dreadnode 2.0.9__tar.gz → 2.0.10__tar.gz

{dreadnode-2.0.9 → dreadnode-2.0.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: dreadnode
-Version: 2.0.9
+Version: 2.0.10
 Summary: Dreadnode SDK
 Project-URL: Homepage, https://dreadnode.io
 Project-URL: Documentation, https://docs.dreadnode.io
@@ -19,7 +19,7 @@ Requires-Dist: gepa>=0.1.1
 Requires-Dist: httpx<1.0.0,>=0.28.0
 Requires-Dist: jsonpath-ng>=1.7.0
 Requires-Dist: jsonref>=1.1.0
-Requires-Dist: litellm<=1.82.3,>=1.80.11
+Requires-Dist: litellm<=1.83.0,>=1.83.0
 Requires-Dist: logfire<=3.20.0,>=3.5.3
 Requires-Dist: loguru>=0.7.3
 Requires-Dist: mcp<2.0.0,>=1.25.0

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/agents/agent.py

@@ -1,7 +1,7 @@
 import asyncio
 import inspect
 import typing as t
-from contextlib import AsyncExitStack, asynccontextmanager, nullcontext, suppress
+from contextlib import AsyncExitStack, asynccontextmanager, suppress
 from copy import deepcopy
 from pathlib import Path
 from textwrap import dedent
@@ -586,7 +586,7 @@ class Agent(Executor[AgentEvent, Trajectory]):
                 agent_name=self.name,
                 inputs={"goal": self._current_input},
                 params={
-                    "session_id": str(self.trajectory.session_id),
+                    "session_id": str(traj.session_id),
                     "max_steps": self.max_steps,
                     **({"model": self.model_name} if self.model_name else {}),
                     **({"tools": [tool.name for tool in self.all_tools]} if self.all_tools else {}),
@@ -683,7 +683,7 @@ class Agent(Executor[AgentEvent, Trajectory]):
                         )
 
                     # Check stop conditions INSIDE span
-                    if any(cond(self.trajectory.steps) for cond in self.stop_conditions):
+                    if any(cond(traj.steps) for cond in self.stop_conditions):
                         logger.info("A stop condition was met. Ending run.")
                         gen_end = GenerationEnd(
                             agent_id=self.agent_id,
@@ -872,7 +872,7 @@ class Agent(Executor[AgentEvent, Trajectory]):
                         )
 
                     # Check stop conditions again INSIDE span
-                    if any(cond(self.trajectory.steps) for cond in self.stop_conditions):
+                    if any(cond(traj.steps) for cond in self.stop_conditions):
                         break
 
             except Continue as e:
@@ -899,7 +899,7 @@ class Agent(Executor[AgentEvent, Trajectory]):
             stop_reason = "error"
         elif max_steps_reached:
             stop_reason = "max_steps_reached"
-        elif self.trajectory.steps and isinstance(self.trajectory.steps[-1], AgentStalled):
+        elif traj.steps and isinstance(traj.steps[-1], AgentStalled):
             stop_reason = "stalled"
 
         async for event in self._dispatch(
@@ -961,17 +961,12 @@ class Agent(Executor[AgentEvent, Trajectory]):
                 if hasattr(tool_container, "__aenter__") and hasattr(tool_container, "__aexit__"):
                     await stack.enter_async_context(tool_container)
 
-            # Parent span only on new conversation (not when borrowing external trajectory)
             display_name = self.name or str(self.agent_id)[:8]
-            ctx = (
-                dreadnode.task_span(
-                    f"agent:{display_name}",
-                    type="agent",
-                    label=display_name,
-                    tags=["agent"],
-                )
-                if reset and trajectory is None
-                else nullcontext()
+            ctx = dreadnode.task_span(
+                f"agent:{display_name}",
+                type="agent",
+                label=display_name,
+                tags=["agent"],
             )
 
             with ctx as parent_span:

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/agents/tools.py

@@ -170,7 +170,8 @@ async def offload_tool_output(
     working_dir: Path,
 ) -> tuple[str, Path]:
     """Write tool output to disk and return middle-out summary plus file path."""
-    output_dir = working_dir / ".dreadnode" / "tool-output"
+    output_root = working_dir if working_dir.name == ".dreadnode" else working_dir / ".dreadnode"
+    output_dir = output_root / "tool-output"
     await asyncio.to_thread(output_dir.mkdir, parents=True, exist_ok=True)
 
     safe_id = re.sub(r"[^a-zA-Z0-9_-]", "_", tool_call_id)
@@ -178,7 +179,8 @@ async def offload_tool_output(
     async with aiofiles.open(file_path, "w") as f:
         await f.write(content)
 
-    relative_path = file_path.relative_to(working_dir)
+    relative_base = working_dir.parent if working_dir.name == ".dreadnode" else working_dir
+    relative_path = file_path.relative_to(relative_base)
 
     half = OFFLOAD_THRESHOLD // 2
     start = content[:half]

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/beast.py

@@ -18,6 +18,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import GenerateParams, Generator, get_generator
@@ -232,6 +233,7 @@ def beast_attack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -247,6 +249,7 @@ def beast_attack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     attack = Study[str](

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/crescendo.py

@@ -11,6 +11,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import Generator, get_generator
@@ -221,6 +222,7 @@ Return ONLY the question text."""
         llm_judge(
             evaluator_model,
             eval_rubric,
+            input=TrialCandidate(),
             min_score=0,
             max_score=1,
         )
@@ -235,6 +237,7 @@ Return ONLY the question text."""
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=objective_judge,
+            include_input=True,
         )
 
     # Resolve model names for span attributes

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/deep_inception.py

@@ -18,6 +18,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import Generator
@@ -135,6 +136,7 @@ def deep_inception_attack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -150,6 +152,7 @@ def deep_inception_attack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     attack = Study[str](

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/drattack.py

@@ -19,6 +19,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import GenerateParams, Generator, get_generator
@@ -137,6 +138,7 @@ def drattack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -152,6 +154,7 @@ def drattack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     attack = Study[str](

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/goat.py

@@ -8,6 +8,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import Generator
@@ -114,6 +115,7 @@ def goat_attack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -141,6 +143,7 @@ def goat_attack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     # Resolve model names for span attributes

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/gptfuzzer.py

@@ -17,6 +17,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import Generator
@@ -150,6 +151,7 @@ def gptfuzzer_attack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -165,6 +167,7 @@ def gptfuzzer_attack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     attack = Study[str](

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/pair.py

@@ -15,6 +15,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import Generator
@@ -142,6 +143,7 @@ def pair_attack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -157,6 +159,7 @@ def pair_attack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     # Resolve model names for span attributes

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/rainbow.py

@@ -18,6 +18,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import Generator
@@ -170,6 +171,7 @@ def rainbow_attack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -185,6 +187,7 @@ def rainbow_attack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     attack = Study[str](

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/airt/renellm.py

@@ -18,6 +18,7 @@ from dreadnode.airt.compliance import (
     tag_attack,
 )
 from dreadnode.airt.prompt import OBJECTIVE_SCORE_KEY, _evaluate_prompt
+from dreadnode.core.meta import TrialCandidate
 from dreadnode.core.scorer import Scorer
 from dreadnode.core.transforms import Transform, TransformsLike
 from dreadnode.generators.generator import GenerateParams, Generator, get_generator
@@ -165,6 +166,7 @@ def renellm_attack(
         llm_judge(
             evaluator_model,
             rubric,
+            input=TrialCandidate(),
             min_score=1,
             max_score=10,
         )
@@ -180,6 +182,7 @@ def renellm_attack(
             target=target,
             fitted_transforms=fitted_transforms,
             scorer=prompt_judge,
+            include_input=True,
         )
 
     attack = Study[str](

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/app/api/client.py

@@ -239,6 +239,45 @@ class ApiClient:
         response = self.request("GET", "/inference")
         return t.cast("list[dict[str, t.Any]]", response.json().get("models", []))
 
+    def list_catalog_models(
+        self,
+        *,
+        query: str | None = None,
+        limit: int | None = None,
+        provider: str | None = None,
+        capabilities: list[str] | None = None,
+        open_weights: bool | None = None,
+        min_context: int | None = None,
+        max_context: int | None = None,
+        min_price: float | None = None,
+        max_price: float | None = None,
+        featured: bool = False,
+    ) -> list[dict[str, t.Any]]:
+        """GET /api/v1/inference/catalog - List featured or searched BYOK catalog models."""
+        params: dict[str, t.Any] = {}
+        if query is not None:
+            params["query"] = query
+        if limit is not None:
+            params["limit"] = limit
+        if provider is not None:
+            params["provider"] = provider
+        if capabilities:
+            params["capabilities"] = capabilities
+        if open_weights is not None:
+            params["open_weights"] = open_weights
+        if min_context is not None:
+            params["min_context"] = min_context
+        if max_context is not None:
+            params["max_context"] = max_context
+        if min_price is not None:
+            params["min_price"] = min_price
+        if max_price is not None:
+            params["max_price"] = max_price
+        if featured:
+            params["featured"] = True
+        response = self.request("GET", "/inference/catalog", params=params)
+        return t.cast("list[dict[str, t.Any]]", response.json().get("models", []))
+
     def get_user_preferences(self) -> dict[str, t.Any]:
         """GET /api/v1/user/preferences - Get user's enabled models with status."""
         response = self.request("GET", "/user/preferences")
@@ -1729,7 +1768,6 @@ class ApiClient:
         *,
         page: int = 1,
         limit: int = 50,
-        category: list[str] | None = None,
         difficulty: list[str] | None = None,
         tags: list[str] | None = None,
         search: str | None = None,
@@ -1741,8 +1779,6 @@ class ApiClient:
         params: dict[str, t.Any] = {}
         params["page"] = page
         params["limit"] = limit
-        if category is not None:
-            params["category"] = category
         if difficulty is not None:
             params["difficulty"] = difficulty
         if tags is not None:
@@ -1762,7 +1798,6 @@ class ApiClient:
         self,
         org: str,
         *,
-        category: list[str] | None = None,
         difficulty: list[str] | None = None,
         search: str | None = None,
         tags: list[str] | None = None,
@@ -1772,8 +1807,6 @@ class ApiClient:
     ) -> dict[str, t.Any]:
         """GET /org/{org}/tasks/facets - Get task filter facets."""
         params: dict[str, t.Any] = {"include_public": include_public}
-        if category is not None:
-            params["category"] = category
         if difficulty is not None:
             params["difficulty"] = difficulty
         if search is not None:
@@ -1900,11 +1933,18 @@ class ApiClient:
         page: int = 1,
         page_size: int = 20,
         project_id: str | None = None,
+        user_id: str | None = None,
     ) -> dict[str, t.Any]:
-        """GET /org/{org}/ws/{workspace}/sessions - List sessions."""
+        """GET /org/{org}/ws/{workspace}/sessions - List sessions.
+
+        The platform endpoint defaults to workspace-wide. Pass ``user_id``
+        to scope to a single creator (e.g. the TUI's "my sessions" view).
+        """
         params: dict[str, t.Any] = {"page": page, "page_size": page_size}
         if project_id is not None:
             params["project_id"] = project_id
+        if user_id is not None:
+            params["user_id"] = user_id
         response = self.request("GET", f"/org/{org}/ws/{workspace}/sessions", params=params)
         return t.cast("dict[str, t.Any]", response.json())
 
@@ -1928,22 +1968,6 @@ class ApiClient:
         )
         return t.cast("dict[str, t.Any]", response.json())
 
-    def ingest_session_events(
-        self,
-        org: str,
-        workspace: str,
-        session_id: str,
-        events: list[dict[str, t.Any]],
-        wal_offset: int,
-    ) -> dict[str, t.Any]:
-        """POST /org/{org}/ws/{workspace}/sessions/{session_id}/events - Ingest events."""
-        response = self.request(
-            "POST",
-            f"/org/{org}/ws/{workspace}/sessions/{session_id}/events",
-            json_data={"events": events, "wal_offset": wal_offset},
-        )
-        return t.cast("dict[str, t.Any]", response.json())
-
     # ------------------------------------------------------------------
     # Session transcripts (message-level persistence)
     # ------------------------------------------------------------------
@@ -1994,6 +2018,7 @@ class ApiClient:
         messages: list[t.Any],
         *,
         context: dict[str, t.Any] | None = None,
+        usage_by_uuid: dict[str, dict[str, int]] | None = None,
     ) -> dict[str, t.Any]:
         """POST /org/{org}/ws/{workspace}/sessions/{session_id}/transcript — Append messages.
 
@@ -2004,10 +2029,20 @@ class ApiClient:
             messages: List of SDK Message objects to persist.
             context: Optional dict with ``model``, ``agent``, ``system_prompt``
                      — include on first call or when agent/model changes.
+            usage_by_uuid: Optional mapping of message uuid (str) → usage dict
+                     with ``input_tokens`` / ``output_tokens`` /
+                     ``cache_read_input_tokens`` / ``cache_creation_input_tokens``.
+                     Entries are attached to the matching message payload.
         """
-        payload: dict[str, t.Any] = {
-            "messages": [self._message_to_transcript_dict(m) for m in messages],
-        }
+        payload_messages: list[dict[str, t.Any]] = []
+        for m in messages:
+            d = self._message_to_transcript_dict(m)
+            msg_id = d.get("id")
+            if usage_by_uuid and isinstance(msg_id, str) and msg_id in usage_by_uuid:
+                d["usage"] = usage_by_uuid[msg_id]
+            payload_messages.append(d)
+
+        payload: dict[str, t.Any] = {"messages": payload_messages}
         if context is not None:
             payload["context"] = context
 

{dreadnode-2.0.9 → dreadnode-2.0.10}/dreadnode/app/api/models.py

@@ -16,6 +16,20 @@ if t.TYPE_CHECKING:
 
 AnyDict = dict[str, t.Any]
 
+
+def _normalize_default_agent_fields(data: t.Any) -> t.Any:
+    """Normalize legacy `default` agent/capability inputs to no override."""
+    if not isinstance(data, dict):
+        return data
+
+    normalized = dict(data)
+    for field in ("agent", "capability"):
+        value = normalized.get(field)
+        if isinstance(value, str) and value.strip().lower() == "default":
+            normalized[field] = None
+    return normalized
+
+
 # -----------------------------------------------------------------------------
 # Type Aliases for Exports
 # -----------------------------------------------------------------------------
@@ -749,6 +763,11 @@ class ChatRequest(BaseModel):
 
     model_config = ConfigDict(populate_by_name=True)
 
+    @model_validator(mode="before")
+    @classmethod
+    def _normalize_default_agent(cls, data: t.Any) -> t.Any:
+        return _normalize_default_agent_fields(data)
+
 
 class ChatResponse(BaseModel):
     """Response body for chat completion."""
@@ -798,6 +817,11 @@ class SessionCreateRequest(BaseModel):
 
     model_config = ConfigDict(populate_by_name=True)
 
+    @model_validator(mode="before")
+    @classmethod
+    def _normalize_default_agent(cls, data: t.Any) -> t.Any:
+        return _normalize_default_agent_fields(data)
+
 
 class SessionRestoreRequest(BaseModel):
     """Request body for restoring a server-side session runtime."""
@@ -809,6 +833,11 @@ class SessionRestoreRequest(BaseModel):
     messages: list[SessionMessage] = Field(default_factory=list)
     trajectory: dict[str, t.Any] | None = None
 
+    @model_validator(mode="before")
+    @classmethod
+    def _normalize_default_agent(cls, data: t.Any) -> t.Any:
+        return _normalize_default_agent_fields(data)
+
 
 class SessionRestoreResponse(BaseModel):
     """Response returned after restoring a server-side session runtime."""
@@ -853,7 +882,7 @@ class CapabilityAgentInfo(BaseModel):
 class ComponentStatusInfo(BaseModel):
     """Component-level load/runtime status."""
 
-    kind: t.Literal["agent", "tool", "skill", "mcp_server", "capability", "check"]
+    kind: t.Literal["agent", "tool", "hook", "skill", "mcp_server", "capability", "check"]
     name: str
     status: t.Literal["ok", "error", "degraded"]
     error: str | None = None
@@ -871,6 +900,7 @@ class CapabilityInfo(BaseModel):
     name: str
     display_name: str
     canonical_name: str | None = None
+    local_path: str | None = None
     source: t.Literal["runtime", "local", "package"] | None = None
     provenance: t.Literal["local", "org", "public"] | None = None
     version: str | None = None
@@ -997,17 +1027,6 @@ class PlatformSessionResponse(BaseModel):
     """Update timestamp."""
 
 
-class PlatformSessionEvent(BaseModel):
-    """Session event for platform ingestion (maps to API SessionEventCreate)."""
-
-    event_type: str
-    """Event type (e.g. 'message', 'tool_use')."""
-    data: dict[str, t.Any] = Field(default_factory=dict)
-    """Event payload."""
-    timestamp: datetime | None = None
-    """Event timestamp."""
-
-
 # -----------------------------------------------------------------------------
 # Training
 # -----------------------------------------------------------------------------