diff --git a/src/clio_agent/arc/cache.py b/src/clio_agent/arc/cache.py
index 24cb9e1..1c74e4e 100644
--- a/src/clio_agent/arc/cache.py
+++ b/src/clio_agent/arc/cache.py
@@ -6,6 +6,7 @@
 
 try:
     from lru import LRU  # lru-dict package
+
     HAS_LRU_DICT = True
 except ImportError:
     HAS_LRU_DICT = False
diff --git a/src/clio_agent/arc/index.py b/src/clio_agent/arc/index.py
index 0985714..a4ae43f 100644
--- a/src/clio_agent/arc/index.py
+++ b/src/clio_agent/arc/index.py
@@ -71,10 +71,7 @@ def search(self, key: Tuple[str, float]) -> Optional[Any]:
         return self._index.get(key)
 
     def range_query(
-        self,
-        start_key: Tuple[str, float],
-        end_key: Tuple[str, float],
-        inclusive: bool = True
+        self, start_key: Tuple[str, float], end_key: Tuple[str, float], inclusive: bool = True
     ) -> List[Any]:
         """
         Retrieve all values within key range [start_key, end_key].
@@ -103,13 +100,13 @@ def range_query(
             ... )
         """
         # irange returns keys, so we need to extract values
-        return [self._index[k] for k in self._index.irange(start_key, end_key, inclusive=(True, inclusive))]
+        return [
+            self._index[k]
+            for k in self._index.irange(start_key, end_key, inclusive=(True, inclusive))
+        ]
 
     def range_query_keys(
-        self,
-        start_key: Tuple[str, float],
-        end_key: Tuple[str, float],
-        inclusive: bool = True
+        self, start_key: Tuple[str, float], end_key: Tuple[str, float], inclusive: bool = True
     ) -> List[Tuple[str, float]]:
         """
         Retrieve all keys within range [start_key, end_key].
@@ -135,10 +132,7 @@ def range_query_keys(
         return list(self._index.irange(start_key, end_key, inclusive=(True, inclusive)))
 
     def range_query_items(
-        self,
-        start_key: Tuple[str, float],
-        end_key: Tuple[str, float],
-        inclusive: bool = True
+        self, start_key: Tuple[str, float], end_key: Tuple[str, float], inclusive: bool = True
     ) -> List[Tuple[Tuple[str, float], Any]]:
         """
         Retrieve all (key, value) pairs within range.
@@ -162,7 +156,10 @@ def range_query_items(
             ...     print(f"{key}: {value}")
         """
         # Return (key, value) tuples for all keys in range
-        return [(k, self._index[k]) for k in self._index.irange(start_key, end_key, inclusive=(True, inclusive))]
+        return [
+            (k, self._index[k])
+            for k in self._index.irange(start_key, end_key, inclusive=(True, inclusive))
+        ]
 
     def delete(self, key: Tuple[str, float]) -> bool:
         """
@@ -249,10 +246,7 @@ def get_session_range(self, session_id: str) -> List[Any]:
         Examples:
             >>> conversations = index.get_session_range("session_1")
         """
-        return self.range_query(
-            (session_id, 0.0),
-            (session_id, float('inf'))
-        )
+        return self.range_query((session_id, 0.0), (session_id, float("inf")))
 
     def get_latest_in_session(self, session_id: str, n: int = 1) -> List[Any]:
         """
@@ -279,11 +273,9 @@ def get_latest_in_session(self, session_id: str, n: int = 1) -> List[Any]:
             >>> recent = index.get_latest_in_session("session_1", n=5)
         """
         # Get keys in reverse order (most recent first)
-        all_keys = list(self._index.irange(
-            (session_id, 0.0),
-            (session_id, float('inf')),
-            reverse=True
-        ))
+        all_keys = list(
+            self._index.irange((session_id, 0.0), (session_id, float("inf")), reverse=True)
+        )
         # Take first n keys (the n most recent)
         latest_keys = all_keys[:n]
         # Return values in chronological order (oldest to newest)
diff --git a/src/clio_agent/arc/lsm.py b/src/clio_agent/arc/lsm.py
index 59eaed8..4d3a4f5 100644
--- a/src/clio_agent/arc/lsm.py
+++ b/src/clio_agent/arc/lsm.py
@@ -230,9 +230,7 @@ def range_scan(self, start_ts: float, end_ts: float) -> List[Dict[str, Any]]:
                 if end_ts < sstable.min_key or start_ts > sstable.max_key:
                     continue
 
-                sstable_results = self._range_scan_sstable(
-                    sstable, start_ts, end_ts
-                )
+                sstable_results = self._range_scan_sstable(sstable, start_ts, end_ts)
                 for ts, metric in sstable_results.items():
                     # MemTable has priority (newer data)
                     if ts not in results:
diff --git a/src/clio_agent/arc/storage.py b/src/clio_agent/arc/storage.py
index 099a3cd..43092ab 100644
--- a/src/clio_agent/arc/storage.py
+++ b/src/clio_agent/arc/storage.py
@@ -66,8 +66,8 @@ def __init__(
 
         # Tier migration policy (days)
         self.tier_policy = tier_policy or {
-            "hot_to_warm": 1,      # 1 day in hot tier before eviction
-            "warm_to_cold": 7,     # 1 week in warm tier
+            "hot_to_warm": 1,  # 1 day in hot tier before eviction
+            "warm_to_cold": 7,  # 1 week in warm tier
             "cold_to_archive": 30,  # 1 month in cold tier
         }
 
@@ -476,9 +476,7 @@ def _maybe_migrate_tiers(self) -> None:
 
             # Parse timestamp
             try:
-                last_accessed = datetime.fromisoformat(
-                    last_accessed_str.replace("Z", "+00:00")
-                )
+                last_accessed = datetime.fromisoformat(last_accessed_str.replace("Z", "+00:00"))
             except Exception:
                 continue
 
diff --git a/src/clio_agent/conversation_manager.py b/src/clio_agent/conversation_manager.py
index 21e5742..a823dd4 100644
--- a/src/clio_agent/conversation_manager.py
+++ b/src/clio_agent/conversation_manager.py
@@ -39,6 +39,7 @@ class ConversationManagerSignature(dspy.Signature):
         - key_topics: Main topics discussed
         - context_for_response: Relevant context
     """
+
     history: dspy.History = dspy.InputField(desc="Conversation history")
     current_question: str = dspy.InputField(desc="Current question")
     summary: str = dspy.OutputField(desc="Summary of context")
@@ -63,7 +64,7 @@ def add_message(self, role: str, content: str):
         self.history_buffer.append({"role": role, "content": content})
         if len(self.history_buffer) > self.max_history_length:
             # Keep only recent messages
-            self.history_buffer = self.history_buffer[-self.max_history_length:]
+            self.history_buffer = self.history_buffer[-self.max_history_length :]
 
     def get_history(self) -> dspy.History:
         """Get current history as dspy.History."""
diff --git a/src/clio_agent/experts/__init__.py b/src/clio_agent/experts/__init__.py
index 56b9d8f..cfc6d68 100644
--- a/src/clio_agent/experts/__init__.py
+++ b/src/clio_agent/experts/__init__.py
@@ -47,6 +47,7 @@
 # EXPERT REGISTRY
 # ============================================================================
 
+
 def get_all_experts() -> Dict[str, dspy.Module]:
     """Get all available expert instances.
 
diff --git a/src/clio_agent/experts/ndp_expert.py b/src/clio_agent/experts/ndp_expert.py
index 457daf7..090ef2e 100644
--- a/src/clio_agent/experts/ndp_expert.py
+++ b/src/clio_agent/experts/ndp_expert.py
@@ -62,9 +62,7 @@ def __init__(self, tool_executor: ToolExecutor | None = None) -> None:
         self._owns_executor = tool_executor is None
         self._tool_executor = tool_executor or create_sync_tool_executor(gateway)
         self._tools = [
-            tool
-            for tool in self._tool_executor.to_dspy_tools()
-            if tool.name.startswith("ndp_")
+            tool for tool in self._tool_executor.to_dspy_tools() if tool.name.startswith("ndp_")
         ]
 
     def forward(self, question: str, file_context: str = "") -> dspy.Prediction:
diff --git a/src/clio_agent/experts/sac_format_expert.py b/src/clio_agent/experts/sac_format_expert.py
index 068cb38..b5eeaab 100644
--- a/src/clio_agent/experts/sac_format_expert.py
+++ b/src/clio_agent/experts/sac_format_expert.py
@@ -58,9 +58,7 @@ def __init__(self, tool_executor: ToolExecutor | None = None) -> None:
         self._owns_executor = tool_executor is None
         self._tool_executor = tool_executor or create_sync_tool_executor(gateway)
         self._tools = [
-            tool
-            for tool in self._tool_executor.to_dspy_tools()
-            if tool.name.startswith("sac_")
+            tool for tool in self._tool_executor.to_dspy_tools() if tool.name.startswith("sac_")
         ]
 
     def forward(self, question: str, file_context: str = "") -> dspy.Prediction:
@@ -229,8 +227,7 @@ def plot_traces(self, filepath: str) -> ExpertResult:
         if isinstance(result, dict) and result.get("error"):
             return ExpertResult(
                 analysis=(
-                    "Could not create SAC waveform plot: "
-                    f"{format_tool_error(result['error'])}"
+                    f"Could not create SAC waveform plot: {format_tool_error(result['error'])}"
                 ),
                 recommendations="Verify the staged file and SAC plotting tool contract.",
                 source="deterministic",
diff --git a/src/clio_agent/gact/events.py b/src/clio_agent/gact/events.py
index f46080a..30fc295 100644
--- a/src/clio_agent/gact/events.py
+++ b/src/clio_agent/gact/events.py
@@ -91,9 +91,7 @@ class EventBus:
     instances which serialize their own operations.
     """
 
-    def __init__(
-        self, *, queue_capacity: int = 256, history_per_session: int = 256
-    ) -> None:
+    def __init__(self, *, queue_capacity: int = 256, history_per_session: int = 256) -> None:
         self._capacity = queue_capacity
         self._history_cap = history_per_session
         # session_id -> list of subscriber queues
@@ -126,9 +124,7 @@ def publish(self, event: Event) -> None:
             except asyncio.QueueFull:
                 pass
 
-    async def subscribe(
-        self, session_id: str, *, last_event_id: int = 0
-    ) -> AsyncIterator[Event]:
+    async def subscribe(self, session_id: str, *, last_event_id: int = 0) -> AsyncIterator[Event]:
         """Yield events for ``session_id`` until the consumer drops.
 
         ``last_event_id`` is the highest event id the client already
diff --git a/src/clio_agent/gact/scheduler.py b/src/clio_agent/gact/scheduler.py
index 4216753..24816de 100644
--- a/src/clio_agent/gact/scheduler.py
+++ b/src/clio_agent/gact/scheduler.py
@@ -120,9 +120,7 @@ def _flush(self) -> None:
         )
         tmp.replace(self._path)
 
-    def add(
-        self, *, session_id: str, cron: str, question: str
-    ) -> Schedule:
+    def add(self, *, session_id: str, cron: str, question: str) -> Schedule:
         sid = "sched_" + uuid.uuid4().hex[:12]
         sch = Schedule(
             id=sid,
@@ -140,9 +138,7 @@ def get(self, sid: str) -> Optional[Schedule]:
         with self._lock:
             return self._schedules.get(sid)
 
-    def list(
-        self, *, session_id: Optional[str] = None
-    ) -> list[Schedule]:
+    def list(self, *, session_id: Optional[str] = None) -> list[Schedule]:
         with self._lock:
             rows = list(self._schedules.values())
         if session_id is not None:
@@ -174,9 +170,7 @@ def due_now(self, when: datetime) -> Iterable[Schedule]:
         for sch in rows:
             if not sch.enabled:
                 continue
-            if sch.last_fired_at and sch.last_fired_at.startswith(
-                when_minute
-            ):
+            if sch.last_fired_at and sch.last_fired_at.startswith(when_minute):
                 continue
             if cron_matches(sch.cron, when):
                 yield sch
diff --git a/src/clio_agent/gact/sessions.py b/src/clio_agent/gact/sessions.py
index 80207f6..dcda289 100644
--- a/src/clio_agent/gact/sessions.py
+++ b/src/clio_agent/gact/sessions.py
@@ -324,7 +324,10 @@ def update(
             if edit_mode is not None and edit_mode in {"diff", "whole", "patch"}:
                 sess.edit_mode = edit_mode
             if routing_mode is not None and routing_mode in {
-                "auto", "chat", "experts", "reasoning_only",
+                "auto",
+                "chat",
+                "experts",
+                "reasoning_only",
             }:
                 sess.routing_mode = routing_mode
             if model is not None:
diff --git a/src/clio_agent/gact/user_agents.py b/src/clio_agent/gact/user_agents.py
index 346df04..1add0a4 100644
--- a/src/clio_agent/gact/user_agents.py
+++ b/src/clio_agent/gact/user_agents.py
@@ -67,20 +67,29 @@ def _load(self) -> None:
             return
         for row in data.get("agents", []):
             try:
-                self._agents[row["id"]] = UserAgent(**{
-                    k: row[k]
-                    for k in (
-                        "id", "title", "description", "source",
-                        "system_prompt", "default_provider", "default_model",
-                        "tier", "specialization",
-                    )
-                    if k in row
-                } | {
-                    "parameters": dict(row.get("parameters", {})),
-                    "keywords": list(row.get("keywords", [])),
-                    "tools": list(row.get("tools", [])),
-                    "metadata": dict(row.get("metadata", {})),
-                })
+                self._agents[row["id"]] = UserAgent(
+                    **{
+                        k: row[k]
+                        for k in (
+                            "id",
+                            "title",
+                            "description",
+                            "source",
+                            "system_prompt",
+                            "default_provider",
+                            "default_model",
+                            "tier",
+                            "specialization",
+                        )
+                        if k in row
+                    }
+                    | {
+                        "parameters": dict(row.get("parameters", {})),
+                        "keywords": list(row.get("keywords", [])),
+                        "tools": list(row.get("tools", [])),
+                        "metadata": dict(row.get("metadata", {})),
+                    }
+                )
             except Exception:
                 continue
 
diff --git a/src/clio_agent/optimizer/instrumentation.py b/src/clio_agent/optimizer/instrumentation.py
index 0b2a7de..6175a8e 100644
--- a/src/clio_agent/optimizer/instrumentation.py
+++ b/src/clio_agent/optimizer/instrumentation.py
@@ -35,6 +35,7 @@ def instrumented_forward(arc_memory: Any, agent_id: str) -> Callable:
         ... def forward(self, question, file_context=""):
         ...     return dspy.Prediction(analysis="...", recommendations="...")
     """
+
     def decorator(forward_fn: Callable) -> Callable:
         @functools.wraps(forward_fn)
         def wrapper(*args, **kwargs):
@@ -86,6 +87,7 @@ def wrapper(*args, **kwargs):
                 arc_memory.store_invocation(invocation)
 
         return wrapper
+
     return decorator
 
 
@@ -147,7 +149,13 @@ def _extract_output(result: Any) -> Dict[str, Any]:
             pass
     else:
         # Fallback: try common expert output fields
-        for field in ("analysis", "recommendations", "visualization_description", "file_path", "answer"):
+        for field in (
+            "analysis",
+            "recommendations",
+            "visualization_description",
+            "file_path",
+            "answer",
+        ):
             val = getattr(result, field, None)
             if val is not None:
                 output_data[field] = _to_safe_text(val)[:500]
diff --git a/src/clio_agent/optimizer/runner.py b/src/clio_agent/optimizer/runner.py
index 8597eb0..ed675d9 100644
--- a/src/clio_agent/optimizer/runner.py
+++ b/src/clio_agent/optimizer/runner.py
@@ -96,8 +96,7 @@ def run(
         """
         if len(trainset) < 5:
             raise ValueError(
-                f"Need at least 5 training examples for 20/80 split. "
-                f"Got {len(trainset)}."
+                f"Need at least 5 training examples for 20/80 split. Got {len(trainset)}."
             )
 
         if metric_fn is None:
diff --git a/src/clio_agent/optimizer/trainer.py b/src/clio_agent/optimizer/trainer.py
index 59830c4..eadd59e 100644
--- a/src/clio_agent/optimizer/trainer.py
+++ b/src/clio_agent/optimizer/trainer.py
@@ -16,10 +16,19 @@
 from clio_agent.arc.schema import Invocation, decode_invocation
 
 # Error keywords that indicate problematic output
-_ERROR_KEYWORDS = frozenset([
-    "error:", "error,", "traceback", "exception:", "failed to",
-    "could not", "unable to", "runtime error", "type error",
-])
+_ERROR_KEYWORDS = frozenset(
+    [
+        "error:",
+        "error,",
+        "traceback",
+        "exception:",
+        "failed to",
+        "could not",
+        "unable to",
+        "runtime error",
+        "type error",
+    ]
+)
 
 
 class TrainingSetGenerator:
@@ -46,9 +55,7 @@ def __init__(self, arc_memory: Any) -> None:
         """
         self._arc = arc_memory
 
-    def generate(
-        self, agent_id: str, min_examples: int = 30
-    ) -> list[dspy.Example]:
+    def generate(self, agent_id: str, min_examples: int = 30) -> list[dspy.Example]:
         """Generate training set from ARC invocations for a specific expert.
 
         Calls arc_memory.get_invocations_by_agent with status="success",
@@ -73,9 +80,7 @@ def generate(
             >>> assert len(examples) >= 30
             >>> assert "question" in examples[0].inputs()
         """
-        invocations = self._arc.get_invocations_by_agent(
-            agent_id, status="success"
-        )
+        invocations = self._arc.get_invocations_by_agent(agent_id, status="success")
 
         if len(invocations) < min_examples:
             raise ValueError(
@@ -127,9 +132,7 @@ def get_available_counts(self) -> dict[str, int]:
         return counts
 
     @staticmethod
-    def _invocation_to_example(
-        inv: Invocation, agent_id: str
-    ) -> dspy.Example | None:
+    def _invocation_to_example(inv: Invocation, agent_id: str) -> dspy.Example | None:
         """Convert a single Invocation to a dspy.Example.
 
         Maps invocation input/output fields to expert signature fields.
@@ -182,9 +185,7 @@ def _invocation_to_example(
             return None
 
 
-def clio_expert_metric(
-    example: dspy.Example, pred: Any, trace: Any = None
-) -> float | bool:
+def clio_expert_metric(example: dspy.Example, pred: Any, trace: Any = None) -> float | bool:
     """Multi-signal metric for CLIO expert optimization.
 
     Scores expert outputs on three weighted signals:
diff --git a/src/clio_agent/optimizer/variants.py b/src/clio_agent/optimizer/variants.py
index d151b82..c5342e9 100644
--- a/src/clio_agent/optimizer/variants.py
+++ b/src/clio_agent/optimizer/variants.py
@@ -137,9 +137,7 @@ def load_variant(
         """
         variant_path = self._variants_dir / f"{variant_id}.json"
         if not variant_path.exists():
-            raise FileNotFoundError(
-                f"Variant file not found: {variant_path}"
-            )
+            raise FileNotFoundError(f"Variant file not found: {variant_path}")
 
         existing_module.load(path=str(variant_path))
         return existing_module
@@ -169,9 +167,7 @@ def deploy(self, variant_id: str, agent_id: str) -> None:
                 found = True
 
         if not found:
-            raise ValueError(
-                f"Variant '{variant_id}' not found for agent '{agent_id}'"
-            )
+            raise ValueError(f"Variant '{variant_id}' not found for agent '{agent_id}'")
 
         # Deactivate all, then activate target
         for record in records:
diff --git a/src/clio_agent/registry/capability_matcher.py b/src/clio_agent/registry/capability_matcher.py
index e822488..bd919f3 100644
--- a/src/clio_agent/registry/capability_matcher.py
+++ b/src/clio_agent/registry/capability_matcher.py
@@ -37,13 +37,64 @@ def __init__(self):
         """Initialize matcher with default stopwords."""
         # Common English stopwords to filter from queries
         self._stopwords = {
-            'the', 'a', 'an', 'in', 'on', 'at', 'to', 'for', 'of', 'with',
-            'is', 'are', 'was', 'were', 'be', 'been', 'being', 'have', 'has',
-            'had', 'do', 'does', 'did', 'will', 'would', 'should', 'could',
-            'can', 'may', 'might', 'must', 'shall', 'i', 'you', 'he', 'she',
-            'it', 'we', 'they', 'them', 'this', 'that', 'these', 'those',
-            'my', 'your', 'his', 'her', 'its', 'our', 'their', 'what', 'which',
-            'who', 'when', 'where', 'why', 'how'
+            "the",
+            "a",
+            "an",
+            "in",
+            "on",
+            "at",
+            "to",
+            "for",
+            "of",
+            "with",
+            "is",
+            "are",
+            "was",
+            "were",
+            "be",
+            "been",
+            "being",
+            "have",
+            "has",
+            "had",
+            "do",
+            "does",
+            "did",
+            "will",
+            "would",
+            "should",
+            "could",
+            "can",
+            "may",
+            "might",
+            "must",
+            "shall",
+            "i",
+            "you",
+            "he",
+            "she",
+            "it",
+            "we",
+            "they",
+            "them",
+            "this",
+            "that",
+            "these",
+            "those",
+            "my",
+            "your",
+            "his",
+            "her",
+            "its",
+            "our",
+            "their",
+            "what",
+            "which",
+            "who",
+            "when",
+            "where",
+            "why",
+            "how",
         }
 
     def extract_keywords(self, query: str) -> List[str]:
@@ -70,18 +121,14 @@ def extract_keywords(self, query: str) -> List[str]:
             ['best', 'compression', 'adios']
         """
         # Lowercase and split on non-alphanumeric (keep numbers for hdf5, etc.)
-        words = re.findall(r'\w+', query.lower())
+        words = re.findall(r"\w+", query.lower())
 
         # Filter stopwords and empty strings
         keywords = [w for w in words if w and w not in self._stopwords]
 
         return keywords
 
-    def match_query(
-        self,
-        query: str,
-        capabilities: Dict[str, Any]
-    ) -> List[Tuple[str, float]]:
+    def match_query(self, query: str, capabilities: Dict[str, Any]) -> List[Tuple[str, float]]:
         """Match query to agent capabilities and return ranked list.
 
         Matching algorithm:
@@ -125,7 +172,7 @@ def match_query(
                 "No agent matching performed. "
                 "Consider rewording the query with more specific terms.",
                 UserWarning,
-                stacklevel=2
+                stacklevel=2,
             )
             return []
 
@@ -133,7 +180,7 @@ def match_query(
         scores = []
         for agent_id, capability in capabilities.items():
             # Get agent's keywords (handle missing keywords gracefully)
-            agent_keywords = capability.get('keywords', [])
+            agent_keywords = capability.get("keywords", [])
             if not agent_keywords:
                 continue
 
@@ -149,11 +196,7 @@ def match_query(
 
         return scores
 
-    def _calculate_score(
-        self,
-        query_keywords: List[str],
-        agent_keywords: List[str]
-    ) -> float:
+    def _calculate_score(self, query_keywords: List[str], agent_keywords: List[str]) -> float:
         """Calculate match score between query and agent keywords.
 
         Score formula:
@@ -183,7 +226,7 @@ def _calculate_score(
         # Example: "parallel io" becomes ["parallel", "io"]
         expanded_keywords = []
         for kw in agent_keywords:
-            expanded_keywords.extend(re.findall(r'\w+', kw.lower()))
+            expanded_keywords.extend(re.findall(r"\w+", kw.lower()))
 
         # Convert to set for O(1) lookup
         agent_kw_set = set(expanded_keywords)
diff --git a/src/clio_agent/runtime/hooks.py b/src/clio_agent/runtime/hooks.py
index e18e46e..835e2ff 100644
--- a/src/clio_agent/runtime/hooks.py
+++ b/src/clio_agent/runtime/hooks.py
@@ -62,9 +62,7 @@ class HookRegistry:
 
     def __init__(self, *, hooks_dir: Optional[Path] = None) -> None:
         self._dir = hooks_dir if hooks_dir is not None else _default_hooks_dir()
-        self._hooks: dict[str, list[Callable[..., Any]]] = {
-            event: [] for event in _KNOWN_EVENTS
-        }
+        self._hooks: dict[str, list[Callable[..., Any]]] = {event: [] for event in _KNOWN_EVENTS}
         self._lock = threading.Lock()
         self._load()
 
@@ -75,9 +73,7 @@ def _load(self) -> None:
             try:
                 module = self._import_path(path)
             except Exception as exc:
-                logger.warning(
-                    "[clio-hooks] failed to load %s: %r", path, exc
-                )
+                logger.warning("[clio-hooks] failed to load %s: %r", path, exc)
                 continue
             for event in _KNOWN_EVENTS:
                 fn = getattr(module, event, None)
@@ -98,9 +94,7 @@ def _import_path(path: Path) -> Any:
         spec.loader.exec_module(module)
         return module
 
-    def fire(
-        self, event: str, /, *args: Any, **kwargs: Any
-    ) -> list[Any]:
+    def fire(self, event: str, /, *args: Any, **kwargs: Any) -> list[Any]:
         """Invoke every registered hook for ``event`` in load order.
 
         Returns the list of return values for hooks that didn't
@@ -119,9 +113,7 @@ def fire(
                 results.append(fn(*args, **kwargs))
             except PermissionError:
                 if event.startswith("post_") or event == "on_error":
-                    logger.warning(
-                        "[clio-hooks] post-event hook raised; swallowing"
-                    )
+                    logger.warning("[clio-hooks] post-event hook raised; swallowing")
                     continue
                 raise
             except Exception as exc:  # noqa: BLE001
diff --git a/src/clio_agent/tools/servers/adios_server.py b/src/clio_agent/tools/servers/adios_server.py
index b537939..9c0846b 100644
--- a/src/clio_agent/tools/servers/adios_server.py
+++ b/src/clio_agent/tools/servers/adios_server.py
@@ -130,7 +130,9 @@ def _adios2_unavailable() -> dict[str, Any]:
     }
 
 
-def _inspect_variables_with_adios2(filepath: Path, variable_name: str | None = None) -> dict[str, Any]:
+def _inspect_variables_with_adios2(
+    filepath: Path, variable_name: str | None = None
+) -> dict[str, Any]:
     """Inspect BP variables through ADIOS2 when the optional dependency exists."""
     try:
         from adios2 import FileReader  # type: ignore[import-not-found]
diff --git a/src/clio_agent/tools/servers/fs_server.py b/src/clio_agent/tools/servers/fs_server.py
index 311884e..2735d7a 100644
--- a/src/clio_agent/tools/servers/fs_server.py
+++ b/src/clio_agent/tools/servers/fs_server.py
@@ -92,19 +92,17 @@ def propose_edit(filepath: str, new_content: str) -> dict[str, Any]:
         p = Path(safe_read)
         old = p.read_text(encoding="utf-8", errors="replace")
     new = new_content if isinstance(new_content, str) else str(new_content)
-    diff_lines = list(difflib.unified_diff(
-        old.splitlines(keepends=True),
-        new.splitlines(keepends=True),
-        fromfile=f"a/{p.name}",
-        tofile=f"b/{p.name}",
-        lineterm="",
-    ))
-    added = sum(
-        1 for ln in diff_lines if ln.startswith("+") and not ln.startswith("+++")
-    )
-    removed = sum(
-        1 for ln in diff_lines if ln.startswith("-") and not ln.startswith("---")
+    diff_lines = list(
+        difflib.unified_diff(
+            old.splitlines(keepends=True),
+            new.splitlines(keepends=True),
+            fromfile=f"a/{p.name}",
+            tofile=f"b/{p.name}",
+            lineterm="",
+        )
     )
+    added = sum(1 for ln in diff_lines if ln.startswith("+") and not ln.startswith("+++"))
+    removed = sum(1 for ln in diff_lines if ln.startswith("-") and not ln.startswith("---"))
     return {
         "path": str(p),
         "unified_diff": "\n".join(diff_lines),
diff --git a/tests/test_arc/test_lsm.py b/tests/test_arc/test_lsm.py
index 91638f2..94c8b08 100644
--- a/tests/test_arc/test_lsm.py
+++ b/tests/test_arc/test_lsm.py
@@ -130,9 +130,7 @@ def test_range_scan_across_flush(self, lsm):
     def test_compaction(self, temp_dir):
         """Test SSTable compaction."""
         # Create LSM with low compaction threshold
-        lsm = LSMTree(
-            data_dir=temp_dir, memtable_size=5, compaction_threshold=3
-        )
+        lsm = LSMTree(data_dir=temp_dir, memtable_size=5, compaction_threshold=3)
 
         try:
             # Write enough data to create multiple SSTables
diff --git a/tests/test_arc/test_memory_coverage.py b/tests/test_arc/test_memory_coverage.py
index 9d14f7d..9fb36ab 100644
--- a/tests/test_arc/test_memory_coverage.py
+++ b/tests/test_arc/test_memory_coverage.py
@@ -357,12 +357,17 @@ def test_clear_all(self, arc):
         """clear_all should remove everything."""
         now = time.time()
         arc.store_invocation(_inv("t1"))
-        arc.store_conversation(Conversation(
-            session_id="s1", user_id="u1",
-            created_at=now, updated_at=now,
-            last_accessed=now, status="active",
-            messages=[Message(role="user", content="hi", timestamp=now)],
-        ))
+        arc.store_conversation(
+            Conversation(
+                session_id="s1",
+                user_id="u1",
+                created_at=now,
+                updated_at=now,
+                last_accessed=now,
+                status="active",
+                messages=[Message(role="user", content="hi", timestamp=now)],
+            )
+        )
         arc.store_context(Context(domain="d1", created_at=now, updated_at=now))
 
         arc.clear_all()
diff --git a/tests/test_arc/test_retrieval.py b/tests/test_arc/test_retrieval.py
index bddc985..9560a8a 100644
--- a/tests/test_arc/test_retrieval.py
+++ b/tests/test_arc/test_retrieval.py
@@ -44,10 +44,7 @@ def _make_conversation(session_id: str, messages: list[tuple[str, str]]) -> Conv
         updated_at=now,
         last_accessed=now,
         status="active",
-        messages=[
-            Message(role=role, content=content, timestamp=now)
-            for role, content in messages
-        ],
+        messages=[Message(role=role, content=content, timestamp=now) for role, content in messages],
     )
 
 
@@ -94,25 +91,19 @@ class TestCalculateRelevanceScore:
     def test_identical_content(self, retriever):
         """Identical query and conversation should score high."""
         conv = _make_conversation("s1", [("user", "optimize HDF5 compression")])
-        score = retriever._calculate_relevance_score(
-            "optimize HDF5 compression", conv
-        )
+        score = retriever._calculate_relevance_score("optimize HDF5 compression", conv)
         assert score > 0.5
 
     def test_no_overlap(self, retriever):
         """No keyword overlap should score 0."""
         conv = _make_conversation("s1", [("user", "weather forecast tomorrow")])
-        score = retriever._calculate_relevance_score(
-            "optimize HDF5 compression", conv
-        )
+        score = retriever._calculate_relevance_score("optimize HDF5 compression", conv)
         assert score == 0.0
 
     def test_partial_overlap(self, retriever):
         """Partial overlap should score between 0 and 1."""
         conv = _make_conversation("s1", [("user", "HDF5 file structure analysis")])
-        score = retriever._calculate_relevance_score(
-            "optimize HDF5 compression", conv
-        )
+        score = retriever._calculate_relevance_score("optimize HDF5 compression", conv)
         assert 0.0 < score < 1.0
 
     def test_empty_query(self, retriever):
@@ -152,10 +143,13 @@ class TestExtractKeyTopics:
 
     def test_extracts_frequent_words(self, retriever):
         """Should extract most frequent meaningful words."""
-        conv = _make_conversation("s1", [
-            ("user", "HDF5 compression HDF5 gzip HDF5"),
-            ("assistant", "HDF5 uses gzip compression by default"),
-        ])
+        conv = _make_conversation(
+            "s1",
+            [
+                ("user", "HDF5 compression HDF5 gzip HDF5"),
+                ("assistant", "HDF5 uses gzip compression by default"),
+            ],
+        )
         topics = retriever.extract_key_topics([conv])
         assert "hdf5" in topics
         assert "compression" in topics
@@ -170,31 +164,28 @@ class TestRetrieveContextForQuery:
 
     def test_returns_context_object(self, retriever, arc):
         """Should return a Context object."""
-        context = retriever.retrieve_context_for_query(
-            "optimize HDF5", "session-1"
-        )
+        context = retriever.retrieve_context_for_query("optimize HDF5", "session-1")
         assert context is not None
         assert context.domain.startswith("query_context_")
 
     def test_with_conversation_history(self, retriever, arc):
         """Should include learned patterns from conversation."""
-        conv = _make_conversation("session-1", [
-            ("user", "How to optimize HDF5 compression?"),
-            ("assistant", "Use gzip compression with chunking."),
-        ])
+        conv = _make_conversation(
+            "session-1",
+            [
+                ("user", "How to optimize HDF5 compression?"),
+                ("assistant", "Use gzip compression with chunking."),
+            ],
+        )
         arc.store_conversation(conv)
 
-        context = retriever.retrieve_context_for_query(
-            "HDF5 compression", "session-1"
-        )
+        context = retriever.retrieve_context_for_query("HDF5 compression", "session-1")
         # Should have learned patterns
         assert len(context.learned_patterns) >= 0
 
     def test_without_conversation(self, retriever, arc):
         """Should handle missing conversation gracefully."""
-        context = retriever.retrieve_context_for_query(
-            "test query", "nonexistent-session"
-        )
+        context = retriever.retrieve_context_for_query("test query", "nonexistent-session")
         assert context is not None
 
 
@@ -203,9 +194,7 @@ class TestCompileExpertContext:
 
     def test_returns_string(self, retriever, arc):
         """Should return a compiled context string."""
-        result = retriever.compile_expert_context(
-            "analyze HDF5", "session-1", tier=2
-        )
+        result = retriever.compile_expert_context("analyze HDF5", "session-1", tier=2)
         assert isinstance(result, str)
 
     def test_lazy_init_compiler(self, retriever):
@@ -220,9 +209,7 @@ class TestGetRelevantToolResults:
 
     def test_no_context_returns_empty(self, retriever):
         """Missing context should return empty list."""
-        results = retriever.get_relevant_tool_results(
-            "analyze HDF5", "hdf5_domain"
-        )
+        results = retriever.get_relevant_tool_results("analyze HDF5", "hdf5_domain")
         assert results == []
 
     def test_with_cached_results(self, retriever, arc):
@@ -251,9 +238,7 @@ def test_with_cached_results(self, retriever, arc):
         )
         arc.store_context(ctx)
 
-        results = retriever.get_relevant_tool_results(
-            "HDF5 compression analysis", "hdf5_domain"
-        )
+        results = retriever.get_relevant_tool_results("HDF5 compression analysis", "hdf5_domain")
         # Should return results with hdf5-related content ranked higher
         assert len(results) >= 1
         assert "tool" in results[0]
diff --git a/tests/test_arc/test_shared_context.py b/tests/test_arc/test_shared_context.py
index a5913b7..4de08af 100644
--- a/tests/test_arc/test_shared_context.py
+++ b/tests/test_arc/test_shared_context.py
@@ -90,9 +90,7 @@ def test_get_session_profiles_multiple(self, arc):
         """Store 3 profiles in same session, retrieve all."""
         arc.store_dataset_profile(_make_profile(filepath="/data/a.parquet"))
         arc.store_dataset_profile(_make_profile(filepath="/data/b.parquet"))
-        arc.store_dataset_profile(
-            _make_profile(filepath="/data/c.hdf5", file_format="hdf5")
-        )
+        arc.store_dataset_profile(_make_profile(filepath="/data/c.hdf5", file_format="hdf5"))
 
         profiles = arc.get_session_profiles("session-1")
         assert len(profiles) == 3
@@ -120,12 +118,8 @@ def test_dataset_profile_cross_expert(self, arc):
 
     def test_dataset_profile_different_sessions(self, arc):
         """Session A profiles are not visible to session B."""
-        arc.store_dataset_profile(
-            _make_profile(session_id="session-A", filepath="/data/a.parquet")
-        )
-        arc.store_dataset_profile(
-            _make_profile(session_id="session-B", filepath="/data/b.parquet")
-        )
+        arc.store_dataset_profile(_make_profile(session_id="session-A", filepath="/data/a.parquet"))
+        arc.store_dataset_profile(_make_profile(session_id="session-B", filepath="/data/b.parquet"))
 
         profiles_a = arc.get_session_profiles("session-A")
         profiles_b = arc.get_session_profiles("session-B")
@@ -171,9 +165,7 @@ def test_store_and_get_procedural_memory(self, arc):
 
     def test_procedural_memory_filter_by_expert(self, arc):
         """Filter procedural memories by expert_id."""
-        arc.store_procedural_memory(
-            _make_procedural(expert_id="data", description="data pattern")
-        )
+        arc.store_procedural_memory(_make_procedural(expert_id="data", description="data pattern"))
         arc.store_procedural_memory(
             _make_procedural(expert_id="analysis", description="analysis pattern")
         )
@@ -192,15 +184,9 @@ def test_procedural_memory_ordering(self, arc):
         t2 = time.time() - 50
         t3 = time.time()
 
-        arc.store_procedural_memory(
-            _make_procedural(description="oldest", learned_at=t1)
-        )
-        arc.store_procedural_memory(
-            _make_procedural(description="middle", learned_at=t2)
-        )
-        arc.store_procedural_memory(
-            _make_procedural(description="newest", learned_at=t3)
-        )
+        arc.store_procedural_memory(_make_procedural(description="oldest", learned_at=t1))
+        arc.store_procedural_memory(_make_procedural(description="middle", learned_at=t2))
+        arc.store_procedural_memory(_make_procedural(description="newest", learned_at=t3))
 
         memories = arc.get_procedural_memories("session-1")
         assert len(memories) == 3
diff --git a/tests/test_arc/test_storage.py b/tests/test_arc/test_storage.py
index f1a16cc..555e06b 100644
--- a/tests/test_arc/test_storage.py
+++ b/tests/test_arc/test_storage.py
@@ -47,9 +47,7 @@ def test_default_tier_policy(self, backend):
     def test_custom_tier_policy(self, tmp_path):
         """Custom tier policy should override defaults."""
         policy = {"hot_to_warm": 2, "warm_to_cold": 14, "cold_to_archive": 60}
-        backend = IOWarpCTEBackend(
-            base_dir=str(tmp_path / "s"), tier_policy=policy
-        )
+        backend = IOWarpCTEBackend(base_dir=str(tmp_path / "s"), tier_policy=policy)
         assert backend.tier_policy["warm_to_cold"] == 14
 
     def test_performance_counters_start_at_zero(self, backend):
diff --git a/tests/test_core/test_codex_provider.py b/tests/test_core/test_codex_provider.py
index 455d112..a1d1573 100644
--- a/tests/test_core/test_codex_provider.py
+++ b/tests/test_core/test_codex_provider.py
@@ -55,9 +55,7 @@ def _reset_registration():
 
 class TestMessagesToCodexPrompt:
     def test_single_user_message(self):
-        prompt = _messages_to_codex_prompt(
-            [{"role": "user", "content": "hello"}]
-        )
+        prompt = _messages_to_codex_prompt([{"role": "user", "content": "hello"}])
         assert "JSON Lines" in prompt
         assert json.loads(prompt.splitlines()[-1]) == {
             "role": "user",
@@ -112,9 +110,7 @@ def test_role_like_content_cannot_create_prompt_boundary(self):
         }
 
     def test_unknown_role_is_downgraded_to_user(self):
-        prompt = _messages_to_codex_prompt(
-            [{"role": "root", "content": "hello"}]
-        )
+        prompt = _messages_to_codex_prompt([{"role": "root", "content": "hello"}])
         row = json.loads(prompt.splitlines()[-1])
         assert row == {"role": "user", "content": "hello"}
 
@@ -218,6 +214,7 @@ def test_unavailable_raises(self):
             from clio_agent.providers.codex_litellm import (
                 _resolve_codex_binary,
             )
+
             _resolve_codex_binary()
 
 
@@ -417,8 +414,6 @@ def test_registered_entry_points_at_codex(self):
         import litellm
 
         ensure_registered()
-        codex_entries = [
-            e for e in litellm.custom_provider_map if e.get("provider") == "codex"
-        ]
+        codex_entries = [e for e in litellm.custom_provider_map if e.get("provider") == "codex"]
         assert len(codex_entries) == 1
         assert isinstance(codex_entries[0]["custom_handler"], CodexLLM)
diff --git a/tests/test_core/test_instrumentation.py b/tests/test_core/test_instrumentation.py
index 698bddd..5a0e352 100644
--- a/tests/test_core/test_instrumentation.py
+++ b/tests/test_core/test_instrumentation.py
@@ -232,12 +232,14 @@ def test_get_invocations_by_agent_filtered():
         arc = _make_arc(tmp)
 
         # Store invocations for two agents
-        for i, (agent, status) in enumerate([
-            ("data", "success"),
-            ("data", "failure"),
-            ("analysis", "success"),
-            ("data", "success"),
-        ]):
+        for i, (agent, status) in enumerate(
+            [
+                ("data", "success"),
+                ("data", "failure"),
+                ("analysis", "success"),
+                ("data", "success"),
+            ]
+        ):
             inv = Invocation(
                 trace_id=f"trace-{i}",
                 session_id="session-1",
diff --git a/tests/test_core/test_runner.py b/tests/test_core/test_runner.py
index 7351734..cdc2632 100644
--- a/tests/test_core/test_runner.py
+++ b/tests/test_core/test_runner.py
@@ -129,8 +129,7 @@ class TestRun:
     def test_run_rejects_small_trainset(self, runner):
         """run() rejects trainset with fewer than 5 examples."""
         small_trainset = [
-            dspy.Example(question="q", analysis="a").with_inputs("question")
-            for _ in range(4)
+            dspy.Example(question="q", analysis="a").with_inputs("question") for _ in range(4)
         ]
 
         with pytest.raises(ValueError, match="at least 5"):
@@ -165,8 +164,15 @@ def test_run_full_pipeline(self, mock_evaluate_cls, mock_simba_cls, runner, tmp_
 
         # Verify result dict has all expected keys
         expected_keys = {
-            "optimized", "before_score", "after_score", "improvement_delta",
-            "p_value", "is_significant", "variant_record", "train_size", "val_size",
+            "optimized",
+            "before_score",
+            "after_score",
+            "improvement_delta",
+            "p_value",
+            "is_significant",
+            "variant_record",
+            "train_size",
+            "val_size",
         }
         assert set(result.keys()) == expected_keys
 
@@ -186,7 +192,9 @@ def test_run_full_pipeline(self, mock_evaluate_cls, mock_simba_cls, runner, tmp_
 
     @patch("clio_agent.optimizer.runner.dspy.SIMBA")
     @patch("clio_agent.optimizer.runner.dspy.evaluate.Evaluate")
-    def test_run_saves_variant_with_correct_args(self, mock_evaluate_cls, mock_simba_cls, runner, tmp_path):
+    def test_run_saves_variant_with_correct_args(
+        self, mock_evaluate_cls, mock_simba_cls, runner, tmp_path
+    ):
         """run() calls variant_manager.save_variant with correct arguments."""
         mock_evaluator = MagicMock()
         mock_evaluator.side_effect = [50.0, 70.0]
@@ -214,4 +222,7 @@ def test_run_saves_variant_with_correct_args(self, mock_evaluate_cls, mock_simba
         # Evaluate was constructed with the custom metric_fn
         mock_evaluate_cls.assert_called()
         call_kwargs = mock_evaluate_cls.call_args
-        assert call_kwargs.kwargs.get("metric") == metric_fn or call_kwargs[1].get("metric") == metric_fn
+        assert (
+            call_kwargs.kwargs.get("metric") == metric_fn
+            or call_kwargs[1].get("metric") == metric_fn
+        )
diff --git a/tests/test_core/test_variants.py b/tests/test_core/test_variants.py
index 0f0e2be..7b83ecb 100644
--- a/tests/test_core/test_variants.py
+++ b/tests/test_core/test_variants.py
@@ -65,9 +65,7 @@ class TestSaveVariant:
 
     def test_save_creates_file_on_disk(self, vm, mock_module, tmp_path):
         """save_variant creates a JSON file on disk."""
-        vm.save_variant(
-            mock_module, "data", 0.6, 0.85, 50, 0.003, True
-        )
+        vm.save_variant(mock_module, "data", 0.6, 0.85, 50, 0.003, True)
 
         variant_path = tmp_path / "variants" / "data_v1.json"
         assert variant_path.exists()
@@ -75,9 +73,7 @@ def test_save_creates_file_on_disk(self, vm, mock_module, tmp_path):
 
     def test_save_stores_variant_record_in_arc(self, vm, arc, mock_module):
         """save_variant stores VariantRecord in ARC memory."""
-        vm.save_variant(
-            mock_module, "data", 0.6, 0.85, 50, 0.003, True
-        )
+        vm.save_variant(mock_module, "data", 0.6, 0.85, 50, 0.003, True)
 
         stored = arc.get_variant_records("data")
         assert len(stored) == 1
@@ -92,9 +88,7 @@ def test_save_stores_variant_record_in_arc(self, vm, arc, mock_module):
 
     def test_save_returns_variant_record(self, vm, mock_module):
         """save_variant returns a VariantRecord with correct metadata."""
-        record = vm.save_variant(
-            mock_module, "data", 0.6, 0.85, 50, 0.003, True
-        )
+        record = vm.save_variant(mock_module, "data", 0.6, 0.85, 50, 0.003, True)
 
         assert isinstance(record, VariantRecord)
         assert record.variant_id == "data_v1"
diff --git a/tests/test_gact/conftest.py b/tests/test_gact/conftest.py
index 722ac36..7d71744 100644
--- a/tests/test_gact/conftest.py
+++ b/tests/test_gact/conftest.py
@@ -59,6 +59,4 @@ def complete_turn(
                 break
         time.sleep(poll_interval)
 
-    raise TimeoutError(
-        f"turn for user message {user_id!r} did not settle within {timeout:g}s"
-    )
+    raise TimeoutError(f"turn for user message {user_id!r} did not settle within {timeout:g}s")
diff --git a/tests/test_gact/test_cluster5.py b/tests/test_gact/test_cluster5.py
index 23a461e..42d3d73 100644
--- a/tests/test_gact/test_cluster5.py
+++ b/tests/test_gact/test_cluster5.py
@@ -88,11 +88,10 @@ def test_share_expiry(client: TestClient) -> None:
     """ttl_s=1 returns a token that 410s after we trick the clock."""
 
     sid = client.post("/v1/sessions", json={"title": "t"}).json()["id"]
-    share = client.post(
-        f"/v1/sessions/{sid}/share", json={"ttl_s": 1}
-    ).json()
+    share = client.post(f"/v1/sessions/{sid}/share", json={"ttl_s": 1}).json()
     # Force expiry by overwriting the row's expires_at.
     import time
+
     state = client.app.state
     state.shared_tokens[share["token"]]["expires_at"] = time.time() - 10
     resp = client.get(f"/v1/shared/{share['token']}")
@@ -118,19 +117,27 @@ def test_extract_agent_from_sessions(client: TestClient) -> None:
         (sid1, ["hdf5_list_datasets", "hdf5_analyze_file"]),
         (sid2, ["hdf5_list_datasets", "parquet_analyze_schema"]),
     ]:
-        state.messages.setdefault(sid, []).append(Message(
-            id="msg_user_x", session_id=sid, role="user",
-            created_at="2026-04-25T00:00:00Z",
-            updated_at="2026-04-25T00:00:00Z",
-            parts=[Part(id="part_x", type="text", text=f"analyze {sid}")],
-        ))
-        state.messages[sid].append(Message(
-            id=f"msg_asst_{sid}", session_id=sid, role="assistant",
-            created_at="2026-04-25T00:00:00Z",
-            updated_at="2026-04-25T00:00:00Z",
-            parts=[Part(id="part_y", type="text", text="done")],
-            metadata={"tools_called": [{"name": t} for t in tools]},
-        ))
+        state.messages.setdefault(sid, []).append(
+            Message(
+                id="msg_user_x",
+                session_id=sid,
+                role="user",
+                created_at="2026-04-25T00:00:00Z",
+                updated_at="2026-04-25T00:00:00Z",
+                parts=[Part(id="part_x", type="text", text=f"analyze {sid}")],
+            )
+        )
+        state.messages[sid].append(
+            Message(
+                id=f"msg_asst_{sid}",
+                session_id=sid,
+                role="assistant",
+                created_at="2026-04-25T00:00:00Z",
+                updated_at="2026-04-25T00:00:00Z",
+                parts=[Part(id="part_y", type="text", text="done")],
+                metadata={"tools_called": [{"name": t} for t in tools]},
+            )
+        )
 
     resp = client.post(
         "/v1/agents/extract",
diff --git a/tests/test_gact/test_commands.py b/tests/test_gact/test_commands.py
index d9a77a9..6145d94 100644
--- a/tests/test_gact/test_commands.py
+++ b/tests/test_gact/test_commands.py
@@ -30,9 +30,7 @@ def get_cache_stats(self) -> dict[str, object]:
 
 @pytest.fixture()
 def client(tmp_path: Path) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_Agent())
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_Agent()))
 
 
 def test_commands_listed(client: TestClient) -> None:
@@ -55,9 +53,7 @@ def test_dispatch_clear_drops_messages(client: TestClient) -> None:
 
     sid = client.post("/v1/sessions", json={"title": "t"}).json()["id"]
     complete_turn(client, sid, "first")
-    assert (
-        len(client.get(f"/v1/sessions/{sid}/messages").json()["messages"]) == 2
-    )
+    assert len(client.get(f"/v1/sessions/{sid}/messages").json()["messages"]) == 2
 
     resp = client.post(f"/v1/sessions/{sid}/commands/clear").json()
     assert resp["command"] == "/clear"
diff --git a/tests/test_gact/test_context_injection.py b/tests/test_gact/test_context_injection.py
index 9c11f06..c21ae84 100644
--- a/tests/test_gact/test_context_injection.py
+++ b/tests/test_gact/test_context_injection.py
@@ -23,10 +23,15 @@ def __init__(self) -> None:
 
     def forward(self, question: str, session_id: str = "default"):
         self.calls.append((question, session_id))
-        return type("Pred", (), {
-            "answer": "ok", "selected_expert": "",
-            "routing_rationale": "",
-        })()
+        return type(
+            "Pred",
+            (),
+            {
+                "answer": "ok",
+                "selected_expert": "",
+                "routing_rationale": "",
+            },
+        )()
 
 
 @pytest.fixture()
@@ -37,9 +42,7 @@ def setup(tmp_path: Path):
     app = build_app(sessions_path=tmp_path / "s.json", agent=agent)
     # Update ws_default's root_path so its files pass the
     # workspace check inside _enrich_with_context_files.
-    app.state.workspaces.update(
-        "ws_default", root_path=str(tmp_path)
-    )
+    app.state.workspaces.update("ws_default", root_path=str(tmp_path))
     return app, TestClient(app), agent, tmp_path
 
 
@@ -134,10 +137,7 @@ def test_read_file_deleted_after_attach_surfaces_error(setup) -> None:
         "retry",
         "exit",
     ]
-    completed = [
-        ev for ev in app.state.bus._history.get(sid, [])
-        if ev.type == "message.completed"
-    ]
+    completed = [ev for ev in app.state.bus._history.get(sid, []) if ev.type == "message.completed"]
     assert completed, "turn did not publish message.completed"
     payload = completed[-1].payload
     assert payload["message_id"] == assistant["id"]
@@ -145,9 +145,7 @@ def test_read_file_deleted_after_attach_surfaces_error(setup) -> None:
     assert payload["error_info"]["error"] == "context_file_error"
 
 
-def test_path_outside_workspace_root_is_inlined_for_reads(
-    setup, tmp_path: Path
-) -> None:
+def test_path_outside_workspace_root_is_inlined_for_reads(setup, tmp_path: Path) -> None:
     """iowarp/clio-agent#5 (revised): the workspace-root check used to
     silently skip context_files outside the root, but the user
     explicitly attaches via the API — they know what they're doing.
diff --git a/tests/test_gact/test_cost_tracking.py b/tests/test_gact/test_cost_tracking.py
index bb3b552..7c84dbe 100644
--- a/tests/test_gact/test_cost_tracking.py
+++ b/tests/test_gact/test_cost_tracking.py
@@ -39,13 +39,12 @@ def forward(self, question: str, session_id: str):
 
 
 def _client(tmp_path: Path, pred) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_Agent(pred))
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_Agent(pred)))
 
 
 def _turn(client: TestClient, sid: str) -> dict:
     from .conftest import complete_turn
+
     return complete_turn(client, sid, "hello")
 
 
diff --git a/tests/test_gact/test_diffs.py b/tests/test_gact/test_diffs.py
index c53b663..213053b 100644
--- a/tests/test_gact/test_diffs.py
+++ b/tests/test_gact/test_diffs.py
@@ -27,13 +27,12 @@ def forward(self, question: str, session_id: str):
 
 
 def _client(tmp_path: Path, diffs) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_Agent(diffs))
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_Agent(diffs)))
 
 
 def _turn(client: TestClient, sid: str) -> dict:
     from .conftest import complete_turn
+
     return complete_turn(client, sid, "propose an edit")
 
 
@@ -47,9 +46,12 @@ def _turn(client: TestClient, sid: str) -> dict:
 
 
 def test_assistant_emits_file_diff_part(tmp_path: Path) -> None:
-    client = _client(tmp_path, diffs=[
-        {"path": "main.go", "unified_diff": SAMPLE_DIFF},
-    ])
+    client = _client(
+        tmp_path,
+        diffs=[
+            {"path": "main.go", "unified_diff": SAMPLE_DIFF},
+        ],
+    )
     sid = client.post("/v1/sessions", json={"title": "t"}).json()["id"]
     a = _turn(client, sid)
     parts = a["parts"]
@@ -61,16 +63,17 @@ def test_assistant_emits_file_diff_part(tmp_path: Path) -> None:
 
 
 def test_apply_flips_status_and_returns_paths(tmp_path: Path) -> None:
-    client = _client(tmp_path, diffs=[
-        {"path": "a.py", "unified_diff": SAMPLE_DIFF},
-        {"path": "b.py", "unified_diff": SAMPLE_DIFF},
-    ])
+    client = _client(
+        tmp_path,
+        diffs=[
+            {"path": "a.py", "unified_diff": SAMPLE_DIFF},
+            {"path": "b.py", "unified_diff": SAMPLE_DIFF},
+        ],
+    )
     sid = client.post("/v1/sessions", json={"title": "t"}).json()["id"]
     _turn(client, sid)
 
-    resp = client.post(
-        f"/v1/sessions/{sid}/diffs/apply", json={"paths": ["a.py"]}
-    ).json()
+    resp = client.post(f"/v1/sessions/{sid}/diffs/apply", json={"paths": ["a.py"]}).json()
     assert resp["applied"] == ["a.py"]
 
     # b.py still pending — apply-all picks it up.
@@ -83,9 +86,12 @@ def test_apply_flips_status_and_returns_paths(tmp_path: Path) -> None:
 
 
 def test_reject_flips_status(tmp_path: Path) -> None:
-    client = _client(tmp_path, diffs=[
-        {"path": "a.py", "unified_diff": SAMPLE_DIFF},
-    ])
+    client = _client(
+        tmp_path,
+        diffs=[
+            {"path": "a.py", "unified_diff": SAMPLE_DIFF},
+        ],
+    )
     sid = client.post("/v1/sessions", json={"title": "t"}).json()["id"]
     _turn(client, sid)
     resp = client.post(f"/v1/sessions/{sid}/diffs/reject", json={}).json()
diff --git a/tests/test_gact/test_doctor_integrations.py b/tests/test_gact/test_doctor_integrations.py
index 275f8ad..c4e7dd6 100644
--- a/tests/test_gact/test_doctor_integrations.py
+++ b/tests/test_gact/test_doctor_integrations.py
@@ -65,11 +65,13 @@ def test_no_agent_no_arc_overall_is_unavailable(tmp_path: Path) -> None:
 
 
 def test_fake_agent_flagged_degraded(tmp_path: Path) -> None:
-    resp = _health_response(build_app(
-        sessions_path=tmp_path / "s.json",
-        agent=_RealishAgent(),
-        arc=_FakeARC(),
-    ))
+    resp = _health_response(
+        build_app(
+            sessions_path=tmp_path / "s.json",
+            agent=_RealishAgent(),
+            arc=_FakeARC(),
+        )
+    )
     assert resp.status_code == 200
     body = resp.json()
     rows = {r["name"]: r for r in body["integrations"]}
@@ -83,11 +85,13 @@ def test_fake_agent_flagged_degraded(tmp_path: Path) -> None:
 
 
 def test_broken_arc_reports_unavailable(tmp_path: Path) -> None:
-    resp = _health_response(build_app(
-        sessions_path=tmp_path / "s.json",
-        agent=_RealishAgent(),
-        arc=_BrokenARC(),
-    ))
+    resp = _health_response(
+        build_app(
+            sessions_path=tmp_path / "s.json",
+            agent=_RealishAgent(),
+            arc=_BrokenARC(),
+        )
+    )
     assert resp.status_code == 503
     body = resp.json()
     rows = {r["name"]: r for r in body["integrations"]}
@@ -96,11 +100,13 @@ def test_broken_arc_reports_unavailable(tmp_path: Path) -> None:
 
 
 def test_ready_health_returns_200(tmp_path: Path) -> None:
-    resp = _health_response(build_app(
-        sessions_path=tmp_path / "s.json",
-        agent=_ProductionLikeAgent(),
-        arc=_FakeARC(),
-    ))
+    resp = _health_response(
+        build_app(
+            sessions_path=tmp_path / "s.json",
+            agent=_ProductionLikeAgent(),
+            arc=_FakeARC(),
+        )
+    )
     assert resp.status_code == 200
     body = resp.json()
     assert body["overall_status"] == "ready"
diff --git a/tests/test_gact/test_fork_and_search.py b/tests/test_gact/test_fork_and_search.py
index bb8282b..5f7c939 100644
--- a/tests/test_gact/test_fork_and_search.py
+++ b/tests/test_gact/test_fork_and_search.py
@@ -30,13 +30,12 @@ def forward(self, question: str, session_id: str):
 
 
 def _client(tmp_path: Path) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_Agent())
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_Agent()))
 
 
 def _turn(client: TestClient, sid: str, text: str) -> dict:
     from .conftest import complete_turn
+
     return complete_turn(client, sid, text)
 
 
@@ -80,9 +79,7 @@ def test_search_returns_ranked_snippets(tmp_path: Path) -> None:
     _turn(client, sid, "load /tmp/alpha.parquet")
     _turn(client, sid, "compare /tmp/alpha.parquet to /tmp/beta.parquet")
 
-    body = client.get(
-        f"/v1/sessions/{sid}/messages/search?q=alpha.parquet"
-    ).json()
+    body = client.get(f"/v1/sessions/{sid}/messages/search?q=alpha.parquet").json()
     matches = body["matches"]
     assert len(matches) >= 2
     for m in matches:
diff --git a/tests/test_gact/test_hooks.py b/tests/test_gact/test_hooks.py
index f089329..f9f1862 100644
--- a/tests/test_gact/test_hooks.py
+++ b/tests/test_gact/test_hooks.py
@@ -42,11 +42,14 @@ def _hook_dir(tmp_path: Path, **events: str) -> Path:
 def test_pre_tool_hook_can_block(tmp_path: Path) -> None:
     """A pre_tool hook that raises PermissionError vetoes the call."""
 
-    d = _hook_dir(tmp_path, pre_tool="""
+    d = _hook_dir(
+        tmp_path,
+        pre_tool="""
 def pre_tool(name, args):
     if name.startswith("hdf5_"):
         raise PermissionError("no hdf5 today")
-""")
+""",
+    )
     reg = HookRegistry(hooks_dir=d)
     with pytest.raises(PermissionError, match="no hdf5"):
         reg.fire("pre_tool", "hdf5_list_datasets", {"path": "/tmp/x"})
@@ -58,10 +61,13 @@ def test_post_tool_hook_swallows_exceptions(tmp_path: Path) -> None:
     """post_* hooks must NEVER crash a turn; exceptions are
     swallowed + logged."""
 
-    d = _hook_dir(tmp_path, post_tool="""
+    d = _hook_dir(
+        tmp_path,
+        post_tool="""
 def post_tool(name, args, result=None, error=None):
     raise RuntimeError("boom")
-""")
+""",
+    )
     reg = HookRegistry(hooks_dir=d)
     # Should not raise.
     reg.fire("post_tool", "fs_read_file", {"x": 1}, result="ok")
@@ -72,16 +78,17 @@ def test_pre_message_hook_blocks_via_app(tmp_path: Path) -> None:
     end-to-end through GACT — the assistant message comes back
     with error_info.error == permission_error."""
 
-    d = _hook_dir(tmp_path, pre_message="""
+    d = _hook_dir(
+        tmp_path,
+        pre_message="""
 def pre_message(session_id, text):
     if "secret" in text.lower():
         raise PermissionError("blocked by policy")
-""")
+""",
+    )
     install_global_registry(HookRegistry(hooks_dir=d))
     try:
-        app = build_app(
-            sessions_path=tmp_path / "s.json", agent=_Agent()
-        )
+        app = build_app(sessions_path=tmp_path / "s.json", agent=_Agent())
         with TestClient(app) as c:
             sid = c.post("/v1/sessions", json={"title": "t"}).json()["id"]
             ack = c.post(
@@ -91,6 +98,7 @@ def pre_message(session_id, text):
             assert ack.status_code == 200
             # Wait for the background turn to settle into error.
             import time as _t
+
             for _ in range(30):
                 sess = c.get(f"/v1/sessions/{sid}").json()
                 if sess["status"] == "error":
@@ -106,15 +114,16 @@ def test_post_message_hook_runs_after_settle(tmp_path: Path) -> None:
     (write a marker file here)."""
 
     marker = tmp_path / "post_message_fired.txt"
-    d = _hook_dir(tmp_path, post_message=f"""
+    d = _hook_dir(
+        tmp_path,
+        post_message=f"""
 def post_message(session_id, assistant):
     open({str(marker)!r}, "w").write(assistant['id'])
-""")
+""",
+    )
     install_global_registry(HookRegistry(hooks_dir=d))
     try:
-        app = build_app(
-            sessions_path=tmp_path / "s.json", agent=_Agent()
-        )
+        app = build_app(sessions_path=tmp_path / "s.json", agent=_Agent())
         with TestClient(app) as c:
             from .conftest import complete_turn
 
diff --git a/tests/test_gact/test_memory_stats.py b/tests/test_gact/test_memory_stats.py
index 5f1204a..3a3a948 100644
--- a/tests/test_gact/test_memory_stats.py
+++ b/tests/test_gact/test_memory_stats.py
@@ -51,9 +51,7 @@ def get_cache_stats(self) -> dict[str, Any]:
 @pytest.fixture()
 def client_with_arc(tmp_path: Path) -> TestClient:
     arc = FakeARC(hits=80, misses=20, conv_index_size=12, inv_index_size=42)
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", arc=arc)
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", arc=arc))
 
 
 def test_memory_stats_reports_cache_counters(client_with_arc: TestClient) -> None:
diff --git a/tests/test_gact/test_metrics.py b/tests/test_gact/test_metrics.py
index c6c4830..a022a0f 100644
--- a/tests/test_gact/test_metrics.py
+++ b/tests/test_gact/test_metrics.py
@@ -32,9 +32,7 @@ def forward(self, question: str, session_id: str):
 
 @pytest.fixture()
 def client(tmp_path: Path) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "sessions.json", agent=_FakeAgent())
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "sessions.json", agent=_FakeAgent()))
 
 
 def test_metrics_empty_state(client: TestClient) -> None:
diff --git a/tests/test_gact/test_nanoagent_runtime.py b/tests/test_gact/test_nanoagent_runtime.py
index fe2a841..c37060a 100644
--- a/tests/test_gact/test_nanoagent_runtime.py
+++ b/tests/test_gact/test_nanoagent_runtime.py
@@ -70,10 +70,9 @@ def test_spawn_many_empty_returns_empty() -> None:
 
 def test_render_input_uses_question_field_when_present() -> None:
     from clio_agent.runtime.nanoagent import _render_input
+
     assert _render_input({"question": "hello"}) == "hello"
-    assert (
-        _render_input({"file": "/tmp/x", "mode": "read"}).startswith("file=")
-    )
+    assert _render_input({"file": "/tmp/x", "mode": "read"}).startswith("file=")
 
 
 def test_analysis_expert_detects_parallel_items() -> None:
@@ -84,12 +83,10 @@ def test_analysis_expert_detects_parallel_items() -> None:
 
     # Multi-item triggers fan-out.
     assert _detect_parallel_items("validate /tmp/a.parquet and /tmp/b.parquet") == [
-        "/tmp/a.parquet", "/tmp/b.parquet",
+        "/tmp/a.parquet",
+        "/tmp/b.parquet",
     ]
-    assert (
-        len(_detect_parallel_items("check schema, statistics, and quality"))
-        == 3
-    )
+    assert len(_detect_parallel_items("check schema, statistics, and quality")) == 3
     # Single-item questions don't fan out.
     assert _detect_parallel_items("validate /tmp/a.parquet") == []
     # No trigger word -> no fan-out.
diff --git a/tests/test_gact/test_permissions.py b/tests/test_gact/test_permissions.py
index d33b59e..3d8ee66 100644
--- a/tests/test_gact/test_permissions.py
+++ b/tests/test_gact/test_permissions.py
@@ -27,9 +27,7 @@ def forward(self, question: str, session_id: str):
 
 
 def _client(tmp_path: Path, perms) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_Agent(perms))
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_Agent(perms)))
 
 
 def _turn(client: TestClient, sid: str) -> dict:
@@ -40,16 +38,19 @@ def _turn(client: TestClient, sid: str) -> dict:
 
 
 def test_permission_requested_then_allowed(tmp_path: Path) -> None:
-    client = _client(tmp_path, perms=[
-        {
-            "tool_call": {
-                "call_id": "c1",
-                "tool_name": "shell.exec",
-                "input": {"cmd": "rm -rf /tmp/scratch"},
-            },
-            "summary": "destructive shell command",
-        }
-    ])
+    client = _client(
+        tmp_path,
+        perms=[
+            {
+                "tool_call": {
+                    "call_id": "c1",
+                    "tool_name": "shell.exec",
+                    "input": {"cmd": "rm -rf /tmp/scratch"},
+                },
+                "summary": "destructive shell command",
+            }
+        ],
+    )
     sid = client.post("/v1/sessions", json={"title": "t"}).json()["id"]
     _turn(client, sid)
 
diff --git a/tests/test_gact/test_session_export.py b/tests/test_gact/test_session_export.py
index d696e71..0363205 100644
--- a/tests/test_gact/test_session_export.py
+++ b/tests/test_gact/test_session_export.py
@@ -23,9 +23,7 @@ def forward(self, *args, **kwargs):
 
 
 def _client(tmp_path: Path) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_Agent())
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_Agent()))
 
 
 def test_export_unknown_session_404s(tmp_path: Path) -> None:
@@ -59,9 +57,7 @@ def test_export_then_import_round_trip(tmp_path: Path) -> None:
     assert len(rows) == 4
     # Original user prompts preserved.
     user_texts = {
-        p["text"] for m in rows
-        for p in m["parts"]
-        if m["role"] == "user" and p["type"] == "text"
+        p["text"] for m in rows for p in m["parts"] if m["role"] == "user" and p["type"] == "text"
     }
     assert {"first", "second"} == user_texts
 
diff --git a/tests/test_gact/test_session_tasks.py b/tests/test_gact/test_session_tasks.py
index 43b66d2..6d33807 100644
--- a/tests/test_gact/test_session_tasks.py
+++ b/tests/test_gact/test_session_tasks.py
@@ -26,9 +26,7 @@ def test_empty_task_list(client: TestClient, sid: str) -> None:
 
 
 def test_create_then_list(client: TestClient, sid: str) -> None:
-    new = client.post(
-        f"/v1/sessions/{sid}/tasks", json={"title": "validate schema"}
-    ).json()
+    new = client.post(f"/v1/sessions/{sid}/tasks", json={"title": "validate schema"}).json()
     assert new["title"] == "validate schema"
     assert new["status"] == "pending"
     assert new["id"].startswith("task_")
@@ -38,21 +36,15 @@ def test_create_then_list(client: TestClient, sid: str) -> None:
 
 
 def test_patch_status(client: TestClient, sid: str) -> None:
-    new = client.post(
-        f"/v1/sessions/{sid}/tasks", json={"title": "x"}
-    ).json()
-    patched = client.patch(
-        f"/v1/tasks/{new['id']}", json={"status": "completed"}
-    ).json()
+    new = client.post(f"/v1/sessions/{sid}/tasks", json={"title": "x"}).json()
+    patched = client.patch(f"/v1/tasks/{new['id']}", json={"status": "completed"}).json()
     assert patched["status"] == "completed"
     rows = client.get(f"/v1/sessions/{sid}/tasks").json()["tasks"]
     assert rows[0]["status"] == "completed"
 
 
 def test_delete(client: TestClient, sid: str) -> None:
-    new = client.post(
-        f"/v1/sessions/{sid}/tasks", json={"title": "x"}
-    ).json()
+    new = client.post(f"/v1/sessions/{sid}/tasks", json={"title": "x"}).json()
     assert client.delete(f"/v1/tasks/{new['id']}").status_code == 204
     rows = client.get(f"/v1/sessions/{sid}/tasks").json()["tasks"]
     assert rows == []
diff --git a/tests/test_gact/test_sse.py b/tests/test_gact/test_sse.py
index c68340c..5074d33 100644
--- a/tests/test_gact/test_sse.py
+++ b/tests/test_gact/test_sse.py
@@ -52,9 +52,7 @@ def test_format_sse_emits_canonical_wire_shape() -> None:
     assert "data: " in raw
     assert raw.endswith("\n\n")
     # The data line is valid JSON matching the envelope.
-    data_line = next(
-        ln for ln in raw.splitlines() if ln.startswith("data: ")
-    )
+    data_line = next(ln for ln in raw.splitlines() if ln.startswith("data: "))
     payload = json.loads(data_line.removeprefix("data: "))
     assert payload["type"] == "message.completed"
     assert payload["payload"] == {"a": 1}
@@ -146,9 +144,7 @@ def forward(self, question: str, session_id: str) -> Any:
 
 @pytest.fixture()
 def client(tmp_path: Path) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_FakeAgent())
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_FakeAgent()))
 
 
 def test_sse_endpoint_404s_for_unknown_session(client: TestClient) -> None:
diff --git a/tests/test_gact/test_thinking_blocks.py b/tests/test_gact/test_thinking_blocks.py
index 0589b4c..487b5a0 100644
--- a/tests/test_gact/test_thinking_blocks.py
+++ b/tests/test_gact/test_thinking_blocks.py
@@ -28,9 +28,7 @@ def forward(self, *args, **kwargs):
 
 
 def _client(tmp_path: Path, pred) -> TestClient:
-    return TestClient(
-        build_app(sessions_path=tmp_path / "s.json", agent=_Agent(pred))
-    )
+    return TestClient(build_app(sessions_path=tmp_path / "s.json", agent=_Agent(pred)))
 
 
 def test_chain_of_thought_reasoning_becomes_thinking_part(
@@ -51,12 +49,14 @@ def test_chain_of_thought_reasoning_becomes_thinking_part(
 def test_react_trajectory_dict_becomes_thinking_part(tmp_path: Path) -> None:
     from .conftest import complete_turn
 
-    pred = _Pred(trajectory={
-        "step_0_thought": "first probe the schema",
-        "step_0_tool_name": "hdf5_list_datasets",
-        "step_1_thought": "now read /sim/temperature",
-        "step_1_tool_name": "hdf5_analyze_dataset",
-    })
+    pred = _Pred(
+        trajectory={
+            "step_0_thought": "first probe the schema",
+            "step_0_tool_name": "hdf5_list_datasets",
+            "step_1_thought": "now read /sim/temperature",
+            "step_1_tool_name": "hdf5_analyze_dataset",
+        }
+    )
     c = _client(tmp_path, pred)
     sid = c.post("/v1/sessions", json={"title": "t"}).json()["id"]
     a = complete_turn(c, sid, "analyze")
diff --git a/tests/test_integration_contract/conftest.py b/tests/test_integration_contract/conftest.py
index b07b094..c53546d 100644
--- a/tests/test_integration_contract/conftest.py
+++ b/tests/test_integration_contract/conftest.py
@@ -105,9 +105,7 @@ def post_user(http: httpx.Client, sid: str, text: str) -> str:
     return ack.json()["message_id"]
 
 
-def turn(
-    http: httpx.Client, sid: str, text: str, *, timeout: float = 180.0
-) -> dict[str, Any]:
+def turn(http: httpx.Client, sid: str, text: str, *, timeout: float = 180.0) -> dict[str, Any]:
     """POST + wait for assistant + return the assistant dict."""
 
     user_id = post_user(http, sid, text)
diff --git a/tests/test_tools/test_file_policy.py b/tests/test_tools/test_file_policy.py
index 338f838..616a103 100644
--- a/tests/test_tools/test_file_policy.py
+++ b/tests/test_tools/test_file_policy.py
@@ -1,6 +1,5 @@
 """Tests for file access policy validation."""
 
-
 from clio_agent.tools import file_policy
 from clio_agent.tools.file_policy import FileAccessPolicy, FilePolicyError
 
diff --git a/tests/test_tools/test_fs_server.py b/tests/test_tools/test_fs_server.py
index 539e90f..4ccfa61 100644
--- a/tests/test_tools/test_fs_server.py
+++ b/tests/test_tools/test_fs_server.py
@@ -84,18 +84,14 @@ def test_gateway_apply_edit_write_respects_late_global_permission_gate(
     seen: list[str] = []
 
     try:
-        set_global_permission_gate(
-            lambda name, _args: seen.append(name) or "allow"
-        )
+        set_global_permission_gate(lambda name, _args: seen.append(name) or "allow")
         executor.call_tool(
             "fs_apply_edit_write",
             {"filepath": str(target), "new_content": "allowed\n"},
         )
         assert target.read_text(encoding="utf-8") == "allowed\n"
 
-        set_global_permission_gate(
-            lambda name, _args: seen.append(name) or "deny"
-        )
+        set_global_permission_gate(lambda name, _args: seen.append(name) or "deny")
         with pytest.raises(PermissionError, match="denied"):
             executor.call_tool(
                 "fs_apply_edit_write",
diff --git a/tests/test_tools/test_gateway.py b/tests/test_tools/test_gateway.py
index 47d57fa..623ce64 100644
--- a/tests/test_tools/test_gateway.py
+++ b/tests/test_tools/test_gateway.py
@@ -160,7 +160,8 @@ async def test_list_capabilities_inside_running_loop_has_no_unawaited_warning(re
 
     gc.collect()
     leaked = [
-        warning for warning in recwarn
+        warning
+        for warning in recwarn
         if issubclass(warning.category, RuntimeWarning)
         and "was never awaited" in str(warning.message)
     ]
@@ -171,8 +172,6 @@ async def test_list_capabilities_inside_running_loop_has_no_unawaited_warning(re
 async def test_gateway_error_handling():
     """Test that errors propagate correctly through gateway."""
     async with Client(gateway) as client:
-        result = await client.call_tool(
-            "hdf5_analyze_file", {"filepath": "/nonexistent/file.h5"}
-        )
+        result = await client.call_tool("hdf5_analyze_file", {"filepath": "/nonexistent/file.h5"})
         data = _parse_result(result)
         assert "error" in data
diff --git a/tests/test_tools/test_shell_server.py b/tests/test_tools/test_shell_server.py
index 4335c8d..0fa0b89 100644
--- a/tests/test_tools/test_shell_server.py
+++ b/tests/test_tools/test_shell_server.py
@@ -23,7 +23,9 @@ def _parse_result(result: object) -> dict:
 
 
 @pytest.mark.asyncio
-async def test_shell_bash_runs_simple_command(tmp_path: Path, monkeypatch: pytest.MonkeyPatch) -> None:
+async def test_shell_bash_runs_simple_command(
+    tmp_path: Path, monkeypatch: pytest.MonkeyPatch
+) -> None:
     """The shell tool should execute a bounded command and return output."""
 
     monkeypatch.setenv("CLIO_ALLOWED_ROOTS", str(tmp_path))