DSPy Agent Skills
Five DSPy 3.2 skills for Claude Code and Codex: signatures, metrics, GEPA optimization, RLM and end-to-end pipeline assembly.
- 278 ★ on the repository
- Last commit
- MIT
Minimal access
Why this level
- The skills supply instructions and examples; a live optimization run spends paid model requests.
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add skills/dspy-agent-skillsDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewMinimal accessThe tool runs nothing and makes no network calls.
- The skills supply instructions and examples; a live optimization run spends paid model requests.
This is third-party code. Review the repository files before installing.
What it does
DSPy Agent Skills gives an agent five skills for DSPy 3.2.x: `dspy-fundamentals`, `dspy-evaluation-harness`, `dspy-gepa-optimizer`, `dspy-rlm-module` and `dspy-advanced-workflow`. Each has a short `SKILL.md`, a detailed `reference.md` and a runnable `example_*.py` with a `--dry-run` mode. The skills trigger by topic. The first covers signatures, modules, ReAct and saving programs, the second metrics and `dspy.Evaluate`, the third optimization through `dspy.GEPA`, the fourth long context through `dspy.RLM`. The fifth chains the others from a baseline to an exported optimized program. The API was checked against DSPy 3.2.1.
Who it is for. Developers of LLM applications on DSPy who want an agent to write against the current 3.2 API instead of stale memory.
Good fit when
- A new classifier needs a signature and a ChainOfThought module.
- A metric and a dev/val split for `dspy.Evaluate` are written from scratch.
- A finished program should be optimized with `dspy.GEPA` and saved as an artifact.
- Questions about a large codebase are handled by recursive exploration through `dspy.RLM`.
Not a fit when
- The project does not use DSPy.
- You need a DSPy older than 3.2: the examples are confirmed only for 3.2.x and partly 3.1.3.
- Data must not go to external models during a live optimization run.
Example request
Build a DSPy sentiment classifier, optimize it with GEPA and save the artifact.Limitations
The skills target DSPy 3.2.x and were checked on 3.2.1. If a package manager resolves an older DSPy, `UV_EXCLUDE_NEWER` or a stale mirror may hide the new release. `dspy.RLM` needs Deno. A live GEPA run needs a key such as `OPENAI_API_KEY` and spends paid requests. The third example rests on a DSPy 3.1.3 artifact. The last commit dates from September 2026. The author checked the Claude Code and Codex CLI specifications as of April 2026.
How to disable. The `--uninstall` flag of `scripts/install.sh` reverses that install. Remove the plugin and manually copied directories from `~/.claude/skills/` or `~/.agents/skills/` with agent tooling.
FAQ
How do I install for Codex?
Through `npx skills add intertwine/dspy-agent-skills --skill '*' -a codex -y` or by cloning and running `./scripts/install.sh --codex-only`.
Which skill triggers for optimization?
`dspy-gepa-optimizer`: it covers `dspy.GEPA`, and the `example_bettertogether.py` file shows a BetterTogether chain.
Can examples run without a key?
Yes. The `example_*.py` scripts run offline with `--dry-run` and need no API key.
What results do the demos show?
The README lists three demos with before and after GEPA numbers: for example RAG QA rose from 80.47 to 100.00. These are the author's figures on chosen models.
Which DSPy version is needed?
Version 3.2.x: checks ran on 3.2.1. `dspy.RLM` additionally needs Deno.
Discussion
No questions or reviews yet.
Related
SkillsA skills library for Claude Code, Cursor and Codex: team code review, TDD, stacked PRs and repository readiness scoring
A set of independent skills against drift in large codebases: hard-cut, root-cause tracing, lean testing and CI right-sizing.
Nine personal engineering skills: an idea-to-PR chain, standards search, a sandboxed review loop and wizard pages for changes
19 skills for Claude Code, Codex and Antigravity CLI: Grok and Exa search, Context7 docs, Serena and task hand-offs between agents
A library of 30-plus instruction skills for coding agents: planning, building, testing, hallucination guards and shipping
14 skills that find the gaps in a task before the code: interviews, plans, assumption tests and proof that a bug fix works
README in short
Five skills, three installation paths, three end-to-end demos with results and tests that validate skill format and versions.