Introspection & Brainstorming Modes
Relevant source files
The following files were used as context for generating this wiki page:
Introduction
Introspection and Brainstorming (referred to as Creative Analysis in some documentation) are specialized behavioral modes within the SuperGemini Framework designed to enhance the Gemini CLI's cognitive processing capabilities. These modes operate through behavioral instruction injection, modifying the AI's persona and logic flow to handle either meta-cognitive reasoning or high-level requirements discovery. While Introspection focuses on transparent reasoning and error recovery, Brainstorming/Creative Analysis is tailored for transforming vague project ideas into structured implementation plans.
These modes are part of the framework's broader orchestration system, which coordinates 13 specialized agents and various Model Context Protocol (MCP) servers. They can be triggered automatically based on task complexity or manually via command-line flags to optimize development workflows.
Sources: Docs/User-Guide/modes.md:5-25, CLAUDE.md:12-20, Docs/Getting-Started/quick-start.md:30-45
Creative Analysis (Brainstorming) Mode
Creative Analysis Mode is the framework's primary mechanism for requirements exploration. It serves as a socratic interface that prevents the AI from rushing into immediate implementation when presented with underspecified or vague requests.
Purpose and Functionality
The core purpose of this mode is to analyze project requirements and provide structured technical recommendations before a single line of code is written. It converts abstract thoughts into actionable development plans by identifying constraints, business goals, and architectural options.
Sources: Docs/User-Guide/modes.md:40-55, Docs/Reference/basic-examples.md:110-125
Trigger Mechanisms
Brainstorming behavior is activated through several pathways:
- Keywords: Using terms like "analyze", "explore", "discuss", or "figure out".
- Uncertainty Indicators: Natural language markers such as "maybe", "possibly", or "could we".
- Manual Flags: The
/sg:analyzecommand or the--analyzeand--exploreflags. - Input Nature: Vague project requests like "I want to build..." automatically trigger this mode.
Sources: Docs/User-Guide/modes.md:44-52, Docs/Reference/quick-start-practices.md:15-25
Operational Flow
The following diagram illustrates the logical progression when Brainstorming/Creative Analysis is activated:
The flow ensures that vague requests result in structured requirements or technical briefs rather than immediate code generation.
Sources: Docs/User-Guide/modes.md:57-80, Docs/Getting-Started/quick-start.md:65-80
Introspection Mode
Introspection Mode provides a meta-cognitive layer to the AI's operations. It is designed for transparency, allowing the user (and the framework itself) to see the underlying reasoning process used to reach a conclusion.
Logic and Reasoning Transparency
When active, the framework exposes its thinking process using specific visual markers. This is critical for debugging complex logic or understanding why the AI chose a specific architectural pattern over another.
| Marker | Meaning |
|---|---|
| 🤔 | Meta-Analysis / Reasoning |
| 🎯 | Decision Logic / Goal Alignment |
| ⚡ | Correction / Action |
| 📊 | Evidence Check |
| 💡 | Learning / Insight |
Sources: Docs/User-Guide/modes.md:85-110, Docs/Reference/README.md:88-95
Meta-Learning Architecture
The architecture of Introspection allows the system to validate actions against SuperGemini principles and quality gates. It evaluates choice logic and considers alternative approaches in real-time, which is particularly useful during error recovery scenarios or unexpected results.
This sequence ensures that every decision is filtered through a transparency layer before being presented to the user.
Sources: Docs/User-Guide/modes.md:112-130, CLAUDE.md:120-135
Technical Implementation and Integration
Both modes are implemented via the instruction injection system. Configuration files stored in the project's source are converted to TOML and installed to the user's local environment.
Command and Flag Reference
| Element | Type | Purpose |
|---|---|---|
/sg:analyze | Command | Primary entry for Brainstorming/Requirements discovery. |
/sg:brainstorm | Command | Collaborative requirement discovery (legacy/alias). |
--introspect | Flag | Manual trigger for Introspection markers and reasoning. |
--analyze | Flag | Manual trigger for Creative Analysis logic. |
--ultrathink | Flag | Advanced analysis enhancement (Advanced). |
Sources: Docs/User-Guide/modes.md:310-330, Docs/Reference/README.md:90-95, CLAUDE.md:85-100
MCP Server Coordination
During Introspection or Brainstorming, the framework often coordinates with specific Model Context Protocol (MCP) servers:
- Sequential: Used for complex multi-step reasoning and expert panel coordination.
- Context7: Activated to pull official documentation and framework-specific patterns for validated analysis.
- Serena: Employed for semantic code understanding and maintaining project memory during deep analysis.
Sources: SuperGemini/Commands/spec-panel.toml:135-150, Docs/Getting-Started/quick-start.md:40-50
Conclusion
Introspection and Brainstorming Modes represent the framework's commitment to "thinking before doing." By providing a structured pathway for requirements exploration (Brainstorming) and a transparent window into AI decision-making (Introspection), SuperGemini ensures that complex development tasks are handled with architectural rigor. These modes are not merely cosmetic; they are integrated into the core orchestration layer to improve information quality and prevent the propagation of errors in the development lifecycle.
Sources: Docs/User-Guide/modes.md:450-470, Docs/Getting-Started/quick-start.md:110-125