fix(runtime): persist resolved model name to DB even when not explicitly provided

[yitang]

Summary

Fixes the model_name column in the runs database table being permanently empty. When a user starts a run without specifying a model_name in the request context, the agent resolves an effective model at runtime, but this was never persisted to the database due to a guard condition that skipped the update entirely when record.model_name was None.

Changes

File: backend/packages/harness/deerflow/runtime/runs/worker.py (lines 237–242)

• Removed the "record.model_name is not None" early-exit guard
• The code now always resolves the effective model name from agent metadata and persists it to the DB when available
• When model_name was already set at creation time, behavior is unchanged (same or different effective model gets written)
• When model_name was never set (the bug case), the resolved default model is now correctly persisted

Test Plan

• All 64 tests in tests/test_run_manager.py pass (including test_model_name_create_or_reject, test_model_name_default_is_none)
• Python syntax check via ruff: clean
• Manual verification: start a run without explicit model_name, query the DB to confirm it's populated

Related Issues

Closes #3109

[Copilot]

Pull request overview

Fixes run metadata persistence so the runs.model_name column is populated even when the user doesn’t explicitly provide a model name, by persisting the agent-resolved effective model name during runtime.

Changes:

• Update run_agent() to always read agent.metadata["model_name"] (when available) and persist it via run_manager.update_model_name(...).
• Add .agents/plans/ to .gitignore.

Reviewed changes

Copilot reviewed 1 out of 2 changed files in this pull request and generated 1 comment.

File	Description
backend/packages/harness/deerflow/runtime/runs/worker.py	Removes the record.model_name is not None guard and persists the resolved effective model name from agent metadata.
.gitignore	Ignores .agents/plans/ directory.

Comments suppressed due to low confidence (1)

backend/packages/harness/deerflow/runtime/runs/worker.py:244

• The new logic updates the run’s model_name whenever effective is truthy, even if it’s identical to the already-persisted value. Because the SQL store’s update_model_name always writes updated_at (and doesn’t short-circuit on same value), this introduces an extra DB write and changes semantics by bumping updated_at for runs that already had the correct model_name. Consider restoring a value-change guard, e.g. only call update_model_name when effective is set and effective != record.model_name (this still fixes the None case since None != effective).

        resolved = getattr(agent, "metadata", {}) or {}
        if isinstance(resolved, dict):
            effective = resolved.get("model_name")
            if effective:
                await run_manager.update_model_name(record.run_id, effective)