refactor(feedback): move the secondary adapters to app/adapters

Group them by bounded context instead of by technology, one file per
port, and align the directory name with the AWS Prescriptive Guidance
layout (entrypoints / domain-with-ports / adapters).

  app/infra/persistence/feedback.py
    -> app/adapters/feedback/feedback_repository.py   owned persistence
    -> app/adapters/feedback/run_lookup.py            anti-corruption layer

`persistence/` promised a technology-first classification that its own
contents contradicted: RunStoreRunLookup lived there while its docstring
said "no new SQL". Splitting per port makes that distinction structural.

SqlFeedbackRepository and _tz_aware move unchanged -- verified by
comparing their AST against the original rather than by eye. run_lookup.py
additionally gains a RunStore annotation behind TYPE_CHECKING (the module
is imported lazily by the composition root, so this keeps the runtime
import cost at zero), a docstring stating that this context owns no table
and writes no SQL against it, and a TODO recording the condition under
which the body is replaced: when the run context publishes a contract of
its own, the RunLookup port itself does not move.

Each module docstring opens with a fixed marker so the two kinds of
secondary adapter stay greppable:

  grep -rl "anti-corruption layer" app/adapters/

Filenames deliberately carry no sql_ / acl_ prefix: a prefix encodes an
implementation property, so switching storage would force a rename even
though the port -- and therefore the import path -- has not changed. The
class name already carries it. A prefix earns its place once one port has
several production implementations, which is not yet the case here.

app/infra/ held nothing else and is removed.

backend/app/adapters/feedback/feedback_repository.py

@@ -1,11 +1,15 @@
-"""SQL adapters for the feedback bounded context.
+"""Secondary adapter (owned persistence) -- the feedback table in SQL.
 
-Secondary adapters implementing the ports declared in
-``deerflow.domain.feedback.ports``. SQL/ORM vocabulary stops at this
-file: methods exchange domain objects, translate ``IntegrityError`` into
-domain errors, and normalize SQLite's tz-naive reads. Queries were
-migrated unchanged from the legacy repository (now removed) to preserve
-behavior.
+Implements ``FeedbackRepository`` from ``deerflow.domain.feedback.ports``.
+This context owns the ``feedback`` table and writes its own queries, so
+SQL/ORM vocabulary stops at this file: methods exchange domain objects,
+translate ``IntegrityError`` into domain errors, and normalize SQLite's
+tz-naive reads. Queries were migrated unchanged from the legacy repository
+(now removed) to preserve behavior.
+
+Sibling adapter: ``run_lookup.py`` serves the same context but owns no
+table and writes no SQL -- see its docstring for why that distinction is
+worth keeping visible.
 """
 
 from __future__ import annotations
@@ -17,10 +21,10 @@ from sqlalchemy.exc import IntegrityError
 from sqlalchemy.ext.asyncio import AsyncSession, async_sessionmaker
 
 from deerflow.domain.feedback.model import DuplicateFeedbackError, Feedback
-from deerflow.domain.feedback.ports import FeedbackRepository, RunLookup
+from deerflow.domain.feedback.ports import FeedbackRepository
 
 # Transitional: the ORM row stays in the harness until PR-N moves engine,
-# models, and migrations into app/infra.
+# models, and migrations into app/adapters.
 from deerflow.persistence.feedback.model import FeedbackRow
 
 
@@ -137,18 +141,3 @@ class SqlFeedbackRepository(FeedbackRepository):
             await session.delete(row)
             await session.commit()
             return True
-
-
-class RunStoreRunLookup(RunLookup):
-    """Adapts the framework ``RunStore`` (wide interface) to the narrow
-    ``RunLookup`` port -- reuses the existing lookup, no new SQL.
-
-    Used by the service for run-ownership checks before writing feedback.
-    """
-
-    def __init__(self, run_store) -> None:
-        self._run_store = run_store
-
-    async def thread_of(self, run_id: str) -> str | None:
-        run = await self._run_store.get(run_id)
-        return run.get("thread_id") if run else None

backend/app/adapters/feedback/run_lookup.py

@@ -0,0 +1,47 @@
+"""Secondary adapter (anti-corruption layer) -- narrowing RunStore to RunLookup.
+
+Implements ``RunLookup`` from ``deerflow.domain.feedback.ports``. Unlike its
+sibling ``feedback_repository.py``, this context does not own the ``runs``
+table and writes no SQL against it: it asks its one question through the
+component the run context already provides.
+
+Being the downstream consumer of an upstream context is the normal shape
+here, not debt. Until that context publishes a contract of its own, a thin
+translation layer is what keeps its wide interface -- and its untyped dict
+rows -- out of the domain.
+"""
+
+from __future__ import annotations
+
+from typing import TYPE_CHECKING
+
+from deerflow.domain.feedback.ports import RunLookup
+
+if TYPE_CHECKING:
+    from deerflow.runtime.runs.store import RunStore
+
+
+class RunStoreRunLookup(RunLookup):
+    """Adapts the framework ``RunStore`` (wide interface) to the narrow
+    ``RunLookup`` port -- reuses the existing lookup, no new SQL.
+
+    Used by the service to confirm a run belongs to the thread before
+    writing feedback. That is referential integrity, not authorization:
+    the router's ``@require_permission(..., owner_check=True)`` proves the
+    caller owns the thread, and this check proves the run belongs to it --
+    only together do they establish that the caller owns the run. Which is
+    why the port takes no ``user_id``.
+
+    TODO(hexagonal): this depends on ``RunStore``, an infrastructure
+    component, rather than on a contract published by the run context --
+    that context has not been through a hexagonal slice yet. When it
+    publishes one (a DTO, not its aggregate and not its repository),
+    replace the body of this class. The ``RunLookup`` port does not move.
+    """
+
+    def __init__(self, run_store: RunStore) -> None:
+        self._run_store = run_store
+
+    async def thread_of(self, run_id: str) -> str | None:
+        run = await self._run_store.get(run_id)
+        return run.get("thread_id") if run else None

backend/app/gateway/deps.py

@@ -400,10 +400,8 @@ async def langgraph_runtime(app: FastAPI, startup_config: AppConfig) -> AsyncGen
         # Initialize repositories — one get_session_factory() call for all.
         sf = get_session_factory()
         if sf is not None:
-            from app.infra.persistence.feedback import (
-                RunStoreRunLookup,
-                SqlFeedbackRepository,
-            )
+            from app.adapters.feedback.feedback_repository import SqlFeedbackRepository
+            from app.adapters.feedback.run_lookup import RunStoreRunLookup
             from deerflow.persistence.run import RunRepository
 
             app.state.run_store = RunRepository(sf)

backend/tests/test_feedback.py

@@ -105,7 +105,7 @@ class TestSqlFeedbackRepository(FeedbackRepositoryContract):
         from sqlalchemy.ext.asyncio import async_sessionmaker, create_async_engine
 
         import deerflow.persistence.feedback.model  # noqa: F401  (register FeedbackRow)
-        from app.infra.persistence.feedback import SqlFeedbackRepository
+        from app.adapters.feedback.feedback_repository import SqlFeedbackRepository
         from deerflow.persistence.base import Base
 
         engine = create_async_engine(f"sqlite+aiosqlite:///{tmp_path / 'test.db'}")

backend/tests/test_owner_isolation.py

@@ -336,7 +336,7 @@ async def test_run_events_cross_user_delete_denied(tmp_path):
 async def test_feedback_cross_user_isolation(tmp_path):
     """Ownership is an explicit user_id filter on the new feedback port —
     the AUTO-sentinel context no longer applies (hexagonal slice)."""
-    from app.infra.persistence.feedback import SqlFeedbackRepository
+    from app.adapters.feedback.feedback_repository import SqlFeedbackRepository
     from deerflow.domain.feedback import Feedback
     from deerflow.persistence.engine import get_session_factory
 
@@ -367,7 +367,7 @@ async def test_feedback_cross_user_isolation(tmp_path):
 @pytest.mark.anyio
 @pytest.mark.no_auto_user
 async def test_feedback_cross_user_delete_denied(tmp_path):
-    from app.infra.persistence.feedback import SqlFeedbackRepository
+    from app.adapters.feedback.feedback_repository import SqlFeedbackRepository
     from deerflow.domain.feedback import Feedback
     from deerflow.persistence.engine import get_session_factory
 

backend/tests/test_persistence_timezone.py

@@ -76,7 +76,7 @@ async def test_run_repository_emits_tz_aware_timestamps(tmp_path):
 
 @pytest.mark.anyio
 async def test_feedback_repository_emits_tz_aware_timestamps(tmp_path):
-    from app.infra.persistence.feedback import SqlFeedbackRepository
+    from app.adapters.feedback.feedback_repository import SqlFeedbackRepository
     from deerflow.domain.feedback import Feedback
 
     repo = SqlFeedbackRepository(await _init_sqlite(tmp_path))