feat: add code-review environment

[vominh1919]

Summary

Adds a new code-review environment for evaluating LLM's ability to review code.

Features

• Bug detection: Identifies potential bugs in code
• Security analysis: Flags security vulnerabilities
• Code quality: Suggests improvements and best practices
• Multi-language: Supports Python, JavaScript, and more

Bounties Program

This environment aligns with Prime Intellect's Environments Program:

• Category: Open Access ($100-500)
• Type: Self-contained benchmark implementation
• Tags: code-review, coding, analysis

Usage

prime env install code-review
prime eval run code-review -m openai/gpt-4.1-mini

Closes #

---

Note

Low Risk
Additive, self-contained environment modules with simple scoring and no changes to shared infrastructure, auth, or data-handling paths.

Overview
Adds three new installable evaluation environments under environments/: api-design, code-review, and sql-query.

Each environment defines a load_environment() that builds small synthetic train/eval HF Datasets, configures a simple keyword-based scoring Rubric, and returns a vf.SingleTurnEnv with an appropriate system prompt. New pyproject.toml files wire packaging metadata, dependencies (verifiers>=0.1.8), and default eval settings; code-review also includes a README with usage instructions.

^{Reviewed by Cursor Bugbot for commit 439fa61. Bugbot is set up for automated code reviews on this repo. Configure here.}

[vominh1919]

Update

Added 2 more environments to this PR:

1. code-review: Code review and bug detection
2. api-design: RESTful API design evaluation
3. sql-query: SQL query writing evaluation

All environments follow Prime Intellect's Environments Program guidelines for Open Access bounties ($100-500 each).

Total: 3 environments ready for review!

[cursor]

Cursor Bugbot has reviewed your changes and found 2 potential issues.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit 439fa61. Configure here.}

[cursor]

Dataset uses prompt string, causing assertion crash

High Severity

All three new environments use "prompt" as the dataset column key with a plain string value, but also pass a system_prompt to SingleTurnEnv. The framework's _ensure_prompt method, when it finds a prompt column already present and system_prompt is set, asserts that prompt must be a list of messages — causing an AssertionError at initialization. The column key needs to be "question" instead of "prompt" so the framework properly wraps the string into a messages list. All existing environments that use string-valued prompts use "question" for this reason.

Additional Locations (2)

• environments/api-design/api-design.py#L15-L35
• environments/sql-query/sql-query.py#L15-L35

 

^{Reviewed by Cursor Bugbot for commit 439fa61. Configure here.}

[cursor]

New environments missing from environments/README.md

Low Severity

Three new environments (code_review, api-design, sql-query) are added to the environments/ folder but environments/README.md is not updated to list them. The project rules require that any PR adding or removing an environment must update environments/README.md to reflect the change under the appropriate category/pattern section.

Additional Locations (2)

• environments/api-design/api-design.py#L1-L63
• environments/sql-query/sql-query.py#L1-L62

 

^{Triggered by project rule: BugBot Instructions}

^{Reviewed by Cursor Bugbot for commit 439fa61. Configure here.}