fix: handle dict word_list in TextArenaEnv.ta_to_hf() + add Supported Patterns docs

[vominh1919]

Changes

1. Fix TextArenaEnv.ta_to_hf() crash with dict word_list (Fixes #1074)

The ta_to_hf() method crashed with KeyError: 1 when word_list was a dict (e.g., TwentyQuestions-v0). This happened because random.choice(words) returned a dict key instead of a word.

Fix: Added check to flatten dict word_list into a list before random selection.

if isinstance(words, dict):
    words = [word for category_words in words.values() for word in category_words]

2. Add "Supported Patterns" section to README (Addresses #1078)

Added comprehensive documentation of supported RL framework design patterns to the README, covering:

• Context management
• User simulations
• Native tool parsing
• Sandboxing (all three modes)
• Reward systems (groupwise, intermediate, rubric composition)
• Multi-environment support
• Resource management
• Custom metrics & error handling
• Offline evals

This makes the design space more discoverable for new users.

Testing

• Verified fix works with both list and dict word_list formats
• README renders correctly on GitHub

Closes #1074
Closes #1078

---

Note

Low Risk
Small, localized bug fix to dataset generation plus documentation-only additions; no security-sensitive or broad behavioral changes.

Overview
Fixes a crash in TextArenaEnv.ta_to_hf() when TextArena provides word_list as a dict (e.g., category -> words) by flattening it before random.choice, ensuring dataset generation works across game types.

Updates README.md with a new Supported Patterns section that summarizes the library’s out-of-the-box RL environment/harness capabilities (context management, tool parsing, sandboxing modes, reward composition, multi-env evals, resource management, and offline eval tooling).

^{Reviewed by Cursor Bugbot for commit fba8562. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vominh1919]

Hi! I'd like to help with this issue.

I can analyze the problem and create a fix. Here's my approach:

1. Understand the issue - Read the description and reproduce the bug
2. Find the root cause - Locate the problematic code
3. Implement a fix - Create a minimal, targeted solution
4. Test the fix - Verify it works as expected
5. Submit a PR - Ready for your review

Would you like me to proceed?

---

Offering help with open issues to improve the project.