Home > Tasks > File > DiscussionTask

DiscussionTask

Directly answer questions or provide insights using the LLM, optionally referencing project files, with support for interactive user feedback.

Side-Effect Safe Interactive Support Markdown Output
⚙️ DiscussionTaskExecutionConfig.json 
{
  "inquiry_questions": [
    "How is the authentication middleware structured?",
    "Are there any potential race conditions in the cache?"
  ],
  "inquiry_goal": "Security and Performance Audit",
  "input_files": ["src/auth/**/*.kt", "src/cache/*.kt"]
}
→
👁️ Session UI (Markdown Render)

Analysis: Security & Performance

Based on the provided files in src/auth and src/cache:

  • Authentication: Uses a JWT-based filter chain. Structure is modular but lacks token revocation logic.
  • Cache: LocalCache.kt uses ConcurrentHashMap, which prevents basic race conditions, but the getOrPut logic is not atomic.

User: "Can you suggest a fix for the cache atomicity?"

Test Workspace Browser

Explore the artifacts and logs generated during automated testing of the DiscussionTask.

Execution Configuration

Field Type Description
inquiry_questions List<String> The specific questions or topics to be addressed in the inquiry.
inquiry_goal String The high-level goal or purpose of the inquiry (e.g., "Code Review").
input_files List<String> Glob patterns (e.g. **/*.kt) to be used as context for the LLM.
task_description String A natural language description of the task's intent.
Token Usage: Medium to High (depends on the number and size of input files).

Task Lifecycle

  1. Context Gathering: The task resolves input_files glob patterns against the workspace.
  2. File Ingestion: Files are read and formatted into a single context block. Non-text files (PDFs, etc.) are processed via specialized readers.
  3. LLM Interaction:
    • If autoFix is enabled: Performs a one-shot analysis and returns the result.
    • If autoFix is disabled: Initiates an interactive Discussable session allowing the user to ask follow-up questions.
  4. Output: Generates a comprehensive Markdown report rendered directly in the Session UI.

Type Configuration

Field Type Description
model ApiChatModel Override the default model used for the "Insight" agent.

Embedded Execution (UnifiedHarness)

To run this task programmatically in a headless environment (CI/CD, CLI), use the UnifiedHarness. 

// 1. Define the Execution Configuration
val executionConfig = DiscussionTask.DiscussionTaskExecutionConfigData(
    inquiry_goal = "Architecture Review",
    inquiry_questions = listOf("Does the service layer follow the Repository pattern?"),
    input_files = listOf("src/main/kotlin/services/*.kt", "src/main/kotlin/repositories/*.kt"),
    task_description = "Reviewing architectural patterns"
)

// 2. Define the Type Configuration (Optional model override)
val typeConfig = DiscussionTask.DiscussionTaskTypeConfig(
    name = "ArchReviewer"
)

// 3. Run via Harness
harness.runTask(
    taskType = DiscussionTask.Discussion,
    typeConfig = typeConfig,
    executionConfig = executionConfig,
    workspace = File("./my-project"),
    autoFix = true // Set to true for non-interactive CI/CD usage
)

Test Case Example

Example of a unit test verifying the discussion logic:

@Test
fun testDiscussionTask() {
    val harness = UnifiedHarness(serverless = true)
    harness.start()
    val result = harness.runTask(
        taskType = DiscussionTask.Discussion,
        executionConfig = DiscussionTask.DiscussionTaskExecutionConfigData(
            inquiry_goal = "Test Inquiry",
            inquiry_questions = listOf("What is the purpose of this test?"),
            input_files = listOf("src/test/kotlin/DiscussionTaskTest.kt")
        ),
        workspace = tempDir,
        autoFix = true
    )
    assert(result.contains("DiscussionTaskTest"))
}

Prompt Segment

The following logic is injected into the orchestrator's system prompt:

Discussion - Directly answer questions or provide insights using the LLM. Reading files is optional and can be included if relevant to the inquiry.
* Specify the questions and the goal of the inquiry.
* Optionally, list input files (supports glob patterns) to be examined.
* User response/feedback and iteration are supported.