Prompt Engineering Patterns
Treats a production prompt as something to measure, not to reword.
Install
Ships scripts/plugin marketplace add wshobson/agents; /plugin install llm-application-dev
When a prompt gives inconsistent output, needs structured JSON, or a template library is being set up.
- Author
- wshobson
- License
- MIT
The techniques are the ones that change output reliability rather than tone: selecting few-shot examples by similarity and diversity instead of picking the first three, eliciting step-by-step reasoning and sampling several paths for self-consistency, and constraining output to JSON so the caller can parse it without regex. Templates with variable interpolation keep prompts in version control rather than scattered through the code. A prompt optimization script ships with the skill, plus reference files per technique and a starting template library.
Similar skills
Prompt Optimizer
SkillTurns a rough request into one finished prompt meant to be pasted into a chat window.
Agent Orchestration
SkillTurns agent tuning into a measured loop with a baseline, classified failure modes and a rollback path.
Taste Skill
SkillPushes agent-built interfaces past the default template look.
Comments(0)
Sign in to comment