Prompt library browser
Good prompts are reusable assets, but they tend to scatter across notes, chat histories, and screenshots. This browser collects a curated set of high-quality prompt templates in one searchable place, tagged by model and task category, so you can find a solid starting point in seconds and copy it with one click.
How it works
The full library is embedded in the page as structured data — each entry has a title, a target model, a category, and a template body with bracketed placeholders. As you type in the search box or change the model and category filters, the list narrows instantly in your browser. There are no network calls, no accounts, and no tracking: filtering is pure client-side string matching over the embedded set.
Each result shows the template text exactly as you would paste it, with
placeholders like [TOPIC] or [CONSTRAINTS] marking where your own input goes.
The copy button puts the template on your clipboard so you can adapt it in your
LLM of choice.
What the categories cover
The library is organised around the task types where templating delivers the most value — situations where the prompt structure is reusable even though the content changes every time:
Writing and editing — templates for blog outlines, first-draft generation, editing passes (tone, clarity, length), and style transfer. These share an instruction skeleton that you fill with topic, audience, and constraints.
Analysis and summarisation — for condensing documents, extracting structured data from unstructured text, and comparing sources. The key structural element these templates carry is an explicit output format, which prevents the model from choosing its own.
Coding — templates for code review, docstring generation, refactoring, test writing, and debugging. They typically include a role definition (“you are a senior engineer”) and a constraint block (“do not change public interfaces”).
Research and Q&A — templates that include chain-of-thought scaffolding, citation requirements, and uncertainty flagging. These are the most sensitive to model selection, since reasoning quality varies.
Data and business — templates for SQL generation, spreadsheet formulas, market research summaries, and proposal drafts. They lean heavily on format instructions to keep output machine-parseable.
How model tags work
Templates tagged for a specific model reflect genuine formatting differences, not just brand preference. For example:
- Claude templates frequently use XML section delimiters (
<instructions>,<context>) because Claude is trained to treat these as clear boundaries. - GPT-4 templates often use numbered instruction lists, which that model follows reliably without extra framing.
- Gemini templates lean toward concise, multi-step prompts where each step is clearly labeled.
A template tagged “Claude” will generally work on GPT-4 too, but you may get cleaner results if you convert the XML tags to a numbered list format, and vice versa. The tag is a suggestion, not a constraint.
Tips for getting the most from templates
Treat every template as a starting skeleton, not a finished prompt. The placeholders are the load-bearing parts — the more specific you make them, the better the output. Use the category filter to discover patterns you would not have thought to write yourself, then keep your own edited versions in a personal note for repeat use.
When a template produces a mediocre result, the first thing to check is whether you have filled every placeholder with enough detail. Vague inputs produce vague outputs; a template with [AUDIENCE] filled in as “people” will underperform the same template filled in as “mid-career accountants evaluating new audit software.”