fix(core): fix passing sendSignal through ProcessorRunner and update temporal markers to use sendSignal (#16927)

## Description

Adds integration tests for the `sendSignal()` processor API and the
`AgentsMDInjector` auto-loading processor, exercising the real
`ProcessorRunner` + `MessageList` path (no mocks/test doubles for the
message list).

**Key finding:** All 16 tests pass at the core processor level,
indicating the `sendSignal` mechanism and `AgentsMDInjector` work
correctly through `ProcessorRunner` with real `MessageList`. The
AGENTS.md auto-loading regression reported in mastracode likely
originates at a higher layer (harness, agent loop, or message
persistence/rehydration).

### sendSignal-integration.test.ts (9 tests)
- Signal from `processInputStep` appears in LLM prompt
- Signal metadata preserved through `MessageList` round-trip
- Data part emitted to stream writer (`data-system-reminder`)
- Response message ID rotation before signal addition
- Multiple processors' signals all preserved in message list
- `sendSignal` available in `processOutputResult` context
- `sendSignal` available in `processAPIError` context
- Works without writer
- Works without `rotateResponseMessageId`

### agents-md-autoload-integration.test.ts (7 tests)
- Auto-loads AGENTS.md when tool call references path near an
instruction file
- Injected AGENTS.md content appears in LLM prompt (via signal → prompt
projection)
- Stream data part emitted for the AGENTS.md signal
- Duplicate suppression across multiple processor runs (signal-based
dedup)
- Directory-based tool calls trigger injection
- No injection when tool call path has no nearby AGENTS.md
- Injection survives across multiple LLM steps

## Related issue(s)

Investigates AGENTS.md auto-loading regression after the signals
refactor (PR #16438).

## Type of change

- [ ] Bug fix (non-breaking change that fixes an issue)
- [ ] New feature (non-breaking change that adds functionality)
- [ ] Breaking change (fix or feature that would cause existing
functionality to change)
- [ ] Documentation update
- [ ] Code refactoring
- [ ] Performance improvement
- [x] Test update

## Checklist

- [x] I have linked the related issue(s) in the description above
- [ ] I have made corresponding changes to the documentation (if
applicable)
- [x] I have added tests that prove my fix is effective or that my
feature works
- [ ] I have addressed all Coderabbit comments on this PR

Link to Devin session:
https://app.devin.ai/sessions/7f81a6b1c3454fe78fe7e6a730cb744d
Requested by: @TylerBarnes

<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->
## ELI5

This PR adds integration tests that confirm processors can emit
persistent "signal" reminders (visible to the model and optionally
streamed) and that AGENTS.md files found near tool-call paths are
auto-injected as those reminders; it also wires a sendSignal helper into
workflow paths so processors get the same sendSignal API as
ProcessorRunner.

## Overview

- Adds 16 Vitest integration tests exercising sendSignal() and
AgentsMDInjector through the real ProcessorRunner + MessageList (no
MessageList mocks).
- Ensures processors invoked via workflow bridges receive a sendSignal
API by exporting and wiring createProcessorSendSignal into workflow/step
execution.
- Tests indicate sendSignal and AgentsMDInjector behave correctly at the
core processor level; the AGENTS.md auto-load regression likely
originates above core (harness/agent loop, or message
persistence/rehydration).

## Test coverage

- sendSignal integration (9 tests —
packages/core/src/processors/sendSignal-integration.test.ts)
- Verifies sendSignal from processInputStep persists a `signal`-role
message and appears in the constructed LLM prompt.
- Confirms signal metadata (type/attributes/path) survives a real
MessageList round-trip.
- Verifies sendSignal emits a `data-system-reminder` stream part when a
writer is provided.
- Asserts response message ID rotation occurs before the signal is added
and propagates to the step result.
- Multiple processors can emit signals and all are preserved in prompts.
- sendSignal is available in processOutputResult and processAPIError
contexts.
- Signals persist even without a writer and when rotateResponseMessageId
is not supplied.

- AgentsMDInjector auto-load integration (7 tests —
packages/core/src/processors/agents-md-autoload-integration.test.ts)
- Confirms AGENTS.md near tool-call paths is detected and injected once
as a `system-reminder` signal with `dynamic-agents-md` metadata.
- Injected AGENTS.md content appears in later LLM prompt text via
signal→prompt projection.
- A `data-system-reminder` stream chunk is emitted for injected
AGENTS.md signals.
- Duplicate suppression prevents re-injection across subsequent
processor runs (signal-based dedup).
- Directory-based tool-call arguments trigger injection; no injection
occurs when no nearby AGENTS.md (readFile not called).
- Injection survives across multiple LLM steps and persisted tool
results.

## Code changes

- Tests added
- packages/core/src/processors/sendSignal-integration.test.ts (+409
lines)
- packages/core/src/processors/agents-md-autoload-integration.test.ts
(+539 lines)
- Changeset
  - .changeset/khaki-cougars-hammer.md (patch bump for `@mastra/core`)
- Production code
  - packages/core/src/processors/runner.ts
- Exports createProcessorSendSignal (previously local) so workflows can
construct sendSignal.
  - packages/core/src/processors/index.ts
    - Re-exports createProcessorSendSignal.
- packages/core/src/workflows/workflow.ts and
packages/core/src/workflows/evented/workflow.ts
- Inject sendSignal into processor baseContext when a messageList is
present, created with rotateResponseMessageId and the processor writer.
  - packages/core/src/workflows/processor-step.test.ts
- New unit test ensuring createStep provides sendSignal to processors
and that rotateResponseMessageId and writer emission behave as expected.
  - packages/core/src/processors/tool-result-reminder.ts
- Minor refactor: uses getCurrentStepResponseMessages(messageList) to
compute completed tool calls.
- (Behavior-preserving) fallback persistence path preserved so reminders
are recorded/streamed even when args.sendSignal is absent.

## Impact & scope

- ~947 lines of tests added; small production surface change to export
and wire sendSignal into workflow contexts.
- No breaking public API changes; one new re-export
(createProcessorSendSignal).
- Confirms core-level sendSignal and AgentsMDInjector functionality;
remaining AGENTS.md regression likely lies in higher-level orchestration
or persistence/rehydration layers.

## Related

- Investigates AGENTS.md auto-load regression after signals refactor
(related to PR `#16438`). Devin session attached in PR for reproduction
details.

<!-- review_stack_entry_start -->

[![Review Change
Stack](https://storage.googleapis.com/coderabbit_public_assets/review-stack-in-coderabbit-ui.svg)](https://app.coderabbit.ai/change-stack/mastra-ai/mastra/pull/16927?utm_source=github_walkthrough&utm_medium=github&utm_campaign=change_stack)

<!-- review_stack_entry_end -->
<!-- end of auto-generated comment: release notes by coderabbit.ai -->

---------

Co-authored-by: Devin AI <158243242+devin-ai-integration[bot]@users.noreply.github.com>
Co-authored-by: tyler <tylerdbarnes@gmail.com>

Author: devin-ai-integration[bot]

.changeset/khaki-cougars-hammer.md

@@ -0,0 +1,6 @@
+---
+'@mastra/core': patch
+'@mastra/memory': patch
+---
+
+Fixed processor workflow steps so `sendSignal` is available when processors inject Agent signals, and updated Observational Memory temporal gap markers to use Agent signals.

packages/core/src/agent/__tests__/agent-signals.test.ts

@@ -126,6 +126,7 @@ describe('Agent signals', () => {
         attributes: { priority: 'high' },
         metadata: { source: 'test', signal: { userProvided: true } },
       },
+      transient: true,
     });
 
     const dbMessage = signal.toDBMessage({ threadId: 'thread-1', resourceId: 'resource-1' });
@@ -689,6 +690,7 @@ describe('Agent signals', () => {
       acceptedAt: signalResult.signal.acceptedAt?.toISOString(),
     });
     expect(signalPart?.data.createdAt).toBeDefined();
+    expect(signalPart?.transient).toBe(true);
 
     subscription.unsubscribe();
   });

packages/core/src/agent/signals.ts

@@ -53,6 +53,7 @@ export type AgentSignalDataPart = {
     attributes?: AgentSignalAttributes;
     metadata?: Record<string, unknown>;
   };
+  transient: true;
 };
 
 /**
@@ -372,6 +373,7 @@ function signalToDataPart(signal: ReturnType<typeof normalizeSignal>, parts: Sig
       ...(signal.attributes ? { attributes: signal.attributes } : {}),
       ...(signal.metadata ? { metadata: signal.metadata } : {}),
     },
+    transient: true,
   };
 }