thinkdeep
Read-only
Analyzes complex problems through multi-stage investigation and reasoning. Supports architecture decisions, debugging, performance, and security analysis with hypothesis testing and expert validation.
Instructions
Performs multi-stage investigation and reasoning for complex problem analysis. Use for architecture decisions, complex bugs, performance challenges, and security analysis. Provides systematic hypothesis testing, evidence-based investigation, and expert validation.
Input Schema
TableJSON Schema
| Name | Required | Description | Default |
|---|---|---|---|
| step | Yes | Current work step content and findings from your overall work | |
| model | Yes | Currently in auto model selection mode. CRITICAL: When the user names a model, you MUST use that exact name unless the server rejects it. If no model is provided, you may use the `listmodels` tool to review options and select an appropriate match. Top models: gemini-2.5-pro (score 100, 1.0M ctx, thinking, code-gen); gemini-3-pro-preview (score 100, 1.0M ctx, thinking, code-gen); gemini-2.5-flash (score 61, 1.0M ctx, thinking); gemini-2.0-flash (score 56, 1.0M ctx, thinking); gemini-2.0-flash-lite (score 42, 1.0M ctx). | |
| images | No | Optional absolute image paths or base64 blobs for visual context. | |
| findings | Yes | Important findings, evidence and insights discovered in this step | |
| confidence | No | Confidence level: exploring (just starting), low (early investigation), medium (some evidence), high (strong evidence), very_high (comprehensive understanding), almost_certain (near complete confidence), certain (100% confidence locally - no external validation needed) | |
| hypothesis | No | Current theory about issue/goal based on work | |
| focus_areas | No | Focus aspects (architecture, performance, security, etc.) | |
| step_number | Yes | Current step number in work sequence (starts at 1) | |
| temperature | No | 0 = deterministic · 1 = creative. | |
| total_steps | Yes | Estimated total steps needed to complete work | |
| issues_found | No | Issues identified with severity levels during work | |
| files_checked | No | List of files examined during this work step | |
| thinking_mode | No | Reasoning depth: minimal, low, medium, high, or max. | |
| relevant_files | No | Files identified as relevant to issue/goal (FULL absolute paths to real files/folders - DO NOT SHORTEN) | |
| continuation_id | No | Unique thread continuation ID for multi-turn conversations. Works across different tools. ALWAYS reuse the last continuation_id you were given—this preserves full conversation context, files, and findings so the agent can resume seamlessly. | |
| problem_context | No | Additional context about problem/goal. Be expressive. | |
| relevant_context | No | Methods/functions identified as involved in the issue | |
| next_step_required | Yes | Whether another work step is needed. When false, aim to reduce total_steps to match step_number to avoid mismatch. | |
| use_assistant_model | No | Use assistant model for expert analysis after workflow steps. False skips expert analysis, relies solely on your personal investigation. Defaults to True for comprehensive validation. |