Merge branch 'main' into fix/mcp-mtls-auth-header-case

Author: h-tsuboi918

.github/workflows/pr-triage.yml

@@ -15,22 +15,24 @@
 name: ADK Pull Request Triaging Agent
 
 on:
-  pull_request_target:
-    types: [opened, reopened, edited]
+  schedule:
+    # Run every 6 hours
+    - cron: '0 */6 * * *'
   workflow_dispatch:
     inputs:
       pr_number:
-        description: 'The Pull Request number to triage'
-        required: true
+        description: 'The Pull Request number to triage (leave empty for batch mode)'
+        required: false
+        type: 'string'
+      pr_count:
+        description: 'Number of PRs to process in batch mode (default: 10)'
+        required: false
+        default: '10'
         type: 'string'
 
 jobs:
   agent-triage-pull-request:
-    if: >-
-      github.event_name == 'workflow_dispatch' || (
-        github.event.pull_request.head.repo.full_name == github.repository &&
-        !contains(github.event.pull_request.labels.*.name, 'google-contributor')
-      )
+    if: github.repository == 'google/adk-python'
     runs-on: ubuntu-latest
     permissions:
       pull-requests: write
@@ -57,7 +59,8 @@ jobs:
           GOOGLE_GENAI_USE_VERTEXAI: 0
           OWNER: 'google'
           REPO: 'adk-python'
-          PULL_REQUEST_NUMBER: ${{ github.event.pull_request.number || github.event.inputs.pr_number }}
+          PULL_REQUEST_NUMBER: ${{ github.event.pull_request.number || github.event.inputs.pr_number || '' }}
+          PR_COUNT_TO_PROCESS: ${{ github.event.inputs.pr_count || '10' }}
           INTERACTIVE: ${{ vars.PR_TRIAGE_INTERACTIVE }}
           PYTHONPATH: contributing/samples/adk_team
         run: python -m adk_pr_triaging_agent.main

AGENTS.md

@@ -1,33 +1,3 @@
-# AI Coding Assistant Context
-
-This document provides context for AI coding assistants (Antigravity, Gemini CLI, etc.) to understand the ADK Python project and assist with development.
-
-## ADK Knowledge, Architecture, and Style
-
-For all matters regarding ADK development, please use the appropriate skill:
-
-- **`adk-architecture`**: Use this skill whenever you need to understand the architecture, event flow, or state management of the ADK system, or when designing or modifying core components and public APIs.
-  - Read `.agents/skills/adk-architecture/SKILL.md` for full instructions.
-- **`adk-style`**: Use this skill whenever writing code, tests, or reviewing PRs for the ADK project to ensure compliance with styling and coding conventions. Also use it for committing, bug fixing, and testing rules.
-  - Read `.agents/skills/adk-style/SKILL.md` for full instructions.
-- **`adk-git`**: Use this skill for any git operation (commit, push, pull, rebase, etc.). It provides guidelines for Conventional Commits and branch naming.
-  - Read `.agents/skills/adk-git/SKILL.md` for full instructions.
-- **`adk-sample-creator`**: Use this skill when creating new samples demonstrating features or agent patterns, or when adding examples to subdirectories under `contributing/`.
-  - Read `.agents/skills/adk-sample-creator/SKILL.md` for full instructions.
-- **`adk-review`**: Use this skill to review local changes for errors, style compliance, unintended outcomes, and to check if associated design docs, guides, samples, or tests need updates.
-  - Read `.agents/skills/adk-review/SKILL.md` for full instructions.
-- **`adk-issue`**: Use this skill when analyzing, triaging, and resolving GitHub issues for the adk-python repository (orchestrating both triage and fix implementation). Do NOT use this skill if the "/adk-issue-analyze" command is explicitly requested.
-  - Read `.agents/skills/adk-issue/SKILL.md` for full instructions.
-- **`adk-issue-analyze`**: Use this skill to fetch, inspect, and analyze a GitHub issue in a strictly read-only manner. Use this skill when the "/adk-issue-analyze" command is explicitly called.
-  - Read `.agents/skills/adk-issue-analyze/SKILL.md` for full instructions.
-- **`adk-issue-fix`**: Use this skill to implement the code changes, unit tests, and documentation updates for an approved GitHub issue fix. Use this skill when the "/adk-issue-fix" command is explicitly called.
-  - Read `.agents/skills/adk-issue-fix/SKILL.md` for full instructions.
-- **`adk-pr-analyze`**: Use this skill to fetch, inspect, and analyze a GitHub pull request in a strictly read-only manner. Use this skill when the "/adk-pr-analyze" command is explicitly called.
-  - Read `.agents/skills/adk-pr-analyze/SKILL.md` for full instructions.
-- **`adk-pr-triage`**: Use this skill to orchestrate triaging and reviewing GitHub pull requests (PRs) (orchestrating both analysis and user review/checkout). Do NOT use this skill if the "/adk-pr-analyze" command is explicitly requested.
-  - Read `.agents/skills/adk-pr-triage/SKILL.md` for full instructions.
-
-
 ## Project Overview
 
 The Agent Development Kit (ADK) is an open-source, code-first Python toolkit for building, evaluating, and deploying sophisticated AI agents.
@@ -45,6 +15,10 @@ The Agent Development Kit (ADK) is an open-source, code-first Python toolkit for
 
 For details on how the Runner works and the invocation lifecycle, please refer to the `adk-architecture` skill and the referenced documentation therein.
 
+## ADK Knowledge, Architecture, and Style
+
+Skills related to ADK development are in `.agents/skills/`.
+
 ## Project Architecture
 
 For detailed architecture patterns, component descriptions, and core interfaces, please refer to the **`adk-architecture`** skill at `.agents/skills/adk-architecture/SKILL.md`.

contributing/samples/adk_team/adk_pr_triaging_agent/agent.py

@@ -21,6 +21,7 @@
 from adk_pr_triaging_agent.settings import REPO
 from adk_pr_triaging_agent.utils import error_response
 from adk_pr_triaging_agent.utils import get_diff
+from adk_pr_triaging_agent.utils import get_request
 from adk_pr_triaging_agent.utils import post_request
 from adk_pr_triaging_agent.utils import read_file
 from adk_pr_triaging_agent.utils import run_graphql_query
@@ -219,6 +220,56 @@ def add_comment_to_pr(pr_number: int, comment: str) -> dict[str, Any]:
   }
 
 
+def list_untriaged_pull_requests(pr_count: int) -> dict[str, Any]:
+  """List open pull requests that need triaging.
+
+  Returns pull requests that need triaging (i.e. do not have google-contributor
+  label and do not have any allowed triage category labels).
+
+  Args:
+    pr_count: number of pull requests to return
+
+  Returns:
+    The status of this request, with a list of pull requests when successful.
+  """
+  url = f"{GITHUB_BASE_URL}/search/issues"
+  query = f"repo:{OWNER}/{REPO} is:open is:pr"
+  params = {
+      "q": query,
+      "sort": "created",
+      "order": "desc",
+      "per_page": 100,
+      "page": 1,
+  }
+
+  try:
+    response = get_request(url, params)
+  except requests.exceptions.RequestException as e:
+    return error_response(f"Error: {e}")
+
+  issues = response.get("items", [])
+  triage_labels = set(ALLOWED_LABELS)
+  untriaged_prs = []
+
+  for pr in issues:
+    pr_labels = {label["name"] for label in pr.get("labels", [])}
+    if "google-contributor" in pr_labels:
+      continue
+    # If it already has any of the ALLOWED_LABELS, skip it.
+    if pr_labels & triage_labels:
+      continue
+
+    untriaged_prs.append({
+        "number": pr["number"],
+        "title": pr["title"],
+    })
+
+    if len(untriaged_prs) >= pr_count:
+      break
+
+  return {"status": "success", "pull_requests": untriaged_prs}
+
+
 root_agent = Agent(
     model="gemini-3.5-flash",
     name="adk_pr_triaging_assistant",
@@ -276,15 +327,16 @@ def add_comment_to_pr(pr_number: int, comment: str) -> dict[str, Any]:
       > This information will help reviewers to review your PR more efficiently. Thanks!
 
       # 4. Steps
-      When you are given a PR, here are the steps you should take:
-      - Call the `get_pull_request_details` tool to get the details of the PR.
-      - Skip the PR (i.e. do not label or comment) if any of the following is true:
-        - the PR is closed
-        - the PR is labeled with "google-contributor"
-        - the PR is already labelled with the above labels (e.g. "documentation", "services", "tools", etc.).
-      - Check if the PR is following the contribution guidelines.
-        - If it's not following the guidelines, recommend or add a comment to the PR that points to the contribution guidelines (https://github.com/google/adk-python/blob/main/CONTRIBUTING.md).
-        - If it's following the guidelines, recommend or add a label to the PR.
+      - If you are asked to find pull requests that need triaging, use `list_untriaged_pull_requests` first.
+      - For each pull request to be triaged:
+        - Call the `get_pull_request_details` tool to get the details of the PR.
+        - Skip the PR (i.e. do not label or comment) if any of the following is true:
+          - the PR is closed
+          - the PR is labeled with "google-contributor"
+          - the PR is already labelled with the above labels (e.g. "documentation", "services", "tools", etc.).
+        - Check if the PR is following the contribution guidelines.
+          - If it's not following the guidelines, recommend or add a comment to the PR that points to the contribution guidelines (https://github.com/google/adk-python/blob/main/CONTRIBUTING.md).
+          - If it's following the guidelines, recommend or add a label to the PR.
 
       # 5. Output
       Present the following in an easy to read format highlighting PR number and your label.
@@ -293,6 +345,7 @@ def add_comment_to_pr(pr_number: int, comment: str) -> dict[str, Any]:
       - The comment you recommended or added to the PR with the justification
     """,
     tools=[
+        list_untriaged_pull_requests,
         get_pull_request_details,
         add_label_to_pr,
         add_comment_to_pr,

contributing/samples/adk_team/adk_pr_triaging_agent/main.py

@@ -18,6 +18,7 @@
 
 from adk_pr_triaging_agent import agent
 from adk_pr_triaging_agent.settings import OWNER
+from adk_pr_triaging_agent.settings import PR_COUNT_TO_PROCESS
 from adk_pr_triaging_agent.settings import PULL_REQUEST_NUMBER
 from adk_pr_triaging_agent.settings import REPO
 from adk_pr_triaging_agent.utils import call_agent_async
@@ -41,13 +42,20 @@ async def main():
   )
 
   pr_number = parse_number_string(PULL_REQUEST_NUMBER)
-  if not pr_number:
+  if pr_number:
+    prompt = f"Please triage pull request #{pr_number}!"
+  else:
+    pr_count = parse_number_string(PR_COUNT_TO_PROCESS, default_value=10)
     print(
-        f"Error: Invalid pull request number received: {PULL_REQUEST_NUMBER}."
+        "No pull request number received. Operating in batch mode (limit:"
+        f" {pr_count})."
+    )
+    prompt = (
+        f"Please use 'list_untriaged_pull_requests' to find {pr_count} pull"
+        " requests that need triaging, then triage each one according to your"
+        " instructions."
     )
-    return
 
-  prompt = f"Please triage pull request #{pr_number}!"
   response = await call_agent_async(runner, USER_ID, session.id, prompt)
   print(f"<<<< Agent Final Output: {response}\n")
 

contributing/samples/adk_team/adk_pr_triaging_agent/settings.py

@@ -28,5 +28,6 @@
 OWNER = os.getenv("OWNER", "google")
 REPO = os.getenv("REPO", "adk-python")
 PULL_REQUEST_NUMBER = os.getenv("PULL_REQUEST_NUMBER")
+PR_COUNT_TO_PROCESS = os.getenv("PR_COUNT_TO_PROCESS", "10")
 
 IS_INTERACTIVE = os.environ.get("INTERACTIVE", "1").lower() in ["true", "1"]

src/google/adk/agents/config_agent_utils.py

@@ -86,8 +86,8 @@ def _load_config_from_path(config_path: str) -> AgentConfig:
   """Load an agent's configuration from a YAML file.
 
   Args:
-    config_path: Path to the YAML config file. Both relative and absolute
-      paths are accepted.
+    config_path: Path to the YAML config file. Both relative and absolute paths
+      are accepted.
 
   Returns:
     The loaded and validated AgentConfig object.
@@ -124,21 +124,31 @@ def resolve_agent_reference(
   Args:
     ref_config: The agent reference configuration (AgentRefConfig).
     referencing_agent_config_abs_path: The absolute path to the agent config
-    that contains the reference.
+      that contains the reference.
 
   Returns:
     The created agent instance.
   """
   if ref_config.config_path:
     if os.path.isabs(ref_config.config_path):
-      return from_config(ref_config.config_path)
-    else:
-      return from_config(
-          os.path.join(
-              os.path.dirname(referencing_agent_config_abs_path),
-              ref_config.config_path,
-          )
+      raise ValueError(
+          "Absolute paths are not allowed in AgentRefConfig config_path:"
+          f" {ref_config.config_path!r}"
       )
+    agent_dir = os.path.dirname(referencing_agent_config_abs_path)
+    resolved_path = os.path.realpath(
+        os.path.join(agent_dir, ref_config.config_path)
+    )
+    canonical_agent_dir = os.path.realpath(agent_dir)
+    if (
+        os.path.commonpath([canonical_agent_dir, resolved_path])
+        != canonical_agent_dir
+    ):
+      raise ValueError(
+          f"Path traversal detected: config_path {ref_config.config_path!r}"
+          " resolves outside the agent directory"
+      )
+    return from_config(resolved_path)
   elif ref_config.code:
     return _resolve_agent_code_reference(ref_config.code)
   else:

src/google/adk/agents/llm_agent.py

@@ -520,9 +520,13 @@ async def _run_async_impl(
       return
 
     should_pause = False
+    output_accumulator = ''
     async with Aclosing(self._llm_flow.run_async(ctx)) as agen:
       async for event in agen:
         self.__maybe_save_output_to_state(event)
+        output_accumulator = self.__maybe_accumulate_streaming_output(
+            event, output_accumulator
+        )
         yield event
         if ctx.should_pause_invocation(event):
           # Do not pause immediately, wait until the long-running tool call is
@@ -544,9 +548,13 @@ async def _run_async_impl(
   async def _run_live_impl(
       self, ctx: InvocationContext
   ) -> AsyncGenerator[Event, None]:
+    output_accumulator = ''
     async with Aclosing(self._llm_flow.run_live(ctx)) as agen:
       async for event in agen:
         self.__maybe_save_output_to_state(event)
+        output_accumulator = self.__maybe_accumulate_streaming_output(
+            event, output_accumulator
+        )
         yield event
       if ctx.end_invocation:
         return
@@ -962,6 +970,47 @@ def __maybe_save_output_to_state(self, event: Event):
         result = validate_schema(self.output_schema, result)
       event.actions.state_delta[self.output_key] = result
 
+  def __maybe_accumulate_streaming_output(
+      self, event: Event, accumulator: str
+  ) -> str:
+    """Accumulates output_key text across a streaming model turn.
+
+    Streaming with tool calls produces non-partial events that carry text
+    alongside a function_call. is_final_response() rejects those, so
+    __maybe_save_output_to_state skips them and the text on those events
+    is dropped from output_key. Accumulate every non-partial text-bearing
+    event from this agent across the model turn so the segments survive
+    in session state. See issue #5590.
+
+    No-op when accumulation doesn't apply (different author, no
+    output_key, output_schema set, partial event, no content, no text).
+    For applicable events, appends the event's text to ``accumulator``
+    and writes the running value to state_delta[output_key], overwriting
+    any value __maybe_save_output_to_state set on the same event.
+    Returns the new accumulator value.
+    """
+    if (
+        not self.output_key
+        or self.output_schema
+        or event.author != self.name
+        or event.partial
+        or not event.content
+        or not event.content.parts
+    ):
+      return accumulator
+
+    text = ''.join(
+        part.text
+        for part in event.content.parts
+        if part.text and not part.thought
+    )
+    if not text:
+      return accumulator
+
+    accumulator += text
+    event.actions.state_delta[self.output_key] = accumulator
+    return accumulator
+
   @model_validator(mode='after')
   def __model_validator_after(self) -> LlmAgent:
     return self

src/google/adk/cli/utils/local_storage.py

@@ -236,6 +236,21 @@ async def append_event(self, session: Session, event: Event) -> Event:
     service = await self._get_service(session.app_name)
     return await service.append_event(session, event)
 
+  async def close(self) -> None:
+    """Closes all underlying session services."""
+    for service in self._services.values():
+      if hasattr(service, "close"):
+        await service.close()
+    self._services.clear()
+
+  async def __aenter__(self) -> PerAgentDatabaseSessionService:
+    """Enters the async context manager."""
+    return self
+
+  async def __aexit__(self, exc_type, exc_val, exc_tb) -> None:
+    """Exits the async context manager and closes the service."""
+    await self.close()
+
 
 class PerAgentFileArtifactService(BaseArtifactService):
   """Routes artifact storage to per-agent `.adk/artifacts` folders."""

src/google/adk/flows/llm_flows/base_llm_flow.py

@@ -578,18 +578,9 @@ async def run_live(
         # initial_history_in_client_content to True. This tells the Live server
         # that the provided history already includes the model's past responses,
         # preventing the server from generating duplicate responses for those replayed turns.
-        #
-        # ``history_config`` is only supported by the Gemini Developer API
-        # backend; the Vertex AI / Gemini Enterprise Agent Platform backend has
-        # no such field on its live setup message and rejects it. On Vertex,
-        # history is instead seeded by ``send_history`` below
-        # (``send_client_content`` with prior turns), so we skip
-        # ``history_config`` for that backend.
         if (
             llm_request.contents
             and not invocation_context.live_session_resumption_handle
-            and isinstance(llm, Gemini)
-            and llm._api_backend == GoogleLLMVariant.GEMINI_API  # pylint: disable=protected-access
         ):
           if not llm_request.live_connect_config:
             llm_request.live_connect_config = types.LiveConnectConfig()