chore: release main

.github/workflows/ci.yml

@@ -18,7 +18,7 @@ jobs:
   lint:
     timeout-minutes: 10
     name: lint
-    runs-on: ${{ github.repository == 'stainless-sdks/agentex-sdk-python' && 'depot-ubuntu-24.04' || 'ubuntu-latest' }}
+    runs-on: ${{ startsWith(github.repository, 'stainless-sdks/') && 'depot-ubuntu-24.04' || 'ubuntu-latest' }}
     if: (github.event_name == 'push' || github.event.pull_request.head.repo.fork) && (github.event_name != 'push' || github.event.head_commit.message != 'codegen metadata')
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
@@ -44,7 +44,7 @@ jobs:
     permissions:
       contents: read
       id-token: write
-    runs-on: ${{ github.repository == 'stainless-sdks/agentex-sdk-python' && 'depot-ubuntu-24.04' || 'ubuntu-latest' }}
+    runs-on: ${{ startsWith(github.repository, 'stainless-sdks/') && 'depot-ubuntu-24.04' || 'ubuntu-latest' }}
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2
 
@@ -87,7 +87,7 @@ jobs:
   test:
     timeout-minutes: 10
     name: test
-    runs-on: ${{ github.repository == 'stainless-sdks/agentex-sdk-python' && 'depot-ubuntu-24.04' || 'ubuntu-latest' }}
+    runs-on: ${{ startsWith(github.repository, 'stainless-sdks/') && 'depot-ubuntu-24.04' || 'ubuntu-latest' }}
     if: github.event_name == 'push' || github.event.pull_request.head.repo.fork
     steps:
       - uses: actions/checkout@de0fac2e4500dabe0009e67214ff5f5447ce83dd # v6.0.2

.release-please-manifest.json

@@ -1,4 +1,4 @@
 {
-  ".": "0.20.0",
-  "adk": "0.20.0"
+  ".": "0.21.0",
+  "adk": "0.21.0"
 }

.stats.yml

@@ -1,4 +1,4 @@
 configured_endpoints: 75
-openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/sgp/agentex-sdk-b2df5f506330ad5fba5a0d518ab8a4bcf876e8c3684a4fe0d0cc5102fd9c569e.yml
-openapi_spec_hash: 132e9efdb0535d9594abadf799431cf5
+openapi_spec_url: https://storage.googleapis.com/stainless-sdk-openapi-specs/sgp/agentex-sdk-644a4ec06aa1f055c614cbef3379684819a4edd84eeb20d2fb29ae01663622a3.yml
+openapi_spec_hash: a6a4dc0c09691ac9783bf38e9653a464
 config_hash: 593e89b291976a5e84e4c3c3f8324354

CHANGELOG.md

@@ -12,6 +12,21 @@
 
 * **tracing:** emit OTel metrics for async span queue depth, batch drain, and SGP export success/failure (HTTP status labels). Disable SDK-side recording with ``AGENTEX_TRACING_METRICS=0``.
 
+## 0.21.0 (2026-07-28)
+
+Full Changelog: [agentex-client-v0.20.0...agentex-client-v0.21.0](https://github.com/scaleapi/scale-agentex-python/compare/agentex-client-v0.20.0...agentex-client-v0.21.0)
+
+### Features
+
+* **api:** add include_live parameter to schedules list method ([199fd6a](https://github.com/scaleapi/scale-agentex-python/commit/199fd6a633bc82626a4629ee1ac52a40e6f10271))
+* **codex:** republish todo_list revisions as they are ticked off ([#473](https://github.com/scaleapi/scale-agentex-python/issues/473)) ([0252fcc](https://github.com/scaleapi/scale-agentex-python/commit/0252fcc1d00c96066e4500d09b07a6b0da836437))
+* **stlc:** configurable CI runner and private-production-repo support in workflow templates ([23c1b6b](https://github.com/scaleapi/scale-agentex-python/commit/23c1b6b8dc60f16c5e19a1aab68d244d286caffc))
+
+
+### Bug Fixes
+
+* **api:** remove params field from task list response ([152bc75](https://github.com/scaleapi/scale-agentex-python/commit/152bc75b242c2ff4179e0cfac0f33edbc6bcae38))
+
 ## 0.20.0 (2026-07-16)
 
 Full Changelog: [agentex-client-v0.19.0...agentex-client-v0.20.0](https://github.com/scaleapi/scale-agentex-python/compare/agentex-client-v0.19.0...agentex-client-v0.20.0)

adk/CHANGELOG.md

@@ -1,5 +1,13 @@
 # Changelog
 
+## 0.21.0 (2026-07-28)
+
+Full Changelog: [agentex-sdk-v0.20.0...agentex-sdk-v0.21.0](https://github.com/scaleapi/scale-agentex-python/compare/agentex-sdk-v0.20.0...agentex-sdk-v0.21.0)
+
+### Chores
+
+* **agentex-sdk:** Synchronize agentex versions
+
 ## 0.20.0 (2026-07-16)
 
 Full Changelog: [agentex-sdk-v0.19.0...agentex-sdk-v0.20.0](https://github.com/scaleapi/scale-agentex-python/compare/agentex-sdk-v0.19.0...agentex-sdk-v0.20.0)

adk/pyproject.toml

@@ -4,7 +4,7 @@
 # (agentex/{__init__.py, _*.py, types/, resources/}) ships from the slim
 # sibling package `agentex-client` which is pinned as a runtime dep.
 name = "agentex-sdk"
-version = "0.20.0"
+version = "0.21.0"
 description = "Agent Development Kit (ADK) overlay for the Agentex API — FastACP server, Temporal workflows, LLM provider integrations, observability"
 license = "Apache-2.0"
 authors = [

examples/tutorials/10_async/10_temporal/020_state_machine/project/workflows/deep_research/performing_deep_research.py

@@ -13,21 +13,28 @@
 
 logger = make_logger(__name__)
 
+# These reference MCP servers still import the mcp 1.x API (``McpError``), which
+# mcp 2.0.0 renamed to ``MCPError``. uvx gives each server its own isolated env and
+# resolves ``mcp`` unpinned there, ignoring the version this project pins, so without
+# this constraint every server dies at import and the agent silently makes zero tool
+# calls. Drop the pin once the servers support mcp 2.x.
+_MCP_PIN = ["--with", "mcp<2"]
+
 MCP_SERVERS = [
     StdioServerParameters(
         command="uvx",
-        args=["mcp-server-time", "--local-timezone", "America/Los_Angeles"],
+        args=[*_MCP_PIN, "mcp-server-time", "--local-timezone", "America/Los_Angeles"],
     ),
     StdioServerParameters(
         command="uvx",
-        args=["openai-websearch-mcp"],
+        args=[*_MCP_PIN, "openai-websearch-mcp"],
         env={
             "OPENAI_API_KEY": os.environ.get("OPENAI_API_KEY", "")
         }
     ),
     StdioServerParameters(
         command="uvx",
-        args=["mcp-server-fetch"],
+        args=[*_MCP_PIN, "mcp-server-fetch"],
     ),
 ]
 

pyproject.toml

@@ -3,7 +3,7 @@
 # overlay (formerly `src/agentex/lib/*`) now lives in `adk/` and ships
 # as the sibling `agentex-sdk` package — see `adk/pyproject.toml`.
 name = "agentex-client"
-version = "0.20.0"
+version = "0.21.0"
 description = "The official Python REST client for the Agentex API"
 dynamic = ["readme"]
 license = "Apache-2.0"

src/agentex/_version.py

@@ -1,4 +1,4 @@
 # File generated from our OpenAPI spec by Stainless. See CONTRIBUTING.md for details.
 
 __title__ = "agentex"
-__version__ = "0.20.0"  # x-release-please-version
+__version__ = "0.21.0"  # x-release-please-version

src/agentex/lib/adk/_modules/_codex_sync.py

@@ -60,7 +60,13 @@
                              a synthetic ToolRequestContent Full is emitted before the response.
   mcp_tool_call           -> same as command_execution
   web_search              -> same as command_execution
-  todo_list               -> same as command_execution
+  todo_list               -> same as command_execution, plus:
+                               item.updated -> StreamTaskMessageFull(ToolResponseContent)
+                             Codex ticks one in-place todo_list item through
+                             item.updated; each revision is republished under
+                             the same tool_call_id so consumers can render the
+                             checklist filling in rather than jumping to its
+                             final state at end of turn.
   collab_tool_call        -> same as command_execution
   error (item type)       -> StreamTaskMessageFull(TextContent, error text) on completed only
 
@@ -75,9 +81,10 @@
                           without this module needing to know about spans.
   item.updated (reasoning): the intermediate cumulative text is discarded;
                             only item.completed carries the final text.
-  item.updated (tool):    tool item types other than agent_message do not
-                          emit updates; item.started opens the request and
-                          item.completed closes it.
+  item.updated (tool):    only todo_list is republished (see above). For the
+                          other tool item types item.started opens the request
+                          and item.completed closes it; any updates in between
+                          carry no state a consumer could act on.
 """
 
 from __future__ import annotations
@@ -112,6 +119,11 @@ def _truncate(text: str, max_len: int = _MAX_RESULT_LENGTH) -> str:
     return str(text)[:max_len]
 
 
+# Tool items codex revises in place rather than reopening. Their item.updated
+# events carry real intermediate state, so they are forwarded as responses.
+_PROGRESSIVE_TOOL_ITEMS = frozenset({"todo_list"})
+
+
 def _tool_name_for(item_type: str, payload: dict[str, Any]) -> str:
     """Derive a canonical tool name from a codex item type."""
     if item_type == "command_execution":
@@ -467,6 +479,33 @@ def _handle_item(self, evt_type: str, item: dict[str, Any]) -> list[StreamTaskMe
                 )
                 out.append(StreamTaskMessageDone(type="done", index=req_idx))
 
+            elif evt_type == "item.updated" and item_type in _PROGRESSIVE_TOOL_ITEMS:
+                # Codex revises its plan in place: one todo_list item is opened
+                # at the start of the turn and ticked off through item.updated,
+                # with item.completed only arriving at the very end. Forwarding
+                # each revision as a response for the SAME tool_call_id lets a
+                # consumer show the checklist filling in as the turn runs; the
+                # last response received is the current state.
+                if item_id in self._tool_open:
+                    actual_type = self._tool_item_types.get(item_id, item_type)
+                    result_text, is_error = _tool_output_for(actual_type, item)
+                    resp_content: dict[str, Any] = {"result": result_text}
+                    if is_error:
+                        resp_content["is_error"] = True
+                    out.append(
+                        StreamTaskMessageFull(
+                            type="full",
+                            index=self._alloc(),
+                            content=ToolResponseContent(
+                                type="tool_response",
+                                author="agent",
+                                tool_call_id=tool_call_id,
+                                name=_tool_name_for(actual_type, item),
+                                content=resp_content,
+                            ),
+                        )
+                    )
+
             elif evt_type == "item.completed":
                 # file_change items may only emit item.completed (no started).
                 if item_id not in self._tool_open:

src/agentex/lib/core/tracing/obs_ids.py

@@ -0,0 +1,83 @@
+"""Correlate adk business spans with the active observability trace.
+
+The adk business ``trace_id`` is the agent **task id** (run-level: it spans the
+whole agent run across many requests -- task/create, then each message/send turn),
+so we must NOT overwrite it with a per-request observability trace_id. Doing so
+would collapse the run-level grouping.
+
+Instead, each business span is *tagged* with the active observability
+trace_id/span_id (this is the OpenTelemetry "span link" pattern -- correlate
+across trace granularities rather than merging them). You can then pivot from a
+persisted business span to the Tempo/Datadog trace for the turn that produced it,
+while the business trace still groups the entire run by task id.
+
+Source selection follows SGP_OBS_MODE, matching egp-api-backend:
+  - unset / "dd_only": ddtrace context (current stack)
+  - "dual":            OTel/LGTM preferred, ddtrace fallback
+  - "lgtm":            OTel/LGTM only
+
+This never fabricates ids -- if no observability context is active, it returns
+an empty dict and the span is simply not tagged.
+"""
+from __future__ import annotations
+
+import os
+from typing import Dict, Tuple, Optional
+
+__all__ = ("get_obs_mode", "obs_correlation")
+
+DD_ONLY = "dd_only"
+DUAL = "dual"
+LGTM = "lgtm"
+_DEFAULT_MODE = DD_ONLY
+_VALID_MODES = (DD_ONLY, DUAL, LGTM)
+
+
+def get_obs_mode() -> str:
+    """Unset/empty/unrecognized -> ``dd_only`` (current behavior)."""
+    raw = os.getenv("SGP_OBS_MODE")
+    if not raw:
+        return _DEFAULT_MODE
+    mode = raw.strip().lower()
+    return mode if mode in _VALID_MODES else _DEFAULT_MODE
+
+
+def _lgtm_ids() -> Optional[Tuple[str, str]]:
+    try:
+        from opentelemetry import trace
+    except ImportError:
+        return None
+    ctx = trace.get_current_span().get_span_context()
+    if ctx and ctx.is_valid:
+        return format(ctx.trace_id, "032x"), format(ctx.span_id, "016x")
+    return None
+
+
+def _ddtrace_ids() -> Optional[Tuple[str, str]]:
+    try:
+        from ddtrace.trace import tracer
+    except ImportError:
+        return None
+    ctx = tracer.current_trace_context()
+    if ctx and ctx.trace_id:
+        return format(ctx.trace_id, "032x"), format(ctx.span_id or 0, "016x")
+    return None
+
+
+def obs_correlation() -> Dict[str, str]:
+    """Return ``{"obs.trace_id": ..., "obs.span_id": ...}`` for the active
+    observability context, or ``{}`` if none is active.
+
+    Never fabricates ids -- this is a correlation tag, not the span's id.
+    """
+    mode = get_obs_mode()
+    if mode == LGTM:
+        ids = _lgtm_ids()
+    elif mode == DUAL:
+        ids = _lgtm_ids() or _ddtrace_ids()
+    else:  # dd_only
+        ids = _ddtrace_ids()
+
+    if not ids:
+        return {}
+    return {"obs.trace_id": ids[0], "obs.span_id": ids[1]}

src/agentex/lib/core/tracing/trace.py

@@ -11,6 +11,7 @@
 from agentex.types.span import Span
 from agentex.lib.utils.logging import make_logger
 from agentex.lib.utils.model_utils import recursive_model_dump
+from agentex.lib.core.tracing.obs_ids import obs_correlation
 from agentex.lib.core.tracing.span_error import set_span_error
 from agentex.lib.core.tracing.span_queue import (
     SpanEventType,
@@ -79,6 +80,12 @@ def start_span(
 
         serialized_input = recursive_model_dump(input) if input else None
         serialized_data = recursive_model_dump(data) if data else None
+        # Tag the business span with the active observability trace_id/span_id
+        # (OTel/ddtrace) so it can be correlated to the per-turn obs trace. The
+        # business trace_id stays the run-level task id -- see obs_ids.py.
+        obs = obs_correlation()
+        if obs:
+            serialized_data = {**(serialized_data or {}), **obs}
         id = str(uuid.uuid4())
 
         span = Span(
@@ -229,6 +236,12 @@ async def start_span(
 
         serialized_input = recursive_model_dump(input) if input else None
         serialized_data = recursive_model_dump(data) if data else None
+        # Tag the business span with the active observability trace_id/span_id
+        # (OTel/ddtrace) so it can be correlated to the per-turn obs trace. The
+        # business trace_id stays the run-level task id -- see obs_ids.py.
+        obs = obs_correlation()
+        if obs:
+            serialized_data = {**(serialized_data or {}), **obs}
         id = str(uuid.uuid4())
 
         span = Span(

src/agentex/resources/agents/schedules.py

@@ -277,6 +277,7 @@ def list(
         self,
         agent_id: str,
         *,
+        include_live: bool | Omit = omit,
         limit: int | Omit = omit,
         # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
         # The extra values given here take precedence over values defined on the client or passed to this method.
@@ -289,6 +290,8 @@ def list(
         List run schedules for an agent.
 
         Args:
+          include_live: Include live Temporal state and upcoming action times.
+
           extra_headers: Send extra headers
 
           extra_query: Add additional query parameters to the request
@@ -306,7 +309,13 @@ def list(
                 extra_query=extra_query,
                 extra_body=extra_body,
                 timeout=timeout,
-                query=maybe_transform({"limit": limit}, schedule_list_params.ScheduleListParams),
+                query=maybe_transform(
+                    {
+                        "include_live": include_live,
+                        "limit": limit,
+                    },
+                    schedule_list_params.ScheduleListParams,
+                ),
             ),
             cast_to=ScheduleListResponse,
         )
@@ -1056,6 +1065,7 @@ async def list(
         self,
         agent_id: str,
         *,
+        include_live: bool | Omit = omit,
         limit: int | Omit = omit,
         # Use the following arguments if you need to pass additional parameters to the API that aren't available via kwargs.
         # The extra values given here take precedence over values defined on the client or passed to this method.
@@ -1068,6 +1078,8 @@ async def list(
         List run schedules for an agent.
 
         Args:
+          include_live: Include live Temporal state and upcoming action times.
+
           extra_headers: Send extra headers
 
           extra_query: Add additional query parameters to the request
@@ -1085,7 +1097,13 @@ async def list(
                 extra_query=extra_query,
                 extra_body=extra_body,
                 timeout=timeout,
-                query=await async_maybe_transform({"limit": limit}, schedule_list_params.ScheduleListParams),
+                query=await async_maybe_transform(
+                    {
+                        "include_live": include_live,
+                        "limit": limit,
+                    },
+                    schedule_list_params.ScheduleListParams,
+                ),
             ),
             cast_to=ScheduleListResponse,
         )