Capture the agent's system prompt as the gen_ai.system_instructions span attribute in the LangChain instrumentation

[rads-1996]

No description provided.

[github-actions]

Performance comparison

Threshold: regressions >15.0% on gating scenarios fail the build. Higher ops/s is better; positive Δ means the PR is slower.

Scenario	Gating	Baseline (ops/s)	Candidate (ops/s)	Δ %	Status
azure_monitor_log	yes	19,033.5	19,088.0	-0.29%	✅
azure_monitor_span	yes	154,990.7	154,750.9	+0.15%	✅
otel_log	no	21,956.3	22,424.5	-2.09%	✅
otel_span	no	38,507.5	38,974.2	-1.20%	✅

[Copilot]

Pull request overview

This PR extends the LangChain GenAI instrumentation to capture an agent/LLM “system prompt” into the OTel semantic convention attribute gen_ai.system_instructions, including routing system messages out of gen_ai.input.messages and gating emission behind the existing sensitive-content controls.

Changes:

• Add _extract_system_instruction support for extracting system content from inputs["messages"] (in addition to inputs["prompts"]), and route system messages out of structured gen_ai.input.messages.
• Emit gen_ai.system_instructions on both LLM spans (_update_span) and agent wrapper spans (_finalize_agent_span) when content capture is enabled.
• Add unit tests for extraction, routing/no-leak guarantees, and gating behavior; update changelog for the new feature.

Reviewed changes

Copilot reviewed 5 out of 5 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/langchain/test_utils.py	Adds tests for system-instruction extraction and ensuring system prompts don’t leak into gen_ai.input.messages.
tests/langchain/test_tracer.py	Adds tests to verify gating and serialization shape of gen_ai.system_instructions on LLM and agent spans.
src/microsoft/opentelemetry/_genai/_langchain/_utils.py	Implements system-instruction extraction and routes system messages out of structured input messages.
src/microsoft/opentelemetry/_genai/_langchain/_tracer.py	Emits gen_ai.system_instructions on spans and aggregates first-child system instructions onto agent spans.
CHANGELOG.md	Documents the new unreleased feature.

Comments suppressed due to low confidence (1)

src/microsoft/opentelemetry/_genai/_langchain/_utils.py:1355

• _extract_structured_input_messages uses next(iter(multiple_messages)), which causes only the first item to be processed when LangChain provides a flat {"messages": [...]} list. With the new route_system_out behavior, a flat list starting with a system message would be fully dropped (system skipped, subsequent user messages never reached). Normalize flat-vs-nested messages like input_messages() to ensure all messages are processed.

    if multiple_messages and isinstance(multiple_messages, Iterable):
        first_messages = next(iter(multiple_messages), None)
        if first_messages is not None:

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.