[devin-ai-integration]

Summary

agentlab-xray and several LLM modules crash with ModuleNotFoundError: No module named 'langchain.schema' because langchain v2 moved BaseMessage/HumanMessage to langchain_core.messages.

This PR updates three files to use a try/except import chain: try langchain_core.messages first → fall back to langchain.schema → gracefully degrade to None. This matches the conditional import pattern already used elsewhere in the codebase (e.g., the else branches already set these to None).

Files changed:

• agent_xray.py — import fix + None guards on isinstance checks
• llm_utils.py — same import fix for LangchainBaseMessage and convert_message_to_dict, plus a None guard on convert_message_to_dict usage with a role-aware dict fallback
• tracking.py — wrapped langchain_community.callbacks import in try/except

Key details

• LangchainBaseMessage and convert_message_to_dict are imported independently (separate try/except blocks), so one can succeed while the other fails.
• When convert_message_to_dict is None but LangchainBaseMessage is available, messages_to_dict() falls back to a manual role mapping (human→user, ai→assistant, system→system) via m.type instead of calling None(m).

Review & Testing Checklist for Human

• Fallback role_map completeness: The manual fallback in llm_utils.py:55 maps human, ai, and system types. Other langchain message types (e.g., tool, function) would pass through as-is via role_map.get(m.type, m.type). Verify this is acceptable or if additional mappings are needed.
• str(m.content) in fallback may lose structured content: If m.content is a list (e.g., multimodal content blocks), str() will serialize it as a Python repr. This only applies in the rare fallback path (langchain_core installed without langchain_community).
• Downstream None handling in agent_xray.py: When BaseMessage=None, format_chat_message falls through to str(message) and submit_action falls through to the ValueError raise. Confirm this is acceptable when langchain is not installed.
• Test plan: Run agentlab-xray with and without langchain installed. With langchain: legacy message formats should still render correctly. Without: the app should launch and function for dict-based messages.

Notes

• langchain is not declared in pyproject.toml or uv.lock — it was always an undeclared optional dependency. This PR doesn't change that; it just stops the hard crash.
• All 68 tests pass locally (15 skipped, matching CI expectations).
• No new tests cover the fallback paths specifically since the fallback is only exercised in unusual dependency configurations (langchain_core without langchain_community).

Link to Devin session: https://app.devin.ai/sessions/991a2a652ca44d36a06d4b59397593d0
Requested by: @ShawnAzman

---

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment and CI monitoring