feat: integrate eval infra and part of the oracles

[ganler]

No description provided.

[Copilot]

Pull Request Overview

This PR integrates the evaluation infrastructure and implements parts of the oracle system for the PurpCode project. The changes focus on building a comprehensive evaluation framework for secure code generation with multiple safety oracles and assessment tools.

Key changes include:

• Addition of a safety-focused system prompt for secure code evaluation
• Implementation of evaluation infrastructure with support for multiple oracles (xscode, malicious assistance detection, etc.)
• Creation of annotation tools for dataset curation and quality assessment

Reviewed Changes

Copilot reviewed 23 out of 23 changed files in this pull request and generated 5 comments.

Show a summary per file

File	Description
utils/init.py	Adds system prompt for safety-focused code generation evaluation
eval/main.py	Main entry point for evaluation pipeline combining generation and assessment
eval/generate.py	Core generation infrastructure with multi-backend support (HF, vLLM, OpenAI, Bedrock)
eval/evaluate.py	Evaluation orchestrator that maps tasks to appropriate oracles
eval/oracles/xscode_overrefuse.py	Oracle for evaluating XSCode dataset refusal and security vulnerabilities
eval/oracles/malicious_assistance_detection.py	Oracle for detecting malicious code assistance in responses
eval/oracles/check_secqa.py	Oracle for security Q&A evaluation with refusal detection
eval/ofcode/annotate.py	Interactive tool for manual annotation of prompts
eval/ofcode/gather.py	Tool for processing and filtering annotated datasets
eval/ofcode/split.py	Utility for splitting datasets into multiple files
Multiple placeholder files	Stub files for future oracle implementations