feat: add dspy — expert skill for compiler-based prompt programming #79

New issue

Closed

opened 2026-07-09 14:40:16 -04:00 by jasper · 1 comment

jasper commented

2026-07-09 14:40:16 -04:00

Contributor

Summary

Add a dspy expert skill to the agent-skills portfolio. DSPy (by Stanford NLP) is a fundamentally different paradigm from our existing skills: it is a compiler for prompt programs, not a chain/RAG framework. Agents write Python programs with typed signatures and DSPy optimizes the prompts automatically.

Why This Skill

Our LlamaIndex and LangChain skills cover pipeline/RAG/agent orchestration. Neither can tell an agent "optimize this prompt systematically with Bayesian search" — that is DSPys domain. The routing guides already reference DSPy in the "optimization-driven prompt programming" row; this skill closes that gap.

Scope

Deep research: DSPy architecture (Predict, ChainOfThought, ReAct modules), teleprompters (BootstrapFewShot, MIPROv2, Bayesian), optimizers, evaluation metrics, program structure
Vault notes: atoms + molecules covering DSPy paradigm, modules, optimizers, evaluation
Expert skill: SKILL.md (thin index) + references (program design, optimizer selection, few-shot bootstrapping, evaluation), templates (classification, RAG, tool-use programs), scripts
Greenfield SkillOpt: 3 epochs (prominence, decision guidance, pattern expansion)
PR: Single PR to magnus/agent-skills

Key References

https://dspy.ai — official docs
https://github.com/stanfordnlp/dspy — GitHub
Stanford NLP DSPy paper (Khattab et al., 2023)

Acceptance Criteria

Deep research pyramid (~8 files) at .hermes/research/dspy/
Vault clipping + 20+ atoms + 4 molecules
Agent skill with 7+ references, 3+ templates, 1+ script
Greenfield SkillOpt 3 epochs
PR merged to main

Routing guide update: Add "Prompt optimization / compiled programs" → DSPy row to the Framework Routing Guide.

## Summary Add a **dspy** expert skill to the agent-skills portfolio. DSPy (by Stanford NLP) is a fundamentally different paradigm from our existing skills: it is a **compiler for prompt programs**, not a chain/RAG framework. Agents write Python programs with typed signatures and DSPy optimizes the prompts automatically. ## Why This Skill Our LlamaIndex and LangChain skills cover pipeline/RAG/agent orchestration. Neither can tell an agent *"optimize this prompt systematically with Bayesian search"* — that is DSPys domain. The routing guides already reference DSPy in the "optimization-driven prompt programming" row; this skill closes that gap. ## Scope - **Deep research:** DSPy architecture (Predict, ChainOfThought, ReAct modules), teleprompters (BootstrapFewShot, MIPROv2, Bayesian), optimizers, evaluation metrics, program structure - **Vault notes:** atoms + molecules covering DSPy paradigm, modules, optimizers, evaluation - **Expert skill:** SKILL.md (thin index) + references (program design, optimizer selection, few-shot bootstrapping, evaluation), templates (classification, RAG, tool-use programs), scripts - **Greenfield SkillOpt:** 3 epochs (prominence, decision guidance, pattern expansion) - **PR:** Single PR to magnus/agent-skills ## Key References - https://dspy.ai — official docs - https://github.com/stanfordnlp/dspy — GitHub - Stanford NLP DSPy paper (Khattab et al., 2023) ## Acceptance Criteria - [ ] Deep research pyramid (~8 files) at .hermes/research/dspy/ - [ ] Vault clipping + 20+ atoms + 4 molecules - [ ] Agent skill with 7+ references, 3+ templates, 1+ script - [ ] Greenfield SkillOpt 3 epochs - [ ] PR merged to main Routing guide update: Add "Prompt optimization / compiled programs" → DSPy row to the Framework Routing Guide.

jasper commented

2026-07-11 16:49:39 -04:00

Author

Contributor

Resolved by merged PR #83: #83

Filed by Jasper (AI agent on behalf of Magnus Hedemark)

Resolved by merged PR #83: https://git.brandyapple.com/magnus/agent-skills/pulls/83 Filed by Jasper (AI agent on behalf of Magnus Hedemark)

jasper closed this issue

2026-07-11 16:49:40 -04:00

No labels

community-feedback

enhancement

skill-upgrade

No milestone

No project

No assignees

1 participant

Notifications

Due date

The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set.

Reference

magnus/agent-skills#79

No description provided.

Rows
Columns