CodeFixer/openenv.yaml at main · Swaroop-Haridas/CodeFixer · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
# openenv.yaml — Required metadata for OpenEnv spec compliance
# Validated by: openenv validate

name: CodeFixerEnv
version: "1.0.0"
description: >
  A reinforcement learning environment where an AI agent receives
  buggy Python functions and must submit corrected code to earn rewards.
  Covers three real-world debugging scenarios: off-by-one loop errors,
  multi-bug list processing, and algorithmic complexity refactoring.

author: "Hackathon Submission"
tags:
  - openenv
  - code-debugging
  - python
  - software-engineering

# ── Interface ──────────────────────────────────────────────────────────────
api:
  reset_endpoint:  "/reset"
  step_endpoint:   "/step"
  state_endpoint:  "/state"
  health_endpoint: "/health"

# ── Observation space ──────────────────────────────────────────────────────
observation_space:
  type: object
  fields:
    input:
      type: string
      description: "Buggy Python code the agent must fix"
    context:
      type: string
      description: "Plain-English description of what the correct code should do"
    history:
      type: array
      description: "Ordered list of previous (action, feedback) dicts this episode"
    step_number:
      type: integer
      description: "Number of steps elapsed since reset"
    max_steps:
      type: integer
      description: "Episode length cap (default 5)"

# ── Action space ───────────────────────────────────────────────────────────
action_space:
  type: object
  fields:
    type:
      type: string
      enum: ["fix", "explain", "give_up"]
      description: "fix=submit corrected code, explain=request hint, give_up=end episode"
    content:
      type: string
      description: "Full corrected Python function (fix), or empty string (explain/give_up)"

# ── Reward space ───────────────────────────────────────────────────────────
reward_space:
  type: object
  fields:
    value:
      type: float
      range: [-0.10, 1.20]
      description: "Net scalar reward for this step"
    grader_score:
      type: float
      range: [0.0, 1.0]
      nullable: true
      description: "Raw grader output (only on fix actions)"
    improvement:
      type: float
      nullable: true
      description: "Score delta vs. previous best"
    efficiency_bonus:
      type: float
      description: "+0.20 if solved in ≤ 2 steps, else 0.0"
    penalty:
      type: float
      description: "Negative component from redundancy / explain / give_up"
    reason:
      type: string
      description: "Human-readable reward explanation"

# ── Tasks ──────────────────────────────────────────────────────────────────
tasks:
  - id: easy_sum_range
    difficulty: easy
    description: >
      Fix an off-by-one error in sum_range(n).
      Agent must change range(n) to range(1, n+1).
    grader_weights:
      syntax: 0.20
      correctness: 0.70
      code_quality: 0.10

  - id: medium_flatten_list
    difficulty: medium
    description: >
      Fix two bugs in flatten(lst): incorrect type check and
      missing inner-list recursion.
    grader_weights:
      syntax: 0.15
      correctness: 0.75
      isinstance_usage: 0.10

  - id: hard_longest_unique_substr
    difficulty: hard
    description: >
      Refactor longest_unique(s) from O(n²) nested loops to
      an O(n) sliding window using a hashmap.
    grader_weights:
      syntax: 0.10
      correctness: 0.70
      linear_complexity: 0.20

# ── Episode config ─────────────────────────────────────────────────────────
episode:
  max_steps: 5
  reward_range: [-0.10, 1.20]
  termination_conditions:
    - "grader_score >= 1.0 on a fix action"
    - "action.type == give_up"
    - "step_number >= max_steps"

# ── Infrastructure ─────────────────────────────────────────────────────────
runtime:
  port: 7860
  framework: fastapi
  python_version: "3.11"