LibreChat/packages/api/src/agents/validation.ts
“Praneeth 949682ef0f
🏪 feat: Agent Marketplace
bugfix: Enhance Agent and AgentCategory schemas with new fields for category, support contact, and promotion status

refactored and moved agent category methods and schema to data-schema package

🔧 fix: Merge and Rebase Conflicts

- Move AgentCategory from api/models to @packages/data-schemas structure
  - Add schema, types, methods, and model following codebase conventions
  - Implement auto-seeding of default categories during AppService startup
  - Update marketplace controller to use new data-schemas methods
  - Remove old model file and standalone seed script

refactor: unify agent marketplace to single endpoint with cursor pagination

  - Replace multiple marketplace routes with unified /marketplace endpoint
  - Add query string controls: category, search, limit, cursor, promoted, requiredPermission
  - Implement cursor-based pagination replacing page-based system
  - Integrate ACL permissions for proper access control
  - Fix ObjectId constructor error in Agent model
  - Update React components to use unified useGetMarketplaceAgentsQuery hook
  - Enhance type safety and remove deprecated useDynamicAgentQuery
  - Update tests for new marketplace architecture
  -Known issues:
  see more button after category switching + Unit tests

feat: add icon property to ProcessedAgentCategory interface

- Add useMarketplaceAgentsInfiniteQuery and useGetAgentCategoriesQuery to client/src/data-provider/Agents/
  - Replace manual pagination in AgentGrid with infinite query pattern
  - Update imports to use local data provider instead of librechat-data-provider
  - Add proper permission handling with PERMISSION_BITS.VIEW/EDIT constants
  - Improve agent access control by adding requiredPermission validation in backend
  - Remove manual cursor/state management in favor of infinite query built-ins
  - Maintain existing search and category filtering functionality

refactor: consolidate agent marketplace endpoints into main agents API and improve data management consistency

  - Remove dedicated marketplace controller and routes, merging functionality into main agents v1 API
  - Add countPromotedAgents function to Agent model for promoted agents count
  - Enhance getListAgents handler with marketplace filtering (category, search, promoted status)
  - Move getAgentCategories from marketplace to v1 controller with same functionality
  - Update agent mutations to invalidate marketplace queries and handle multiple permission levels
  - Improve cache management by updating all agent query variants (VIEW/EDIT permissions)
  - Consolidate agent data access patterns for better maintainability and consistency
  - Remove duplicate marketplace route definitions and middleware

selected view only agents injected in the drop down

fix: remove minlength validation for support contact name in agent schema

feat: add validation and error messages for agent name in AgentConfig and AgentPanel

fix: update agent permission check logic in AgentPanel to simplify condition

Fix linting WIP

Fix Unit tests WIP

ESLint fixes

eslint fix

refactor: enhance isDuplicateVersion function in Agent model for improved comparison logic

- Introduced handling for undefined/null values in array and object comparisons.
- Normalized array comparisons to treat undefined/null as empty arrays.
- Added deep comparison for objects and improved handling of primitive values.
- Enhanced projectIds comparison to ensure consistent MongoDB ObjectId handling.

refactor: remove redundant properties from IAgent interface in agent schema

chore: update localization for agent detail component and clean up imports

ci: update access middleware tests

chore: remove unused PermissionTypes import from Role model

ci: update AclEntry model tests

ci: update button accessibility labels in AgentDetail tests

refactor: update exhaustive dep. lint warning

🔧 fix: Fixed agent actions access

feat: Add role-level permissions for agent sharing people picker

  - Add PEOPLE_PICKER permission type with VIEW_USERS and VIEW_GROUPS permissions
  - Create custom middleware for query-aware permission validation
  - Implement permission-based type filtering in PeoplePicker component
  - Hide people picker UI when user lacks permissions, show only public toggle
  - Support granular access: users-only, groups-only, or mixed search modes

refactor: Replace marketplace interface config with permission-based system

  - Add MARKETPLACE permission type to handle marketplace access control
  - Update interface configuration to use role-based marketplace settings (admin/user)
  - Replace direct marketplace boolean config with permission-based checks
  - Modify frontend components to use marketplace permissions instead of interface config
  - Update agent query hooks to use marketplace permissions for determining permission levels
  - Add marketplace configuration structure similar to peoplePicker in YAML config
  - Backend now sets MARKETPLACE permissions based on interface configuration
  - When marketplace enabled: users get agents with EDIT permissions in dropdown lists  (builder mode)
  - When marketplace disabled: users get agents with VIEW permissions  in dropdown lists (browse mode)

🔧 fix: Redirect to New Chat if No Marketplace Access and Required Agent Name Placeholder (#8213)

* Fix: Fix the redirect to new chat page if access to marketplace is denied

* Fixed the required agent name placeholder

---------

Co-authored-by: Atef Bellaaj <slalom.bellaaj@external.daimlertruck.com>

chore: fix tests, remove unnecessary imports

refactor: Implement permission checks for file access via agents

- Updated `hasAccessToFilesViaAgent` to utilize permission checks for VIEW and EDIT access.
- Replaced project-based access validation with permission-based checks.
- Enhanced tests to cover new permission logic and ensure proper access control for files associated with agents.
- Cleaned up imports and initialized models in test files for consistency.

refactor: Enhance test setup and cleanup for file access control

- Introduced modelsToCleanup array to track models added during tests for proper cleanup.
- Updated afterAll hooks in test files to ensure all collections are cleared and only added models are deleted.
- Improved consistency in model initialization across test files.
- Added comments for clarity on cleanup processes and test data management.

chore: Update Jest configuration and test setup for improved timeout handling

- Added a global test timeout of 30 seconds in jest.config.js.
- Configured jest.setTimeout in jestSetup.js to allow individual test overrides if needed.
- Enhanced test reliability by ensuring consistent timeout settings across all tests.

refactor: Implement file access filtering based on agent permissions

- Introduced `filterFilesByAgentAccess` function to filter files based on user access through agents.
- Updated `getFiles` and `primeFiles` functions to utilize the new filtering logic.
- Moved `hasAccessToFilesViaAgent` function from the File model to permission services, adjusting imports accordingly
- Enhanced tests to ensure proper access control and filtering behavior for files associated with agents.

fix: make support_contact field a nested object rather than a sub-document

refactor: Update support_contact field initialization in agent model

- Removed handling for empty support_contact object in createAgent function.
- Changed default value of support_contact in agent schema to undefined.

test: Add comprehensive tests for support_contact field handling and versioning

refactor: remove unused avatar upload mutation field and add informational toast for success

chore: add missing SidePanelProvider for AgentMarketplace and organize imports

fix: resolve agent selection race condition in marketplace HandleStartChat
- Set agent in localStorage before newConversation to prevent useSelectorEffects from auto-selecting previous agent

fix: resolve agent dropdown showing raw ID instead of agent info from URL

  - Add proactive agent fetching when agent_id is present in URL parameters
  - Inject fetched agent into agents cache so dropdowns display proper name/avatar
  - Use useAgentsMap dependency to ensure proper cache initialization timing
  - Prevents raw agent IDs from showing in UI when visiting shared agent links

Fix: Agents endpoint renamed to "My Agent" for less confusion with the Marketplace agents.

chore: fix ESLint issues and Test Mocks

ci: update permissions structure in loadDefaultInterface tests

- Refactored permissions for MEMORY and added new permissions for MARKETPLACE and PEOPLE_PICKER.
- Ensured consistent structure for permissions across different types.

feat:  support_contact validation to allow empty email strings
2025-08-13 16:24:18 -04:00

161 lines
4.7 KiB
TypeScript

import { z } from 'zod';
import { ViolationTypes, ErrorTypes } from 'librechat-data-provider';
import type { Agent, TModelsConfig } from 'librechat-data-provider';
import type { Request, Response } from 'express';
/** Avatar schema shared between create and update */
export const agentAvatarSchema = z.object({
filepath: z.string(),
source: z.string(),
});
/** Base resource schema for tool resources */
export const agentBaseResourceSchema = z.object({
file_ids: z.array(z.string()).optional(),
files: z.array(z.any()).optional(), // Files are populated at runtime, not from user input
});
/** File resource schema extends base with vector_store_ids */
export const agentFileResourceSchema = agentBaseResourceSchema.extend({
vector_store_ids: z.array(z.string()).optional(),
});
/** Tool resources schema matching AgentToolResources interface */
export const agentToolResourcesSchema = z
.object({
image_edit: agentBaseResourceSchema.optional(),
execute_code: agentBaseResourceSchema.optional(),
file_search: agentFileResourceSchema.optional(),
ocr: agentBaseResourceSchema.optional(),
})
.optional();
/** Support contact schema for agent */
export const agentSupportContactSchema = z
.object({
name: z.string().optional(),
email: z.union([z.literal(''), z.string().email()]).optional(),
})
.optional();
/** Base agent schema with all common fields */
export const agentBaseSchema = z.object({
name: z.string().nullable().optional(),
description: z.string().nullable().optional(),
instructions: z.string().nullable().optional(),
avatar: agentAvatarSchema.nullable().optional(),
model_parameters: z.record(z.unknown()).optional(),
tools: z.array(z.string()).optional(),
agent_ids: z.array(z.string()).optional(),
end_after_tools: z.boolean().optional(),
hide_sequential_outputs: z.boolean().optional(),
artifacts: z.string().optional(),
recursion_limit: z.number().optional(),
conversation_starters: z.array(z.string()).optional(),
tool_resources: agentToolResourcesSchema,
support_contact: agentSupportContactSchema,
category: z.string().optional(),
});
/** Create schema extends base with required fields for creation */
export const agentCreateSchema = agentBaseSchema.extend({
provider: z.string(),
model: z.string().nullable(),
tools: z.array(z.string()).optional().default([]),
});
/** Update schema extends base with all fields optional and additional update-only fields */
export const agentUpdateSchema = agentBaseSchema.extend({
provider: z.string().optional(),
model: z.string().nullable().optional(),
projectIds: z.array(z.string()).optional(),
removeProjectIds: z.array(z.string()).optional(),
isCollaborative: z.boolean().optional(),
});
interface ValidateAgentModelParams {
req: Request;
res: Response;
agent: Agent;
modelsConfig: TModelsConfig;
logViolation: (
req: Request,
res: Response,
type: string,
errorMessage: Record<string, unknown>,
score?: number | string,
) => Promise<void>;
}
interface ValidateAgentModelResult {
isValid: boolean;
error?: {
message: string;
};
}
/**
* Validates an agent's model against the available models configuration.
* This is a non-middleware version of validateModel that can be used
* in service initialization flows.
*
* @param params - Validation parameters
* @returns Object indicating whether the model is valid and any error details
*/
export async function validateAgentModel(
params: ValidateAgentModelParams,
): Promise<ValidateAgentModelResult> {
const { req, res, agent, modelsConfig, logViolation } = params;
const { model, provider: endpoint } = agent;
if (!model) {
return {
isValid: false,
error: {
message: `{ "type": "${ErrorTypes.MISSING_MODEL}", "info": "${endpoint}" }`,
},
};
}
if (!modelsConfig) {
return {
isValid: false,
error: {
message: `{ "type": "${ErrorTypes.MODELS_NOT_LOADED}" }`,
},
};
}
const availableModels = modelsConfig[endpoint];
if (!availableModels) {
return {
isValid: false,
error: {
message: `{ "type": "${ErrorTypes.ENDPOINT_MODELS_NOT_LOADED}", "info": "${endpoint}" }`,
},
};
}
const validModel = !!availableModels.find((availableModel) => availableModel === model);
if (validModel) {
return { isValid: true };
}
const { ILLEGAL_MODEL_REQ_SCORE: score = 1 } = process.env ?? {};
const type = ViolationTypes.ILLEGAL_MODEL_REQUEST;
const errorMessage = {
type,
model,
endpoint,
};
await logViolation(req, res, type, errorMessage, score);
return {
isValid: false,
error: {
message: `{ "type": "${ViolationTypes.ILLEGAL_MODEL_REQUEST}", "info": "${endpoint}|${model}" }`,
},
};
}