name: ai-code-review-env
version: "1.0.0"
description: >
An RL environment where an AI agent performs structured code review.
The agent reads buggy JavaScript code, identifies the bug category,
and proposes a fix. Three tasks of progressive difficulty: syntax,
logic, and performance. Rewards are shaped continuously across both
the identify and fix phases of each task.

author: yaxh64
tags:

* openenv
* code-review
* javascript
* reinforcement-learning
* hackathon

inference: inference.py

interface:
reset: reset
step: step
state: state

action:
type: object
description: >
An agent action consists of a phase ("identify" or "fix") and
free-text content describing the finding or the corrected code.
properties:
action_type:
type: string
enum: [identify, fix]
content:
type: string

observation:
type: object
description: >
The agent observes the buggy code, a task description including
the current phase instruction, and the history of prior steps.
properties:
code:
type: string
task:
type: string
history:
type: array
items:
type: string
task_id:
type: integer
language:
type: string
difficulty:
type: string
enum: [easy, medium, hard]
category:
type: string
enum: [syntax, logic, performance]

reward:
type: float
range: [0.0, 1.0]

tasks:

* id: 1
  name: missing-parenthesis
  difficulty: easy
  category: syntax

* id: 2
  name: wrong-even-check
  difficulty: medium
  category: logic

* id: 3
  name: inefficient-array-loop
  difficulty: hard
  category: performance