Add archive-guided mutation prompts via TreeSHAP

[nikivanstein]

Summary

• add a feature-guided mutation mode that trains an XGBoost model on AST features and steers prompts with TreeSHAP insights
• expose reusable AST and complexity feature extraction utilities and a lightweight TreeSHAP wrapper
• expand automated tests to cover the new guidance workflow and ensure mutation prompts include guidance

[codecov]

Codecov Report

❌ Patch coverage is 82.33438% with 56 lines in your changes missing coverage. Please review.

Files with missing lines	Patch %	Lines
llamea/feature_guidance.py	80.14%	28 Missing ⚠️
llamea/ast_features.py	82.57%	23 Missing ⚠️
llamea/llamea.py	86.66%	4 Missing ⚠️
llamea/treeshap.py	92.85%	1 Missing ⚠️

📢 Thoughts on this report? Let us know!

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7ead1c592d

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Clear stale guidance when parent guidance fails

In _update_feature_guidance, when compute_feature_guidance(...) returns None for a specific parent, the code only clears feature_guidance_message if parent is None. That means a parent-specific guidance failure leaves the previous message intact, and construct_prompt will append stale guidance to the mutation prompt for a different parent. This misguides mutations whenever guidance cannot be computed for a parent (e.g., insufficient training data or SHAP failure), so the guidance should be reset even when parent is provided.

Useful? React with 👍 / 👎.