This repository provides a production-grade suite of five DSPy agent skills designed for coding agents, specifically targeting Claude Code and Codex CLI. It offers a unified, validated set of tools for implementing DSPy fundamentals, evaluation, optimization (GEPA), recursive learning (RLM), and advanced workflows, enabling developers to build more sophisticated AI agents with DSPy.

How It Works

The project packages five core DSPy functionalities as agentskills.io-compatible skills. These skills are designed for progressive disclosure, featuring concise SKILL.md and detailed reference.md files. The dspy-advanced-workflow skill orchestrates the others for end-to-end agent pipelines. The approach emphasizes validation against DSPy 3.2.1 and provides runnable Python example scripts with offline --dry-run capabilities for local testing.

Quick Start & Requirements

• Installation:
  • Claude Code: /plugin marketplace add intertwine/dspy-agent-skills
  • Agent Skills CLI: npx skills add intertwine/dspy-agent-skills --skill '*' -a codex -y
  • From Source: git clone https://github.com/intertwine/dspy-agent-skills && cd dspy-agent-skills && ./scripts/install.sh
• Prerequisites: DSPy 3.2.x (tested against 3.2.1), Python 3.10+, Claude Code, Codex CLI. Deno is required for dspy.RLM examples. API keys (e.g., OPENAI_API_KEY) are necessary for live model interactions.
• Links: Official DSPy docs: https://dspy.ai/. Claude Code skill spec: https://code.claude.com/docs/en/skills.md. Codex skill spec: https://developers.openai.com/codex/skills.

Highlighted Details

• Skills are validated against DSPy 3.2.1, ensuring spec compliance.
• Provides a single source of truth for both Claude Code and Codex CLI agent skills.
• Includes runnable example scripts with offline --dry-run functionality.
• Features end-to-end examples with committed baseline vs. GEPA-optimized performance metrics and DSPy version comparisons.
• Offers plugin manifests for one-click installation in compatible marketplaces.

Maintenance & Community

• Primary contributor: Bryan Young (@intertwine).
• Validation performed by Claude Opus 4.7.
• No community channels (e.g., Discord, Slack) are explicitly mentioned in the README.

Licensing & Compatibility

• License: MIT.
• Compatibility: DSPy 3.2.x, Claude Code (current), Codex CLI (current), Python 3.10+. Deno is required for RLM examples. The MIT license permits commercial use and linking.

Limitations & Caveats

• The npx skills add dspy-agent-skills command requires upstream CLI support for source aliases; use the full intertwine/dspy-agent-skills identifier.
• One example artifact (03-invoice-extraction) uses DSPy 3.1.3 due to limited optimization headroom in DSPy 3.2.0 for that specific task.
• Live execution of examples necessitates valid API keys for language models.
• The UV_EXCLUDE_NEWER environment variable may interfere with resolving the correct DSPy 3.2.1 wheel.