Class: Ace::Test::EndToEndRunner::Atoms::PromptBuilder

Inherits:
Object
  • Object
show all
Defined in:
lib/ace/test/end_to_end_runner/atoms/prompt_builder.rb

Overview

Builds LLM prompts for E2E test execution

Creates a system prompt that instructs the LLM to execute a test scenario and return structured JSON results, along with the user prompt containing the test scenario content.

Constant Summary collapse

TC_SYSTEM_PROMPT =

System prompt for TC-level (single test case) execution

<<~PROMPT
  You are an E2E test executor for the ACE (Agentic Coding Environment) toolkit.

  Your task is to execute a single test case in a pre-populated sandbox and return structured results.

  ## Instructions

  1. The test sandbox is pre-populated at the path provided — do NOT create or modify the sandbox setup
  2. Read the test case steps carefully
  3. Execute the test case steps in the sandbox
  4. Record pass/fail status
  5. Return results as JSON

  ## Output Format

  You MUST return a JSON block wrapped in ```json fences with these fields:

  ```json
  {
    "test_id": "TS-XXX-NNN",
    "tc_id": "TC-NNN",
    "status": "pass|fail",
    "actual": "What actually happened",
    "notes": "Any additional observations",
    "summary": "Brief result"
  }
  ```

  ## Rules

  - Execute ONLY the single test case provided
  - Execute in the pre-populated sandbox (do not modify setup files)
  - Record actual output/behavior, not just expected
  - If the test case cannot be executed (missing tool, permission error), mark as "fail" with explanation
PROMPT
SYSTEM_PROMPT = 
<<~PROMPT
  You are an E2E test executor for the ACE (Agentic Coding Environment) toolkit.

  Your task is to execute the provided test scenario step by step and return structured results.

  ## Instructions

  1. Read the test scenario carefully
  2. Execute the Environment Setup commands
  3. Create any Test Data as specified
  4. Execute each Test Case (TC-NNN) in order
  5. Record pass/fail status for each test case
  6. Return results as JSON

  ## Output Format

  You MUST return a JSON block wrapped in ```json fences with these fields:

  ```json
  {
    "test_id": "TS-XXX-NNN",
    "status": "pass|fail|partial",
    "test_cases": [
      {
        "id": "TC-001",
        "description": "Brief description",
        "status": "pass|fail",
        "actual": "What actually happened",
        "notes": "Any additional observations"
      }
    ],
    "summary": "Brief execution summary",
    "observations": "Any friction points or issues discovered"
  }
  ```

  ## Rules

  - Execute ALL test cases, even if earlier ones fail
  - Record actual output/behavior, not just expected
  - Use "partial" status if some test cases pass and some fail
  - Include meaningful observations about tool behavior
  - If a test case cannot be executed (missing tool, permission error), mark as "fail" with explanation
PROMPT

Instance Method Summary collapse

Instance Method Details

#build(scenario, test_cases: nil) ⇒ String

Build the user prompt for a test scenario

Parameters:

  • scenario (Models::TestScenario) —

    The test scenario to execute

  • test_cases (Array<String>, nil) (defaults to: nil) —

    Optional test case IDs to filter

Returns:

  • (String) —

    The user prompt containing the test scenario



150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
 
# File 'lib/ace/test/end_to_end_runner/atoms/prompt_builder.rb', line 150

def build(scenario, test_cases: nil)
  filter_instruction = if test_cases&.any?
    "\n**IMPORTANT:** Execute ONLY the following test cases: #{test_cases.join(", ")}. Skip all other test cases.\n"
  else
    ""
  end

  pending_instruction = build_pending_tc_skip_instruction(scenario)

  execute_instruction = if test_cases&.any?
    "Execute only the specified test cases (#{test_cases.join(", ")}) and return the JSON results as specified in your instructions."
  else
    "Execute all test cases in this scenario and return the JSON results as specified in your instructions."
  end

  <<~PROMPT
    # Execute E2E Test: #{scenario.test_id}

    **Package:** #{scenario.package}
    **Title:** #{scenario.title}
    **Priority:** #{scenario.priority}
    #{filter_instruction}#{pending_instruction}
    ## Test Scenario

    #{scenario.content}

    ---

    #{execute_instruction}
  PROMPT
end

#build_tc(test_case:, scenario:, sandbox_path:) ⇒ String

Build a TC-level user prompt for a single test case

Parameters:

  • test_case (Models::TestCase) —

    The single test case to execute

  • scenario (Models::TestScenario) —

    The parent scenario for metadata

  • sandbox_path (String) —

    Path to the pre-populated sandbox

Returns:

  • (String) —

    The TC-level user prompt



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
 
# File 'lib/ace/test/end_to_end_runner/atoms/prompt_builder.rb', line 101

def build_tc(test_case:, scenario:, sandbox_path:)
  if test_case.pending?
    return <<~PROMPT
      # SKIP Test Case: #{scenario.test_id} / #{test_case.tc_id}

      **Package:** #{scenario.package}
      **Scenario:** #{scenario.title}
      **Test Case:** #{test_case.title}
      **Status:** PENDING — #{test_case.pending}

      This test case is marked as pending and should NOT be executed.
      Return the following JSON result:

      ```json
      {
        "test_id": "#{scenario.test_id}",
        "tc_id": "#{test_case.tc_id}",
        "status": "skip",
        "actual": "Skipped — pending",
        "notes": "#{test_case.pending}",
        "summary": "Pending: #{test_case.pending}"
      }
      ```
    PROMPT
  end

  <<~PROMPT
    # Execute Test Case: #{scenario.test_id} / #{test_case.tc_id}

    **Package:** #{scenario.package}
    **Scenario:** #{scenario.title}
    **Test Case:** #{test_case.title}
    **Sandbox Path:** #{sandbox_path}

    ## Test Case Content

    #{test_case.content}

    ---

    Execute the test case steps in the sandbox at `#{sandbox_path}` and return JSON results as specified in your instructions.
  PROMPT
end