agentsop-dspy

What problem does it solve?

It turns brittle, hand-written prompt pipelines into a reusable decision system that can be optimized with evaluation metrics instead of guesswork.

Core Features & Use Cases

• When to activate: Detects DSPy-relevant intent (e.g., “use DSPy”, “compile prompts”, “MIPRO/GEPA”) and explicitly avoids one-shot or audit-forced cases.
• SOP for DSPy compilation: Guides coder-agents through Signature → Module → Metric → Compile using DSPy’s Stage 1/2/3 gate with clear exit criteria.
• Dilemma-based decision rules: Covers optimizer cost vs gain, LM swapping and recompilation, metric bias risks, and handling stuck compile trials.
• Evidence-driven guidance: Uses documented DSPy/GEPA practices and warns against common anti-patterns (no metric, too few examples, wrong optimizer stage).

Quick Start

Ask an AI coder-agent to “apply the agentsop-dspy skill to turn my current prompt pipeline into a DSPy program by defining a Signature, selecting an appropriate Module, building a trustworthy metric, and running a Stage 3 compile with auto='light' first.”