Autonomous Synthetic Visual Dataset Tool-Calling Matrix

Role: Principal Synthetic Data Architect specializing in autonomous tool-calling pipelines for computer vision model training.

Context

• Target Domain: {{target_model_domain}}
• Edge-Case Taxonomy: {{edge_case_taxonomy}}
• Agent Orchestrator: {{agent_orchestrator}}
• Generation Tool Interface: {{tool_call_spec}}
• Render Engine Target: {{rendering_engine}}
• Acceptance Criteria: {{validation_criteria}}

Task

Design a structured, tool-executable prompt matrix that autonomous agents can invoke dynamically to generate balanced, artifact-free synthetic image datasets across complex edge-case variations.

Method

1. Deconstruct the {{edge_case_taxonomy}} into distinct environmental, lighting, occlusion, and perspective states.
2. Map every edge-case state to specific schema parameters required by {{tool_call_spec}}.
3. Formulate positive visual prompts utilizing the stylistic syntax and keyword weighting optimal for {{rendering_engine}}.
4. Define targeted negative prompt embeddings to suppress hallucinations and non-physical geometry in {{target_model_domain}}.
5. Calibrate diffusion seed strategies, sampler steps, and CFG scale values per scenario category.
6. Specify dynamic variable substitution slots that {{agent_orchestrator}} will populate at runtime.
7. Establish programmatic trigger rules based on {{validation_criteria}} to automate retry logic upon failed generation.

Constraints

• MUST output strict, machine-parsable matrix rows containing complete JSON-ready tool arguments.
• MUST NOT use generic qualitative descriptors; specify explicit spatial, volumetric, and optical parameters.
• Every prompt string MUST contain explicit composition, lighting vector, and camera lens specifications.
• All matrix entries MUST adhere to the argument limits defined in {{tool_call_spec}}.

Output format

1. Tool Call Schema Alignment

Summary of payload keys and parameter mappings (max 150 words).

2. Autonomous Prompt Execution Matrix

A markdown table with the following exact columns: | State ID | Edge-Case Scenario | Tool Payload (function_call) | Positive Prompt Body | Negative Constraint String | Execution Parameters (CFG/Steps/Sampler) | Pass/Fail Evaluation Vector | (Provide exactly 6 comprehensive, production-ready rows).

3. Agentic Runtime Triggers

Numbered list of 3 conditional tool-chain rules for automated fallback when {{validation_criteria}} fails.

Self-review

• Are all 6 matrix rows directly ingestible by {{agent_orchestrator}}?
• Does every positive prompt include concrete camera, lighting, and spatial tokens?
• Are negative prompts tailored to eliminate artifacts specific to {{target_model_domain}}?