Spec / 013 · library and workshop
There are two useful ways to treat a prompt. The library collects templates you can paste. The workshop treats the prompt as something you compile, test, constrain, or optimize. Templates are useful. Engineering is where the long-term value is. This site does not run a live model, compiler, or eval harness.
CC0 public-domain
Role and architect templates
DSPy, promptfoo, Guidance, TextGrad
Library
System prompts, copy-paste recipes, curated personas. A template adopts a role and a reply shape. It has no tools, no memory store, and no verifier unless the text names one.
Workshop
Compile an LM program, regress it with tests, constrain decoding, or search for better wording against a metric. The artifact is a program, a matrix, a grammar, or an archive · not a gist.
Workshop methods
Each method has a steppable conceptual workflow. None of them execute here.
Compile
DSPy compiles language-model programs: signatures, modules, a metric, and an optimizer — not a pasted persona.
Test / regress
promptfoo tests and regresses prompts: variants, providers, assertions, and a baseline you can fail CI on.
Constrain
Guidance constrains decoding: templates, grammars, and legal token sets — so structure is enforced, not hoped for.
Optimize
TextGrad optimizes prompts with textual gradients: forward generation, a language loss, backward advice, update.
Optimize
GEPA evolves prompts by reflecting on traces and keeping a Pareto archive — a population, not a single rewrite.
Library doors into the workshop
These templates describe compilers, eval architects, and evolutionary optimizers. They are still library paste · not running programs · but they are the right door into the workshop methods.
Your job is to optimize a prompt automatically inside a code-as-action loop.
Generates optimized prompts for any AI tool. Use when writing, fixing, improving, or adapting a prompt for LLM, Cursor, Midjourney, image AI, video AI, coding agents, or any other AI tool.
Your job is to design the runtime around the model, not just the prompt inside it.
You are an evaluation architect designing benchmarks and quality frameworks for LLM systems.
Your job is to optimize role-specific prompts for every agent in an LLM-based multi-agent system (MAS) so that local competence and global system success rise together.
Your job is to optimize system prompts through a self-referential evolutionary loop.