Home > Tasks > Social > MultiPerspectiveAnalysisTask

MultiPerspectiveAnalysisTask

Analyze complex topics from multiple independent viewpoints and synthesize findings into a unified, high-confidence conclusion using parallel agent execution.

Side-Effect Safe Multi-Agent Synthesis GPT-4o Recommended
⚙️ MultiPerspectiveAnalysisConfig.json
{
  "analysis_subject": "Migrating to Microservices",
  "perspectives": [
    "Technical Debt",
    "Operational Complexity",
    "Developer Velocity"
  ],
  "input_files": ["src/**/*.kt"],
  "synthesize": true,
  "consensus_threshold": 0.8
}
→
👁️ Session UI (TabbedDisplay)
Overview
Technical Debt
Synthesis

Unified Conclusion

The analysis shows a 0.85 consensus that migration is viable if operational tooling is prioritized.

  • Agreement: High risk in deployment overhead.
  • Conflict: Developer velocity may dip initially.

Live Results Showcase

Explore actual analysis artifacts generated by this task in the test workspace.

Execution Configuration

Field Type Description
analysis_subject* String The topic or problem to analyze from multiple viewpoints.
perspectives* List<String> List of viewpoints (e.g., technical, business, ethical, user).
input_files List<String> Glob patterns (e.g. **/*.kt) for source context.
synthesize Boolean If true, generates a unified conclusion. Default: true.
consensus_threshold Double Minimum agreement level (0.0-1.0). Default: 0.7.
related_files List<String> Additional files for context (supports glob patterns).

Task Lifecycle

  1. Initialization: Validates that the subject and perspectives are provided and the threshold is within bounds (0.0-1.0).
  2. Context Gathering: Resolves glob patterns in input_files and related_files to provide the LLM with relevant source code or documentation.
  3. Parallel Analysis: Spawns independent ChatAgent instances for each perspective. This prevents "groupthink" or cross-contamination of ideas between viewpoints.
  4. Synthesis: A specialized synthesis agent reviews all perspective results to identify common themes, conflicts, and overall consensus.
  5. UI Rendering: Results are streamed into a TabbedDisplay, allowing users to toggle between individual analyses and the final synthesis.

Embedded Execution (Scenario B)

Invoke this task directly using the UnifiedHarness in headless mode:

import com.simiacryptus.cognotik.plan.TaskType
import com.simiacryptus.cognotik.plan.tools.social.MultiPerspectiveAnalysisTask.Companion.MultiPerspectiveAnalysis
import com.simiacryptus.cognotik.plan.tools.social.MultiPerspectiveAnalysisTask.MultiPerspectiveAnalysisTaskExecutionConfigData
import com.simiacryptus.cognotik.plan.tools.TaskTypeConfig

// 1. Define Runtime Input
val executionConfig = MultiPerspectiveAnalysisTaskExecutionConfigData(
    analysis_subject = "Migrating to Microservices",
    perspectives = listOf("Technical Debt", "Operational Complexity"),
    input_files = listOf("src/**/*.kt"),
    synthesize = true,
    consensus_threshold = 0.8
)

// 2. Run via Harness
harness.runTask(
    taskType = MultiPerspectiveAnalysis,
    typeConfig = TaskTypeConfig(), // Use default static config
    executionConfig = executionConfig,
    workspace = File("./my-project"),
    autoFix = true
)

Kotlin Task Instantiation

val task = MultiPerspectiveAnalysisTask(
    orchestrationConfig = config,
    planTask = MultiPerspectiveAnalysisTaskExecutionConfigData(
        analysis_subject = "Adopting Rust for performance modules",
        perspectives = listOf("Security", "Performance", "Hiring"),
        synthesize = true,
        consensus_threshold = 0.8
    )
)

Prompt Segment

The following instructions are injected into the orchestrator to guide the LLM:

MultiPerspectiveAnalysis - Analyze problems from multiple viewpoints with synthesis
** Specify the subject to analyze in analysis_subject
** Provide a list of perspectives to consider (e.g., technical, business, ethical, user experience)
** Optionally, list input files (supports glob patterns) to provide context for the analysis
** Set synthesize=true to generate a unified conclusion from all perspectives
** Configure consensus_threshold (0.0-1.0) to determine minimum agreement level
** Additional context files can be specified via input_files
** Each perspective will be analyzed independently, then synthesized
** Useful for:
   - Architectural decision making
   - Code review from multiple angles
   - Strategic planning
   - Risk assessment
   - Feature evaluation