agentsop-output-format-by-model

What problem does it solve?

It helps you stop losing model quality by wrapping outputs in the wrong format (e.g., forcing code into JSON), and instead choose an output wire format that matches the content, the consumer, and the target model.

Core Features & Use Cases

• Per-task format selection, not reflex: pick plain text, diff, markdown, JSON, tool-call envelopes, or grammar-constrained formats based on the generated content and how it will be consumed.
• Empirical decision rules: uses benchmark findings (like Aider’s code-in-JSON vs unified-diff differences) and research on format-restricted reasoning to guide tradeoffs.
• Concrete SOP + anti-patterns: provides a step-by-step workflow for deciding format across code edits, tool calls, SQL payloads, extraction schemas, and mixed-content tasks, including what to avoid.

Quick Start

Ask an AI to redesign your agent’s output schema by telling it what the model must produce, who consumes it next (parser, human, compiler, tool dispatcher), and which model you are using.