fix: factor execution env crashes when no conda environment is active

[chrisbatso]

Problem

get_factor_env() in rdagent/components/coder/factor_coder/config.py unconditionally builds a conda-based environment:

if hasattr(conf, "python_bin"):
    env = LocalEnv(conf=(CondaConf(conda_env_name=os.environ.get("CONDA_DEFAULT_ENV"))))

hasattr(conf, "python_bin") is always True (python_bin is a declared settings field with a default), so on any host where the process is not running inside an activated conda environment, CONDA_DEFAULT_ENV is unset and CondaConf raises:

ValidationError: 1 validation error for CondaConf
conda_env_name
  Input should be a valid string [type=string_type, input_value=None, ...]

This makes rdagent fin_factor crash at loop 0 on any conda-less installation (e.g. a plain venv/uv setup on a fresh Ubuntu server), even though nothing about factor execution actually requires conda.

Fix

Keep the conda path when CONDA_DEFAULT_ENV is set; otherwise fall back to a LocalEnv whose bin_path is the current interpreter's own bin directory, so factor code runs with the environment RD-Agent itself was installed into.

Testing

Reproduced on Ubuntu 26.04 with RD-Agent installed in a uv-managed venv (no conda present): rdagent fin_factor crashed with the ValidationError above; with this patch, factor implementation, execution, and full qlib R&D loops (15 loops, multiple LLM backends) run to completion. The conda path is unchanged for users inside an activated env.

🤖 Generated with Claude Code

---

📚 Documentation preview 📚: https://RDAgent--1462.org.readthedocs.build/en/1462/