feat: Refactor agent request context handling and improve agent definitions

- Introduced a new `request-context.ts` file to centralize request context management for agents.
- Created a `BaseAgentRequestContext` interface to standardize user tier and language handling across agents.
- Implemented utility functions `getUserTierFromContext` and `getLanguageFromContext` to simplify context retrieval.
- Updated all agents (reportAgent, researchAgent, researchPaperAgent, scriptWriterAgent, seoAgent, socialMediaAgent, stockAnalysisAgent, translationAgent, weatherAgent, webResearchAgent) to use the new context handling methods, removing redundant code.
- Replaced `asNestedAgents` usage in network definitions with direct agent assignments to improve type safety and clarity.
- Added a new `chat-request.ts` file defining a schema for chat transport requests using Zod for validation.
- Updated documentation in `AGENTS.md` to reflect changes in agent wiring and context handling.
- Enhanced type definitions for agent contexts to support additional properties and improve type safety.

Author: ssdeanx

app/chat/providers/chat-context.tsx

@@ -163,7 +163,7 @@ export function ChatProvider({
                             id: selectedAgent,
                             messages: outgoingMessages,
                             parts: outgoingMessages.flatMap(
-                                (m) => (m.parts ?? []) as UIMessage['parts']
+                                (m) => (m.parts ?? [])
                             ),
                             trigger,
                             messageId,

lib/types/chat-request.ts

@@ -0,0 +1,28 @@
+import { z } from 'zod'
+
+export const chatTransportRequestSchema = z
+  .object({
+    id: z.string().optional(),
+    messages: z.array(z.unknown()).default([]),
+    trigger: z.string().optional(),
+    messageId: z.string().optional(),
+    requestMetadata: z.unknown().optional(),
+    requestContext: z.record(z.string(), z.unknown()).optional(),
+    memory: z
+      .object({
+        thread: z.string().optional(),
+        resource: z.string().optional(),
+      })
+      .partial()
+      .optional(),
+    data: z
+      .object({
+        agentId: z.string().min(1).optional(),
+        input: z.string().optional(),
+      })
+      .loose()
+      .optional(),
+  })
+  .loose()
+
+export type ChatTransportRequest = z.infer<typeof chatTransportRequestSchema>

memory-bank/activeContext.md

@@ -1,3 +1,61 @@
+## Active Context Update (2026-03-17 - codingAgents explicit tool-generic cleanup)
+
+- User flagged that `codingAgents.ts` still contained an explicit `new Agent<...>` constructor and that the per-agent tool objects had been disturbed.
+- Fixed `src/mastra/agents/codingAgents.ts` by:
+  - removing the remaining explicit generic constructor from `codeReviewerAgent`
+  - restoring the local per-agent tool objects (`codeArchitectTools`, `codeReviewerTools`, `testEngineerTools`, `refactoringTools`) with concrete tool names
+  - normalizing `codeArchitectAgent` model selection back to shared request-context helper usage rather than raw string access
+  - removing the stray unused `scrapingSchedulerTool` import
+- Validation:
+  - targeted `get_errors` on `src/mastra/agents/codingAgents.ts` returned **No errors found**
+  - folder-level `get_errors` on `src/mastra/agents` returned **No errors found**
+
+## Active Context Update (2026-03-17 - networks/a2a nested-agent cleanup completed without ToolsInput adaptering)
+
+- Completed the follow-up fix pass for the remaining `src/mastra/networks` and `src/mastra/a2a` nested child-agent assignment failures.
+- User direction was explicit: do **not** rely on any shared adapter or `as Record<string, Agent>` cast, and do **not** introduce gratuitous `ToolsInput` typing into agents.
+- Source-level fix applied across the remaining child agents by:
+  - removing narrowed `RequestContext<SpecificContext>` instruction callback annotations from public agent surfaces
+  - pinning affected child agents to public `new Agent<..., unknown>(...)` request-context generics
+  - keeping specialized runtime-context parsing local inside instruction bodies via explicit narrowing/constants
+  - removing the extra `ToolsInput` usage introduced in the touched agents; tool maps now use direct inferred object types (`typeof toolMap`) or `Record<string, never>` for tool-less agents
+- Additional cleanup included:
+  - `codingAgents.ts` refactoring-tool typing cleanup
+  - `codingA2ACoordinator.ts` / `a2aCoordinatorAgent.ts` unused-parameter cleanup
+  - `codingTeamNetwork.ts` synchronous quality gate fix
+- Source agents updated in this pass:
+  - `dataExportAgent.ts`
+  - `dataIngestionAgent.ts`
+  - `dataTransformationAgent.ts`
+  - `reportAgent.ts`
+  - `stockAnalysisAgent.ts`
+  - `recharts.ts`
+  - `researchPaperAgent.ts`
+  - `documentProcessingAgent.ts`
+  - `learningExtractionAgent.ts`
+  - `scriptWriterAgent.ts`
+  - `package-publisher.ts`
+  - plus cleanup in previously touched `editorAgent.ts`, `copywriterAgent.ts`, `knowledgeIndexingAgent.ts`, `researchAgent.ts`, `contentStrategistAgent.ts`, `evaluationAgent.ts`, and `codingAgents.ts`
+- Validation result:
+  - targeted `get_errors` returned clean for all edited agent files
+  - targeted `get_errors` for `src/mastra/a2a` and `src/mastra/networks` no longer surfaced the earlier nested-agent assignment failures
+
+## Active Context Update (2026-03-17 - nested agent typing without adapter)
+
+- Resolved the `seoAgent.ts` nested child-agent type errors without using a shared adapter or local cast.
+- Root cause: Mastra inferred narrower child `Agent<..., TRequestContext>` generics from `researchAgent`, `contentStrategistAgent`, and `evaluationAgent`, which made them fail assignment to parent `agents: Record<string, Agent<..., unknown>>` slots.
+- Fix applied:
+  - centralized shared request-context keys in `src/mastra/agents/request-context.ts`
+  - replaced raw string usage in touched agents with declared constants/helpers
+  - explicitly pinned the public child-agent generic to `unknown` via `new Agent<..., unknown>(...)`
+  - kept runtime-safe parsing of specialized context values inside the agent instruction bodies
+- Files updated:
+  - `src/mastra/agents/request-context.ts`
+  - `src/mastra/agents/researchAgent.ts`
+  - `src/mastra/agents/contentStrategistAgent.ts`
+  - `src/mastra/agents/evaluationAgent.ts`
+- Validation: targeted `get_errors` is clean for all edited files and `src/mastra/agents/seoAgent.ts`.
+
 ## Active Context Update (2026-03-16 - use-mastra-query full client-js surface expansion)
 
 - Expanded `lib/hooks/use-mastra-query.ts` beyond dataset/eval coverage to expose the remaining installed `@mastra/client-js` surfaces needed by the frontend hook factory.

memory-bank/progress.md

@@ -1,3 +1,41 @@
+# 2026-03-17 codingAgents normalization
+
+- Restored `codingAgents.ts` after a bad intermediate state:
+  - local tool objects are back in place by concrete tool name
+  - no explicit `new Agent<...>` tool generic remains in that file
+  - targeted validation for `codingAgents.ts` is clean
+- Rechecked `src/mastra/agents` after the fix:
+  - ✅ **No errors found**
+
+# 2026-03-17 full nested-agent source fix for networks/a2a
+
+- Completed the remaining nested-agent typing cleanup the user requested across `src/mastra/networks` and `src/mastra/a2a`.
+- Removed the bad `ToolsInput` additions from the touched agent implementations.
+- Reworked the remaining child agents so parent registration compiles by source design rather than parent-side casting/adaptering.
+- Fixed adjacent diagnostics encountered during the pass:
+  - invalid `unknown` template interpolation in `knowledgeIndexingAgent.ts`
+  - `codingAgents.ts` unused import + `projectRoot` stringification issue
+  - unused `requestContext` / `userId` in A2A coordinators
+  - `codingTeamNetwork.ts` async/no-await quality-gate bug
+- Validation:
+  - ✅ edited agent files clean
+  - ✅ `src/mastra/a2a` clean
+  - ✅ `src/mastra/networks` clean
+
+# 2026-03-17 nested-agent typing fix without adapter
+
+- Fixed the reported `seoAgent.ts` nested child-agent assignment failures for:
+  - `researchAgent`
+  - `contentStrategistAgent`
+  - `evaluationAgent`
+- Kept the solution at the source rather than adding a parent-side adapter:
+  - shared request-context keys now come from `src/mastra/agents/request-context.ts`
+  - touched child agents parse specialized runtime context internally
+  - touched child agents now expose `unknown` as their public `Agent` request-context generic so direct parent registration compiles cleanly
+- Validation:
+  - ✅ targeted `get_errors` on `request-context.ts`, `researchAgent.ts`, `contentStrategistAgent.ts`, `evaluationAgent.ts`, and `seoAgent.ts` returned clean
+  - ⚠️ folder-wide `get_errors` on `src/mastra/agents` still shows unrelated pre-existing diagnostics in `knowledgeIndexingAgent.ts`
+
 # 2026-03-16 use-mastra-query full client-js hook expansion
 
 - Completed the requested `@mastra/client-js` parity pass in `lib/hooks/use-mastra-query.ts` without replacing the file’s existing hook-factory pattern.

src/mastra/a2a/AGENTS.md

@@ -100,15 +100,16 @@ coordinate_a2a_task({
 
 ## Typing Note
 
-- Parent coordinators that register child agents should use `@/src/mastra/agents/nestedAgents` for the `agents` config.
-- This keeps the current Mastra nested-agent typing workaround isolated to one boundary while upstream typings remain restrictive.
+- Do **not** rely on parent-side adapter helpers or `as Record<string, Agent>` casts for child-agent registration.
+- Fix child agents at the source by exposing the public `Agent<..., unknown>` request-context generic and keeping any specialized runtime-context parsing inside the agent implementation.
 
 ---
 
 ## Change Log
 
 | Version | Date (UTC) | Changes                                         |
 | ------- | ---------- | ----------------------------------------------- |
+| 2.0.2   | 2026-03-17 | Updated guidance to use source-level public `unknown` request-context generics for child agents instead of adapter-based registration. |
 | 2.0.1   | 2026-03-16 | Added note about `nestedAgents` boundary adapter for child-agent registration typing. |
 | 2.0.0   | 2025-11-26 | Major update: documented all coordinated agents |
 | 1.0.0   | 2025-11-14 | Initial A2A coordinator implementation          |

src/mastra/a2a/a2aCoordinatorAgent.ts

@@ -10,7 +10,6 @@ import {
   refactoringAgent,
   testEngineerAgent,
 } from '../agents/codingAgents'
-import { asNestedAgents } from '@/src/mastra/agents/nestedAgents'
 import { contentStrategistAgent } from '../agents/contentStrategistAgent'
 import { copywriterAgent } from '../agents/copywriterAgent'
 import { editorAgent } from '../agents/editorAgent'
@@ -40,7 +39,7 @@ export const a2aCoordinatorAgent = new Agent({
   name: 'a2aCoordinator',
   description:
     'A2A Coordinator that orchestrates multiple specialized agents in parallel. Routes tasks dynamically, coordinates workflows, and synthesizes results using the A2A protocol.',
-  instructions: ({ requestContext }) => {
+  instructions: () => {
    // const userId = requestContext.get('userId')
     return {
       role: 'system',
@@ -108,7 +107,7 @@ Use knowledgeIndexingAgent to provide semantic context for complex queries.
   model: googleAI,
   memory: pgMemory,
   options: {},
-  agents: asNestedAgents({
+  agents: {
     researchAgent,
     knowledgeIndexingAgent,
     editorAgent,
@@ -119,7 +118,7 @@ Use knowledgeIndexingAgent to provide semantic context for complex queries.
     refactoringAgent,
     contentStrategistAgent,
     projectManagementAgent,
-  }),
+  },
   workflows: {
     researchSynthesisWorkflow,
     repoIngestionWorkflow,

src/mastra/a2a/codingA2ACoordinator.ts

@@ -16,7 +16,6 @@ import {
   refactoringAgent,
   testEngineerAgent,
 } from '../agents/codingAgents'
-import { asNestedAgents } from '@/src/mastra/agents/nestedAgents'
 import {
   checkFileExists,
   createDirectory,
@@ -59,8 +58,7 @@ export const codingA2ACoordinator = new Agent({
   name: 'Coding A2A Coordinator',
   description:
     'A2A Coordinator that orchestrates multiple coding agents in parallel for complex development tasks like full feature development, comprehensive reviews, and refactoring with tests.',
-  instructions: ({ requestContext }) => {
-    const userId = requestContext.get('userId')
+  instructions: () => {
     return {
       role: 'system',
       content: `You are a Coding A2A (Agent-to-Agent) Coordinator that orchestrates multi-agent coding workflows.
@@ -206,12 +204,12 @@ When a user's request requires prolonged, structured work across multiple subtas
   model: googleAIFlashLite,
   memory: pgMemory,
 
-  agents: asNestedAgents({
+  agents: {
     codeArchitectAgent,
     codeReviewerAgent,
     testEngineerAgent,
     refactoringAgent,
-  }),
+  },
   workflows: {
     researchSynthesisWorkflow,
     financialReportWorkflow,

src/mastra/agents/AGENTS.md

@@ -90,7 +90,8 @@ This directory contains 22+ agent definitions that map use-case intents to seque
 - **Logging**: Use the provided logger for consistent logging
 - **Testing**: Write unit tests for all agent functionality
 - **Documentation**: Document your agent's purpose, inputs, and outputs
-- **Nested Agent Registration**: When an agent exposes child agents via the `agents` config, use `@/src/mastra/agents/nestedAgents` to adapt Mastra's current nested-agent typing without spreading one-off workarounds across files.
+- **Nested Agent Registration**: Prefer fixing child-agent public generics at the source. If a child agent needs specialized runtime context internally, keep that parsing inside its instruction/helpers while pinning the public `Agent<..., unknown>` request-context generic so parent `agents` registration remains directly assignable.
+- **Tool Typing**: Avoid adding `ToolsInput` annotations to agent definitions unless they are truly required. Prefer inferred tool maps (`const tools = { ... }`) with `typeof tools`, or `Record<string, never>` for tool-less agents.
 
 ## Execution & Testing
 
@@ -121,6 +122,9 @@ npm test src/mastra/__tests__/agents/your-agent.test.ts
 | Version | Date (UTC) | Changes                                                                                                                              |
 | ------- | ---------- | ------------------------------------------------------------------------------------------------------------------------------------ |
 | 2.2.0   | 2025-12-15 | Added 5 new specialized agents: socialMediaAgent, seoAgent, translationAgent, customerSupportAgent, projectManagementAgent.          |
+| 2.2.4   | 2026-03-17 | Normalized `codingAgents.ts` by restoring concrete per-agent tool maps and removing the last explicit `new Agent<...>` tool generic; `src/mastra/agents` now validates clean. |
+| 2.2.3   | 2026-03-17 | Removed unnecessary `ToolsInput` usage from touched agents and completed the source-level `unknown` request-context fix for all remaining network/A2A child-agent registrations. |
+| 2.2.2   | 2026-03-17 | Fixed direct nested-agent registration typing by centralizing request-context keys and pinning touched child agents to public `unknown` request-context generics instead of using an adapter in parent registrations. |
 | 2.2.1   | 2026-03-16 | Added `nestedAgents.ts` boundary adapter for Mastra nested sub-agent typing and cleaned folder-level agent errors.                  |
 | 2.1.0   | 2025-11-28 | Added 4 Financial Chart agents: chartTypeAdvisorAgent, chartDataProcessorAgent, chartGeneratorAgent, chartSupervisorAgent.           |
 | 2.0.0   | 2025-11-26 | Major update: 22 agents documented. Added data pipeline, research paper, document processing, knowledge indexing agents.             |

src/mastra/agents/academicResearchAgent.ts

@@ -1,6 +1,7 @@
 import type { GoogleGenerativeAIProviderOptions } from '@ai-sdk/google'
 import { Agent } from '@mastra/core/agent'
 import type { RequestContext } from '@mastra/core/request-context'
+import type { AgentRequestContext } from './request-context'
 
 import { google } from '../config/google'
 import { log } from '../config/logger'
@@ -18,14 +19,9 @@ import {
 import { extractLearningsTool } from '../tools/extractLearningsTool'
 import { InternalSpans } from '@mastra/core/observability'
 
-type UserTier = 'free' | 'pro' | 'enterprise'
-
-export interface AcademicResearchRuntimeContext {
-    'user-tier': UserTier
-    language: 'en' | 'es' | 'ja' | 'fr'
-    userId?: string
+export type AcademicResearchRuntimeContext = AgentRequestContext<{
     researchType?: 'quick' | 'deep' | 'systematic'
-}
+}>
 
 log.info('Initializing Academic Research Agent...')
 

src/mastra/agents/acpAgent.ts

@@ -38,11 +38,11 @@ import type { RequestContext } from '@mastra/core/request-context'
 import { PGVECTOR_PROMPT } from '@mastra/pg'
 import { TokenLimiterProcessor } from '@mastra/core/processors'
 import { InternalSpans } from '@mastra/core/observability'
+import type { AgentRequestContext } from './request-context'
 
-export interface ACPContext {
-    userId?: string
+export type ACPContext = AgentRequestContext<{
     userRole?: string
-}
+}>
 
 export const acpAgent = new Agent({
     id: 'acpAgent',