test(tracing): cover Markdown trace export

tests/test_litellm/tracing/test_markdown.py

@@ -0,0 +1,139 @@
+"""
+Tests for the Markdown trace export (GET /v1/traces/{id}?format=md).
+"""
+
+import json
+import os
+import sys
+
+sys.path.insert(0, os.path.abspath("../../.."))
+
+from litellm.tracing.markdown import trace_to_markdown
+from litellm.tracing.types import LiteLLMRequest, Span, Trace, TraceSummary
+
+
+def _span(span_id, parent, name, type_, start, **extra) -> Span:
+    return Span(
+        span_id=span_id,
+        parent_span_id=parent,
+        name=name,
+        type=type_,
+        agent=extra.pop("agent", "lead"),
+        start_offset_ms=start,
+        duration_ms=extra.pop("duration_ms", 1000.0),
+        status=extra.pop("status", "ok"),
+        error=extra.pop("error", None),
+        input_preview="",
+        model=extra.pop("model", None),
+        input_tokens=0,
+        output_tokens=0,
+        litellm=extra.pop("litellm", None),
+    )
+
+
+REQUEST = LiteLLMRequest(
+    request_id="chatcmpl-1",
+    model="openai/claude-sonnet-4-5",
+    model_group="claude-sonnet-4-5",
+    provider="openai",
+    key_alias="research-bot",
+    team_alias="research-agents",
+    spend=0.0087,
+    prompt_tokens=3378,
+    completion_tokens=514,
+    cache_read_tokens=0,
+    cache_write_tokens=0,
+    latency_ms=8278,
+    ttft_ms=None,
+    status="success",
+)
+
+SPANS = [
+    _span("root", None, "lead", "agent", 0),
+    _span("mw", "root", "FilesystemMiddleware.wrap_model_call", "framework", 1),
+    _span("llm", "mw", "ChatOpenAI", "llm", 2, litellm=REQUEST, model="claude-sonnet-4-5"),
+    _span("task", "root", "task", "tool", 3),
+    _span("sub", "task", "researcher", "agent", 4, agent="researcher"),
+    _span(
+        "grep",
+        "sub",
+        "grep_code",
+        "tool",
+        5,
+        agent="researcher",
+        status="error",
+        error="TimeoutError('Timeout after 30s')Traceback (most recent call last):\n  File x",
+    ),
+]
+TRACE = Trace(
+    summary=TraceSummary(
+        trace_id="t1",
+        name="lead",
+        service="research-agent",
+        input_preview="",
+        start_time="2026-09-30T00:00:00+00:00",
+        duration_ms=10_000,
+        status="ok",
+        span_count=6,
+        agent_count=2,
+        agent_invocations=2,
+        llm_calls=1,
+        tool_calls=2,
+        error_count=1,
+        input_tokens=3378,
+        output_tokens=514,
+        spend=0.0087,
+        models=["claude-sonnet-4-5"],
+    ),
+    agents=[],
+    spans=SPANS,
+)
+IO = {
+    "root": (
+        json.dumps([{"role": "user", "content": "Fix the flaky login test"}]),
+        json.dumps({"role": "assistant", "content": "Fixed it."}),
+    ),
+    "llm": (
+        "",
+        json.dumps(
+            {
+                "role": "assistant",
+                "content": "",
+                "tool_calls": [{"name": "task", "args": {"subagent_type": "researcher"}}],
+            }
+        ),
+    ),
+    "grep": ('{"q": "session"}', ""),
+}
+
+
+def test_markdown_reads_input_steps_output():
+    md = trace_to_markdown(TRACE, IO)
+    assert md.startswith("# Agent trace: lead")
+    assert "## Input\n\n**user**: Fix the flaky login test" in md
+    assert "## Output\n\n**assistant**: Fixed it." in md
+    assert "1 errors" in md
+
+
+def test_framework_spans_are_skipped_and_children_lifted():
+    md = trace_to_markdown(TRACE, IO)
+    assert "wrap_model_call" not in md
+    # the LLM call is lifted to depth 0 under the root, with LiteLLM cost and the tool call it made
+    assert "- **llm** `ChatOpenAI` · 1.00s · claude-sonnet-4-5 · 3378→514 tok · $0.0087" in md
+    assert '**assistant** → `task({"subagent_type": "researcher"})`' in md
+
+
+def test_subagents_nest_and_failures_show_first_error_line():
+    md = trace_to_markdown(TRACE, IO)
+    assert "  - **subagent** `researcher`" in md
+    assert "    - **tool** `grep_code` · 1.00s · **FAILED**" in md
+    assert "error: `TimeoutError('Timeout after 30s')`" in md
+    assert "Traceback" not in md
+
+
+def test_span_id_exports_only_that_subtree():
+    md = trace_to_markdown(TRACE, IO, span_id="sub")
+    assert "## Subtree of `researcher`" in md
+    assert "grep_code" in md
+    assert "ChatOpenAI" not in md
+    assert "## Input" not in md