loongsuite-python 0.8.0

loongsuite-python 0.8.0

Installation

pip install loongsuite-distro==0.8.0
loongsuite-bootstrap -a install --version 0.8.0

Package Versions

• loongsuite-* packages: 0.8.0
• opentelemetry-* packages: 0.60b1

PyPI packages

The following distributions are built and uploaded to PyPI for this release:

• loongsuite-otel-util-genai
• loongsuite-distro
• loongsuite-site-bootstrap
• loongsuite-instrumentation-agentscope
• loongsuite-instrumentation-algotune
• loongsuite-instrumentation-autogen
• loongsuite-instrumentation-bfclv4
• loongsuite-instrumentation-claude-agent-sdk
• loongsuite-instrumentation-claw-eval
• loongsuite-instrumentation-crewai
• loongsuite-instrumentation-dashscope
• loongsuite-instrumentation-deepagents
• loongsuite-instrumentation-google-adk
• loongsuite-instrumentation-hermes-agent
• loongsuite-instrumentation-langchain
• loongsuite-instrumentation-langgraph
• loongsuite-instrumentation-litellm
• loongsuite-instrumentation-mem0
• loongsuite-instrumentation-microsoft-agent-framework
• loongsuite-instrumentation-minisweagent
• loongsuite-instrumentation-openhands
• loongsuite-instrumentation-qwen-agent
• loongsuite-instrumentation-qwenpaw
• loongsuite-instrumentation-slop-code
• loongsuite-instrumentation-terminus2
• loongsuite-instrumentation-vita
• loongsuite-instrumentation-webarena
• loongsuite-instrumentation-widesearch
• loongsuite-instrumentation-wildtool

---

loongsuite

Changed

• Adopt upstream semantic conventions for LoongSuite instrumentation packages.

---

loongsuite-otel-util-genai

Added

• Add multimodal runtime config snapshot and generation-aware uploader hot-reload
  (MultimodalRuntimeConfig, update_multimodal_runtime_config,
  get_or_rebuild_uploader_pair).
• Add optional MultimodalUsageRecorder hooks for FsUploader upload
  success/error recording. Default is no-op.
• Add hook_advice and async_hook_advice fail-open decorators for
  instrumentation-only logic, with explicit rejection of deferred generator
  lifecycles.

Fixed

• Keep multimodal tests compatible with downstream Python 3.8 consumers.

---

loongsuite-instrumentation-agentscope

Fixed

• Derive AgentScope v2 ReAct rounds from the agent iteration state instead of
  a ContextVar token that could fail when QwenPaw advances a reply stream
  from different asyncio tasks.
• Restore Agent, ReAct, and tool span contexts for each stream advance so
  cross-task QwenPaw handoffs preserve parentage, and treat GeneratorExit as
  a successful stream close instead of a tool failure.
• Start AgentScope v2 streaming LLM spans before invoking the underlying model,
  restore their context for each stream operation, and treat cross-task
  GeneratorExit as a successful close while preserving TTFT.
• Model AgentScope v2 ReAct iterations from on_reasoning so each react step
  span parents both its LLM call and tool executions, including concurrent
  tools and QwenPaw cross-task stream handoffs.
• Capture AgentScope v2 string message content as text parts so LLM input and
  output message attributes are populated when content capture is enabled.

Added

• Capture gen_ai.skill.name, gen_ai.skill.id, and available skill metadata
  on AgentScope v2 built-in Skill viewer tool spans.

---

loongsuite-instrumentation-agno

Changed

• Split Agno 2.x tool-call loops into ReAct step spans with one LLM span per
  provider request, and populate streaming agent token usage from child LLM
  spans when Agno run events do not include metrics.
• Propagate Agno run user and session identity to model and tool spans while
  preferring ENTRY baggage identity over Agno framework-provided values.

Added

• Capture gen_ai.skill.* attributes on Agno skill tool spans such as
  get_skill_instructions, get_skill_reference, and get_skill_script.

---

loongsuite-instrumentation-autogen

Fixed

• Avoid enriching non-AutoGen spans that use overlapping GenAI operation names.

---

loongsuite-instrumentation-claude-agent-sdk

Fixed

• Capture Claude Agent SDK default model names from stream metadata when the
  client options and environment do not provide an explicit model.
• Fail open LLM spans when Claude Agent SDK streams raise exceptions, preserving
  parent context cleanup for subsequent calls.
• Preserve tool-only assistant messages as tool spans without fabricating LLM
  spans.

---

loongsuite-instrumentation-google-adk

Added

• Capture gen_ai.skill.* attributes on Google ADK SkillToolset
  load_skill and load_skill_resource execute-tool spans.

Changed

• Align the supported Python and CI matrix with Google ADK's Python 3.10
  runtime floor.

---

loongsuite-instrumentation-litellm

Fixed

• Route streamed chunk accumulation and finalization callbacks through shared
  fail-open advice while keeping stream iteration and lifecycle handling in the
  LiteLLM wrapper.
• Isolate prepare, response mapping, success, and error advice from application
  completion calls, and detach streaming context before cross-task consumption.

---

loongsuite-instrumentation-microsoft-agent-framework

Fixed

• Avoid enriching non-MAF spans that use overlapping GenAI operation names.
• Preserve parent-child context for legacy Microsoft Agent Framework streaming
  agent spans.
• Stop exporting process-cumulative GenAI ObservableGauges and use Microsoft
  Agent Framework's native metric instruments instead.
• Keep non-streaming MAF spans current for nested work and scope Robin provider
  suppression to the active LLM call when that optional integration is present.

---

loongsuite-instrumentation-qwenpaw

Added

• Added QwenPaw 2 Entry telemetry around each Runtime.run request while
  preserving the QwenPaw 1 and legacy CoPaw entry points.
• Added task-safe stream context handoff so QwenPaw heartbeat tasks inherit the
  Entry span without keeping an attached context token across yields.
• Records QwenPaw 2 Entry TTFT only after user-visible output and preserves
  business stream cleanup when Entry telemetry setup fails.

Fixed

• Restore the Entry context while closing downstream streams and preserve
  downstream aclose() errors instead of finalizing the Entry as successful.