code-workout/model_capabilities.yaml at master · web-cat/code-workout · 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
# Model Capabilities Registry
#
# This file is OPTIONAL. GSD works without it.
# If present, it provides guidance for model selection.
#
# Usage: Reference this file when choosing models by capability.
# Do NOT make any workflow depend on this file existing.

# Capability definitions
capabilities:
  thinking_mode:
    description: "Extended reasoning with internal monologue"
    benefits:
      - Better for complex planning
      - Improved debugging hypotheses
      - More thorough architecture analysis

  long_context:
    description: "Context window > 100k tokens"
    benefits:
      - Can analyze multiple files at once
      - Better for large refactors
      - Full codebase reviews

  tools:
    description: "Function/tool calling support"
    benefits:
      - Execute verification commands
      - Interact with external systems
      - Structured outputs

  speed_tier:
    description: "Response latency tier"
    values:
      - fast: "< 2s average, good for iteration"
      - standard: "2-5s average, balanced"
      - reasoning: "> 5s average, deep analysis"

# Example model profiles (illustrative, not exhaustive)
# These show how to categorize models by capability type.

profiles:
  # Fast iteration models
  fast_coder:
    speed_tier: fast
    thinking_mode: false
    long_context: false
    tools: true
    best_for:
      - Quick edits
      - Simple implementations
      - Frequent iteration cycles
    examples:
      - "Flash/Turbo variants"
      - "Smaller parameter models"

  # Standard balanced models
  standard:
    speed_tier: standard
    thinking_mode: false
    long_context: true
    tools: true
    best_for:
      - Most development tasks
      - Moderate refactoring
      - General coding
    examples:
      - "Pro/Standard variants"
      - "Mid-tier models"

  # Deep reasoning models
  reasoning:
    speed_tier: reasoning
    thinking_mode: true
    long_context: true
    tools: true
    best_for:
      - Architecture planning
      - Complex debugging
      - Security analysis
    examples:
      - "Thinking/Reasoning variants"
      - "Advanced reasoning models"

# Phase recommendations (suggestions, not requirements)
phase_recommendations:
  planning:
    preferred_profile: reasoning
    reason: "Complex decisions benefit from extended thinking"

  implementation:
    preferred_profile: fast_coder
    reason: "Frequent iteration needs speed"

  refactoring:
    preferred_profile: standard
    reason: "Balance of context and speed"

  debugging:
    preferred_profile: reasoning
    reason: "Hypothesis generation needs depth"

  review:
    preferred_profile: standard
    reason: "Large context for full diffs"

# IMPORTANT: This file is for guidance only.
# Never fail a workflow because a specific model capability is missing.