Dspy Advanced Workflow

This is the DSPy orchestration skill you reach for when you're building a real pipeline from scratch, not just prototyping one predictor. It walks you through the seven-step loop: spec out your task, write the typed signature and module, split train and validation data, write a rich metric that returns feedback (not just a score), baseline it, run GEPA optimization with a reflection LM, then export the artifact. The template is opinionated about details that matter: valset should be separate and representative, the metric must return a dspy.Prediction object or aggregation breaks, and you should always baseline before optimizing so you have a claim to make. If you're saying "end to end" or "from scratch," this is the entry point.

GitHub stars: 250
last updated: May 25, 2026

View on GitHub

Install

$ npx -y skills add intertwine/dspy-agent-skills --skill dspy-advanced-workflow --agent claude-code

Installs into .claude/skills of the current project.

Files

DSPy Advanced Workflow (2026)

This skill runs the seven-step loop that turns a natural-language task description into an optimized, saved, deployable DSPy program. Use the relevant steps in order. Stop at a validated baseline for a prototype; optimizer runs require an appropriate authorized budget and evidence of need. Exporting a local artifact does not authorize deployment.

The seven steps

1. Spec

Rephrase the user's task in one sentence. Identify inputs, outputs, the quality axis that matters, and any constraints (latency, cost, tool access, context size). Pick predictor shape:

Task shape	Predictor
Single-step structured I/O	`dspy.Predict` / `dspy.ChainOfThought`
Tool use / multi-step	`dspy.ReAct`
Code execution	`dspy.ProgramOfThought`
Long context / codebase	`dspy.RLM` → `dspy-rlm-module`

2. Program

Write the typed dspy.Signature + dspy.Module subclass per dspy-fundamentals. No hard-coded prompts. Keep predictors named so GEPA can target them.