refactor(agents): relocate boundary-only infra out of shared/

Neither module is imported by any sibling agent package, so neither belongs in
the cross-agent shared kernel:

- checkpointer.py -> app/agents/runtime/checkpointer.py
  LangGraph Postgres checkpoint saver. It's cross-agent *runtime infra* wired by
  the boundary (app lifespan + anonymous_chat & multi_agent_chat flows), not
  agent code. New app/agents/runtime/ layer holds boundary-wired agent infra.

- shared/system_prompt.py + shared/prompts/ -> app/prompts/
  The legacy single-agent prompt composer. The live agents don't use it
  (main_agent has its own system_prompt/ builder; anonymous_chat builds inline);
  its only consumer is new_llm_config_routes for displaying default instructions.
  Moved to the existing non-agent prompt domain:
    system_prompt.py        -> app/prompts/default_system_instructions.py
    prompts/                 -> app/prompts/system_prompt_composer/

app/agents/shared/ now contains only genuinely cross-agent code: context,
middleware/{compaction,retry_after,dedup_tool_calls}, tools/.

NOTE: get_default_system_instructions() (LLM-config UI) composes from the legacy
library, which differs from what the live agents actually run -- pre-existing
latent staleness, not changed here.

surfsense_backend/app/agents/runtime/__init__.py

@@ -0,0 +1,6 @@
+"""Agent runtime infrastructure wired by the boundary (not agent code).
+
+Modules here are cross-agent infra used to *run* agents (e.g. the LangGraph
+Postgres checkpointer), as opposed to ``app/agents/shared/`` which holds code
+imported by 2+ sibling agent packages.
+"""

surfsense_backend/app/app.py

@@ -23,7 +23,7 @@ from starlette.requests import Request as StarletteRequest
 from starlette.responses import Response as StarletteResponse
 from uvicorn.middleware.proxy_headers import ProxyHeadersMiddleware
 
-from app.agents.shared.checkpointer import (
+from app.agents.runtime.checkpointer import (
     close_checkpointer,
     setup_checkpointer_tables,
 )

surfsense_backend/app/prompts/default_system_instructions.py

@@ -1,5 +1,5 @@
 """
-Thin compatibility wrapper around :mod:`app.agents.shared.prompts.composer`.
+Thin compatibility wrapper around :mod:`app.prompts.system_prompt_composer.composer`.
 
 The composer split the previous monolithic prompt string into a fragment
 tree under ``prompts/`` plus a model-family dispatch step (see the
@@ -11,7 +11,7 @@ that existing call sites — the multi-agent chat factory, anonymous chat
 routes, and the configurable-prompt admin path — keep working without churn.
 
 For new call sites prefer importing ``compose_system_prompt`` directly
-from :mod:`app.agents.shared.prompts.composer`.
+from :mod:`app.prompts.system_prompt_composer.composer`.
 """
 
 from __future__ import annotations
@@ -20,7 +20,7 @@ from datetime import UTC, datetime
 
 from app.db import ChatVisibility
 
-from .prompts.composer import (
+from .system_prompt_composer.composer import (
     _read_fragment,
     compose_system_prompt,
     detect_provider_variant,
@@ -55,7 +55,7 @@ def build_surfsense_system_prompt(
 ) -> str:
     """Build the default SurfSense system prompt (citations on, defaults).
 
-    See :func:`app.agents.shared.prompts.composer.compose_system_prompt`
+    See :func:`app.prompts.system_prompt_composer.composer.compose_system_prompt`
     for full parameter docs.
     """
     return compose_system_prompt(
@@ -84,7 +84,7 @@ def build_configurable_system_prompt(
 ) -> str:
     """Build a configurable SurfSense system prompt (NewLLMConfig path).
 
-    See :func:`app.agents.shared.prompts.composer.compose_system_prompt`
+    See :func:`app.prompts.system_prompt_composer.composer.compose_system_prompt`
     for full parameter docs.
     """
     return compose_system_prompt(
@@ -108,7 +108,9 @@ def get_default_system_instructions() -> str:
     The output reflects the current fragment tree, not a baked-in constant.
     """
     resolved_today = datetime.now(UTC).date().isoformat()
-    from .prompts.composer import _build_system_instructions  # local import
+    from .system_prompt_composer.composer import (
+        _build_system_instructions,  # local import
+    )
 
     return _build_system_instructions(
         visibility=ChatVisibility.PRIVATE,

surfsense_backend/app/prompts/system_prompt_composer/composer.py

@@ -2,7 +2,7 @@
 Prompt composer for the SurfSense ``new_chat`` agent.
 
 This module assembles the agent's system prompt from the markdown fragments
-under :mod:`app.agents.shared.prompts`. It replaces the monolithic
+under :mod:`app.prompts.system_prompt_composer`. It replaces the monolithic
 ``system_prompt.py`` with a clean, fragment-based composition:
 
 ::
@@ -119,7 +119,7 @@ def detect_provider_variant(model_name: str | None) -> ProviderVariant:
 # -----------------------------------------------------------------------------
 
 
-_PROMPTS_PACKAGE = "app.agents.shared.prompts"
+_PROMPTS_PACKAGE = "app.prompts.system_prompt_composer"
 
 
 def _read_fragment(subpath: str) -> str:

surfsense_backend/app/routes/anonymous_chat_routes.py

@@ -352,7 +352,7 @@ async def stream_anonymous_chat(
         from langchain_core.messages import AIMessage, HumanMessage
 
         from app.agents.anonymous_chat import create_anonymous_chat_agent
-        from app.agents.shared.checkpointer import get_checkpointer
+        from app.agents.runtime.checkpointer import get_checkpointer
         from app.db import shielded_async_session
         from app.services.new_streaming_service import VercelStreamingService
         from app.services.token_tracking_service import start_turn

surfsense_backend/app/routes/new_chat_routes.py

@@ -1934,7 +1934,7 @@ async def regenerate_response(
     """
     from langchain_core.messages import HumanMessage
 
-    from app.agents.shared.checkpointer import get_checkpointer
+    from app.agents.runtime.checkpointer import get_checkpointer
 
     try:
         # Verify thread exists and user has permission

surfsense_backend/app/routes/new_llm_config_routes.py

@@ -13,7 +13,6 @@ from fastapi import APIRouter, Depends, HTTPException
 from sqlalchemy.ext.asyncio import AsyncSession
 from sqlalchemy.future import select
 
-from app.agents.shared.system_prompt import get_default_system_instructions
 from app.config import config
 from app.db import (
     NewLLMConfig,
@@ -21,6 +20,7 @@ from app.db import (
     User,
     get_async_session,
 )
+from app.prompts.default_system_instructions import get_default_system_instructions
 from app.schemas import (
     DefaultSystemInstructionsResponse,
     GlobalNewLLMConfigRead,

surfsense_backend/app/tasks/chat/streaming/flows/shared/pre_stream_setup.py

@@ -4,7 +4,7 @@ from __future__ import annotations
 
 from sqlalchemy.ext.asyncio import AsyncSession
 
-from app.agents.shared.checkpointer import get_checkpointer
+from app.agents.runtime.checkpointer import get_checkpointer
 from app.db import SearchSourceConnectorType
 from app.services.connector_service import ConnectorService
 
@@ -33,7 +33,7 @@ async def setup_connector_and_firecrawl(
 async def get_chat_checkpointer():
     """Resolve the PostgreSQL checkpointer for persistent conversation memory.
 
-    Thin wrapper around ``app.agents.shared.checkpointer.get_checkpointer`` so
+    Thin wrapper around ``app.agents.runtime.checkpointer.get_checkpointer`` so
     flow orchestrators can rely on a streaming-local symbol and we have a hook
     point if the checkpointer source ever needs to vary per flow.
     """

surfsense_backend/tests/unit/agents/new_chat/prompts/test_composer.py

@@ -6,12 +6,12 @@ from datetime import UTC, datetime
 
 import pytest
 
-from app.agents.shared.prompts.composer import (
+from app.db import ChatVisibility
+from app.prompts.system_prompt_composer.composer import (
     ALL_TOOL_NAMES_ORDERED,
     compose_system_prompt,
     detect_provider_variant,
 )
-from app.db import ChatVisibility
 
 pytestmark = pytest.mark.unit
 
@@ -64,7 +64,7 @@ class TestProviderVariantDetection:
         ``gpt-5`` reasoning regex first. Codex is the more specialised
         prompt and mirrors OpenCode's dispatch order.
         """
-        from app.agents.shared.prompts.composer import detect_provider_variant
+        from app.prompts.system_prompt_composer.composer import detect_provider_variant
 
         assert detect_provider_variant("openai:gpt-5-codex") == "openai_codex"
         assert detect_provider_variant("openai:gpt-5") == "openai_reasoning"