feat: Add LangGraph workflow auto-detection and agent tag parsing

[NikitaVoitov]

Summary

This PR adds automatic detection of LangGraph workflows and parsing of agent attributes from tags.

Problem

The current instrumentation lacks:

1. LangGraph auto-detection: Workflows using LangGraph don't get workflow_type or framework attributes
2. Agent tag parsing: Agent metadata passed via tags (common pattern) is ignored
3. Workflow descriptions: No way to capture workflow/agent descriptions for better observability

Root Cause

1. on_chain_start doesn't check for LangGraph-specific metadata keys (langgraph_version, graph_id)
2. Agent attributes from tags (agent_type:, agent_description:, agent_tools:) aren't parsed
3. description: tags and ls_description metadata aren't extracted

Solution

1. LangGraph Auto-Detection

# In on_chain_start - detect LangGraph from metadata
if metadata:
    if metadata.get("langgraph_version") or metadata.get("graph_id"):
        wf.workflow_type = "graph"
        wf.framework = "langgraph"
    # Extract description from metadata
    desc = metadata.get("ls_description") or metadata.get("description")
    if desc:
        wf.description = _safe_str(desc)

2. Agent Tag Parsing

# In _start_agent_invocation - parse agent attributes from tags
if tags:
    for tag in tags:
        if tag.startswith("agent_type:"):
            agent.agent_type = tag[11:]
        elif tag.startswith("agent_description:"):
            agent.description = tag[18:]
        elif tag.startswith("agent_tools:"):
            agent.tools = tag[12:].split(",")

3. Workflow Description from Tags

# Support description via tag
if not wf.description:
    for tag in tags:
        if tag.startswith("description:"):
            wf.description = tag[len("description:"):]
            break
# Auto-generate from graph_id if not set
if not wf.description and metadata.get("graph_id"):
    wf.description = f"LangGraph workflow: {metadata['graph_id']}"

Evidence

Before Fix (trace 27ba4165c26889bb6d328fd356d9381d)

Attribute	Value
gen_ai.workflow.type	❌ Missing
gen_ai.framework	❌ Missing
gen_ai.workflow.description	❌ Missing
gen_ai.agent.type	❌ Missing
gen_ai.agent.description	❌ Missing
gen_ai.agent.tools	❌ Missing

After Fix

Agent trace (fe915fd88fd3c8542e2adee30b42559f):

Attribute	Value
gen_ai.agent.name	✅ healthcare_agent
gen_ai.agent.type	✅ react
gen_ai.agent.description	✅ Healthcare agent
gen_ai.agent.tools	✅ ["query_member_data","search_knowledge_base"]
gen_ai.framework	✅ langchain

Workflow trace (8593063fe9ba82df06d5f7a126e658b2):

Attribute	Value
gen_ai.workflow.type	✅ graph
gen_ai.workflow.description	✅ Healthcare processing workflow
gen_ai.framework	✅ langgraph

Testing

Added 5 unit tests:

• test_langgraph_workflow_auto_detection_from_metadata
• test_agent_tag_parsing_name
• test_agent_tag_parsing_type_and_description
• test_agent_tag_parsing_tools
• test_description_tag_for_workflow

Files Changed

File	Changes
callback_handler.py	LangGraph detection + agent tag parsing logic
test_callback_handler_agent.py	+5 tests

Why This Matters

1. Agent Flow Visualization: workflow_type enables proper visualization in Splunk O11y
2. LangGraph Support: LangGraph dev server replaces custom metadata but preserves tags - tag parsing ensures attributes survive
3. Better Observability: Agent descriptions and tools visible in traces aid debugging

[github-actions]

Thank you for your submission, we really appreciate it. Like many open-source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution. You can sign the CLA by just posting a Pull Request Comment same as the below format.

---

I have read the CLA Document and I hereby sign the CLA

---

_{You can retrigger this bot by commenting recheck in this Pull Request. Posted by the CLA Assistant Lite bot.}