Class: Aidp::Execute::PromptEvaluator

Inherits:
Object
  • Object
show all
Defined in:
lib/aidp/execute/prompt_evaluator.rb

Overview

Evaluates prompt effectiveness using ZFC after multiple iterations

FIX for issue #391: When the work loop reaches 10+ iterations without completion, this evaluator assesses prompt quality and suggests improvements.

Uses Zero Framework Cognition (ZFC) to analyze:

  • Whether the prompt clearly defines completion criteria

  • If task breakdown instructions are adequate

  • Whether the agent has sufficient context

  • If there are blockers preventing progress

Examples:

evaluator = PromptEvaluator.new(config)
result = evaluator.evaluate(
  prompt_content: prompt_manager.read,
  iteration_count: 12,
  task_summary: persistent_tasklist.summary,
  recent_failures: all_results
)
# => { effective: false, issues: [...], suggestions: [...] }

Constant Summary collapse

EVALUATION_ITERATION_THRESHOLD =

Threshold for triggering evaluation

10
EVALUATION_INTERVAL =

Re-evaluate periodically after threshold

5

Instance Attribute Summary collapse

Instance Method Summary collapse

Constructor Details

#initialize(config, ai_decision_engine: nil) ⇒ PromptEvaluator 



38
39
40
41
 
# File 'lib/aidp/execute/prompt_evaluator.rb', line 38

def initialize(config, ai_decision_engine: nil)
  @config = config
  @ai_decision_engine = ai_decision_engine || safely_build_ai_decision_engine
end

Instance Attribute Details

#ai_decision_engine ⇒ Object (readonly)

Expose for testability



36
37
38
 
# File 'lib/aidp/execute/prompt_evaluator.rb', line 36

def ai_decision_engine
  @ai_decision_engine
end

Instance Method Details

#evaluate(prompt_content:, iteration_count:, task_summary:, recent_failures:, step_name: nil) ⇒ Hash

Evaluate prompt effectiveness



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
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
 
# File 'lib/aidp/execute/prompt_evaluator.rb', line 73

def evaluate(prompt_content:, iteration_count:, task_summary:, recent_failures:, step_name: nil)
  Aidp.log_debug("prompt_evaluator", "starting_evaluation",
    iteration: iteration_count,
    step: step_name,
    prompt_size: prompt_content&.length || 0)

  # When AI decision engine is unavailable (e.g., in tests with mock configs),
  # return a neutral result that doesn't trigger feedback appending
  unless @ai_decision_engine
    Aidp.log_debug("prompt_evaluator", "skipping_evaluation_no_ai_engine")
    return {
      effective: true,  # Assume effective to avoid unnecessary feedback
      issues: [],
      suggestions: [],
      likely_blockers: [],
      recommended_actions: [],
      confidence: 0.0,
      skipped: true,
      skip_reason: "AI decision engine not available"
    }
  end

  prompt = build_evaluation_prompt(
    prompt_content: prompt_content,
    iteration_count: iteration_count,
    task_summary: task_summary,
    recent_failures: recent_failures
  )

  schema = {
    type: "object",
    properties: {
      effective: {
        type: "boolean",
        description: "True if the prompt is likely to lead to completion within a few more iterations"
      },
      issues: {
        type: "array",
        items: {type: "string"},
        description: "Specific problems identified with the current prompt"
      },
      suggestions: {
        type: "array",
        items: {type: "string"},
        description: "Actionable suggestions to improve prompt effectiveness"
      },
      likely_blockers: {
        type: "array",
        items: {type: "string"},
        description: "Potential blockers preventing progress"
      },
      recommended_actions: {
        type: "array",
        items: {
          type: "object",
          properties: {
            action: {type: "string"},
            priority: {type: "string", enum: ["high", "medium", "low"]},
            rationale: {type: "string"}
          }
        },
        description: "Specific actions to take, prioritized"
      },
      confidence: {
        type: "number",
        minimum: 0.0,
        maximum: 1.0,
        description: "Confidence in this assessment"
      }
    },
    required: ["effective", "issues", "suggestions", "confidence"]
  }

  begin
    result = @ai_decision_engine.decide(
      :prompt_evaluation,
      context: {prompt: prompt},
      schema: schema,
      tier: :mini,
      cache_ttl: nil  # Each evaluation is context-specific
    )

    Aidp.log_info("prompt_evaluator", "evaluation_complete",
      iteration: iteration_count,
      effective: result[:effective],
      issue_count: result[:issues]&.size || 0,
      confidence: result[:confidence])

    result
  rescue => e
    Aidp.log_error("prompt_evaluator", "evaluation_failed",
      error: e.message,
      error_class: e.class.name)

    build_fallback_result("Evaluation failed: #{e.message}")
  end
end

#generate_template_improvements(evaluation_result:, original_template:) ⇒ Hash

Generate improvement recommendations for the prompt template Used for AGD pattern - generating improved templates based on evaluation



176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
 
# File 'lib/aidp/execute/prompt_evaluator.rb', line 176

def generate_template_improvements(evaluation_result:, original_template:)
  return nil unless @ai_decision_engine

  Aidp.log_debug("prompt_evaluator", "generating_template_improvements",
    issue_count: evaluation_result[:issues]&.size || 0)

  prompt = build_improvement_prompt(evaluation_result, original_template)

  schema = {
    type: "object",
    properties: {
      improved_sections: {
        type: "array",
        items: {
          type: "object",
          properties: {
            section_name: {type: "string"},
            original: {type: "string"},
            improved: {type: "string"},
            rationale: {type: "string"}
          }
        }
      },
      additional_sections: {
        type: "array",
        items: {
          type: "object",
          properties: {
            section_name: {type: "string"},
            content: {type: "string"},
            rationale: {type: "string"}
          }
        }
      },
      completion_criteria_improvements: {
        type: "array",
        items: {type: "string"},
        description: "Specific improvements to completion criteria definitions"
      }
    },
    required: ["improved_sections", "completion_criteria_improvements"]
  }

  @ai_decision_engine.decide(
    :template_improvement,
    context: {prompt: prompt},
    schema: schema,
    tier: :standard,  # Use standard tier for more thoughtful improvements
    cache_ttl: nil
  )
rescue => e
  Aidp.log_error("prompt_evaluator", "template_improvement_failed",
    error: e.message)
  nil
end

#safely_build_ai_decision_engine ⇒ Object

Safely build AIDecisionEngine, returning nil if config doesn’t support it This allows tests with mock configs to work without AI calls



45
46
47
48
49
50
51
52
53
54
 
# File 'lib/aidp/execute/prompt_evaluator.rb', line 45

def safely_build_ai_decision_engine
  # Check if config supports the methods AIDecisionEngine needs
  return nil unless @config.respond_to?(:default_provider)

  build_default_ai_decision_engine
rescue => e
  Aidp.log_debug("prompt_evaluator", "skipping_ai_decision_engine",
    reason: e.message)
  nil
end

#should_evaluate?(iteration_count) ⇒ Boolean

Check if evaluation should be triggered based on iteration count



59
60
61
62
63
64
 
# File 'lib/aidp/execute/prompt_evaluator.rb', line 59

def should_evaluate?(iteration_count)
  return false unless iteration_count >= EVALUATION_ITERATION_THRESHOLD

  # Evaluate at threshold and every EVALUATION_INTERVAL after
  (iteration_count - EVALUATION_ITERATION_THRESHOLD) % EVALUATION_INTERVAL == 0
end