Prompt Foundry
Stop writing prompts. Engineer them. 32 XML-tagged templates, eval files, a TS loader, and a 28-page methodology PDF.
**Stop writing prompts. Engineer them.**
A library of 32 production-grade prompt templates — each XML-tagged, versioned, and shipped with golden-example eval files. Plus a TypeScript loader, a Notion dashboard, a Claude Code subagent, and *The Prompt Spec v1*: a 28-page methodology guide.
This is not a PDF of 1,000 ChatGPT prompts. It's prompts-as-code: versioned, evaluable, hot-reloadable. Compatible with Claude Opus 4.7, GPT-5, Gemini 2.5, and any model behind Vercel AI Gateway.
The story behind it
## Why Prompt Foundry
In 2026 most prompts still live as inline strings in app code. They have no version, no eval coverage, no contract. When the model gets something wrong, you patch the string, redeploy, and pray. When the model updates, you don't know which prompts regressed until users complain.
Foundry's 32 templates are the 32 contracts we use most. Each is structured, named, and shipped with a `.jsonl` of golden cases. The TypeScript loader gives you `loadPrompt(id, vars)` with type-safe interpolation. The Notion dashboard gives the rest of your team a searchable view.
The 28-page Prompt Spec PDF is the methodology behind the library — read it once, refer to it forever.
What's inside
- 32 XML-tagged production prompts across 8 categories
- Each prompt ships with 5–7 golden eval cases (.jsonl)
- Zero-dependency TypeScript loader with hot reload
- Compatible with Claude / GPT-5 / Gemini via AI Gateway
- 28-page methodology PDF: The Prompt Spec v1
- Commercial license, lifetime updates
▸ prompt-foundry/▸ prompts/· code-review.md 3.2KB· refactor.md 1.6KB· test-gen.md 1.8KB· doc-gen.md 1.6KB▸ analysis/▸ writing/▸ summarization/▸ classification/▸ extraction/▸ tool-use/▸ multi-turn/▸ evals/▸ promptfoundry-loader/· load.ts 3.4KB· types.ts 1.5KB· prompt-engineer.md 2.4KB· the-prompt-spec-v1.md 10.7KB· notion-template.md 1.0KB· CHANGELOG.md· LICENSE.md· README.md