
|
|
memoryInstructions: 'remember the customer preference',
|
|
accessToken: 'secret-api-key',
|
|
credential: 'credential-1',
|
|
agent: 'agent-1',
|
|
environment: 'environment-1',
|
|
memoryStoreId: 'memory-1',
|
|
sessionParameters: { requestId: 'request-1' },
|
|
})
|
|
).toEqual({
|
|
userMessage: 'investigate this issue',
|
|
memoryInstructions: 'remember the customer preference',
|
|
})
|
|
})
|
|
|
|
it('selects only Managed Agent create-session text supplied to the model', () => {
|
|
expect(
|
|
selectModelInput(managedAgentCreateSessionTool, {
|
|
userMessage: 'start the investigation',
|
|
memoryInstructions: 'use the account history',
|
|
accessToken: 'secret-api-key',
|
|
credential: 'credential-1',
|
|
agent: 'agent-1',
|
|
environment: 'environment-1',
|
|
vaults: ['vault-1'],
|
|
files: [{ fileId: 'file-1' }],
|
|
})
|
|
).toEqual({
|
|
userMessage: 'start the investigation',
|
|
memoryInstructions: 'use the account history',
|
|
})
|
|
})
|
|
|
|
it('selects only the message sent to an existing Managed Agent session', () => {
|
|
expect(
|
|
selectModelInput(managedAgentSendMessageTool, {
|
|
userMessage: 'continue with the next step',
|
|
accessToken: 'secret-api-key',
|
|
credential: 'credential-1',
|
|
sessionId: 'session-1',
|
|
})
|
|
).toEqual({ userMessage: 'continue with the next step' })
|
|
})
|
|
|
|
it('selects only the custom-tool result returned to the Managed Agent', () => {
|
|
expect(
|
|
selectModelInput(managedAgentRespondCustomToolTool, {
|
|
result: 'the operation completed',
|
|
isError: false,
|
|
customToolUseId: 'tool-use-1',
|
|
accessToken: 'secret-api-key',
|
|
credential: 'credential-1',
|
|
sessionId: 'session-1',
|
|
})
|
|
).toEqual({ result: 'the operation completed' })
|
|
})
|
|
|
|
it('selects only the denial reason surfaced to the Managed Agent', () => {
|
|
expect(
|
|
selectModelInput(managedAgentRespondToolConfirmationTool, {
|
|
denyMessage: 'the operation is not authorized',
|
|
decision: 'deny',
|
|
toolUseIds: ['tool-use-1'],
|
|
accessToken: 'secret-api-key',
|
|
credential: 'credential-1',
|
|
sessionId: 'session-1',
|
|
})
|
|
).toEqual({ denyMessage: 'the operation is not authorized' })
|
|
|
|
expect(
|
|
selectModelInput(managedAgentRespondToolConfirmationTool, {
|
|
denyMessage: 'must not be sent',
|
|
decision: 'allow',
|
|
toolUseIds: ['tool-use-1'],
|
|
})
|
|
).toEqual({})
|
|
})
|
|
})
|