Prompt Library

A prompt library is a curated collection of reusable prompt patterns for AI language models. Each pattern includes the prompt text, placeholder variables, example values, expected output shape, limitations, and adaptation notes. A library helps you avoid rewriting prompts from scratch and gives you tested starting points for common tasks.

No. These patterns are starting points, not guarantees. Prompt behaviour depends on the specific model, its version, the system context, and how the filled values interact with the instructions. Always test a pattern with your intended model before relying on it in production.

Yes. Click "Copy prompt" on any pattern to copy the raw template text, including placeholder variables. Use "Copy with example values" to copy a version with the illustrative values already filled in. You can then paste it into ChatGPT, Claude, Gemini, or any other interface.

Double-brace placeholders ({{variable_name}}) mark positions where you should insert your own content. Replace {{title}} with your actual title, {{audience}} with your actual audience, and so on. The "Placeholder variables" table in each detail panel shows an example value for each variable.

Read the "Adaptation notes" section of each pattern. It explains which parts to change, which constraints to relax, and how to combine the pattern with follow-up prompts. The Prompt Template Builder tool lets you load a pattern, define named variables with defaults, and preview the filled result.

No. The Prompt Library runs entirely in your browser. The dataset is embedded in the static page — there is no runtime API call, no content fetch, and no upload of any kind. Copying to clipboard and downloading (if applicable) are the only operations that leave the browser.

Each pattern was written by the ToolsSonic editorial team based on common, practical tasks. Selection criteria include: distinct approach (not superficial variants), checkable output shape, explicit limitations, no proprietary or submitted content, and a clear audience. Patterns are versioned and dated for transparency.

The current collection covers 19 categories: Writing (blog post, email reply, press release, LinkedIn post, newsletter, case study, YouTube script), Summarization (article bullets, meeting notes, executive summary), Extraction (structured data, requirements), Classification (support tickets, sentiment), Brainstorming (idea generation, pre-mortem), Coding Assistance (code review, explanation, unit tests, refactoring, SQL query, API design, regex generator), Debugging (error diagnosis), Documentation (function docstring, README), Research Planning (research plan, user interview guide), Customer Support (escalation, FAQ answers, apology response), Marketing Review (copy review, A/B test brief), Structured Output (JSON schema, CSV data generator), SEO (meta tags, content brief, schema markup, keyword cluster, internal linking plan), HR & Recruiting (job descriptions, interview questions, performance review), Education (lesson plans, quizzes, flashcard set, assessment rubric), Sales & Outreach (cold email, objection playbook, follow-up email), Social Media (Twitter thread, content calendar, Instagram caption), Product Management (user stories, PRD sections, OKRs, competitive analysis, sprint goal), and Legal Review (contract risk flagging, privacy policy drafting).

The Prompt Library is a browsable catalogue of curated patterns you can copy and adapt. The Prompt Template Builder is a workspace for authoring your own reusable templates with named variables, defaults, and a live preview. You can open any library pattern in the Template Builder to customize it as your own template.

The Prompt Optimizer analyses your own prompt text for clarity, structure, and heuristic issues. The Prompt Library provides pre-written patterns. They complement each other: find a pattern here, adapt it, then run it through the Optimizer to check formatting before sending to a model.

No. The current release is a curated, static dataset — there is no submission form, community voting, or user-generated content. This is intentional: it keeps the collection editorially controlled and avoids safety, copyright, and quality issues from open submissions.

Each pattern includes example values for its placeholder variables. "Copy with example values" fills every {{variable}} with its example value before copying. The result shows you what a completed prompt looks like. It is intended for illustration — replace the example values with your own before sending to a model.

The version and date indicate when the pattern was written or last reviewed by the editorial team. It is not automatically updated. If you notice a pattern that needs revision, it can be flagged for review in the next editorial pass.

No standalone system prompts in the current release. The patterns are instruction-style prompts intended for the user turn. Some patterns (like Code Review) function as a de-facto system context, but none are labelled as hidden or pre-injected system prompts.

No. The current collection covers text-in / text-out language models only. Image generation, speech synthesis, and multimodal prompts are a different domain with different conventions and are not included.

Use the search box to search by keyword, category, task description, or audience. The category filter chips let you narrow to a specific domain (Writing, Coding, Classification, etc.). The status bar shows how many patterns match your current filter.

Output shape describes the expected structure or format of the model's response — for example "Markdown outline with H2 headings" or "JSON object with four keys." It helps you know what to check for in the output and how to validate it programmatically.

The prompt patterns are provided as editorial examples. You are free to adapt them for your own use, including commercial projects. We do not claim that any pattern will produce a specific result or that it is "the best" approach for any task. You remain responsible for the outputs you generate.

The patterns are written to be model-agnostic and should work with most instruction-following language models. Specific model names are not referenced in the prompt text. Some patterns (like the Structured Output / JSON pattern) may perform better with models that have native JSON mode support.

In the detail panel of any pattern, click "Open in Template Builder ↗". This opens the Template Builder tool in a new tab. You can then paste the prompt text into the editor, click "Extract variables" to build the variable inventory, and configure labels and defaults.

The empty state shows a "reset filters" link that clears both the search query and the category filter. You can also clear the search input manually or click the ✕ button at the right of the search field.

Yes. Open any pattern's detail panel and you will find a "Fill in variables" form with an input field for each placeholder. As you type, the live preview below updates in real time with your values highlighted. Click "Copy filled prompt" to copy the completed text to your clipboard — no need to open a separate tool.

After filling in your variables, these buttons open the respective model's interface in a new tab. Your filled prompt is already copied to your clipboard — just paste it in. It is a convenience shortcut, not an API integration: no data is sent from ToolsSonic to any AI provider.

Complexity is derived from the number of placeholder variables in a pattern: Simple (1–2 variables), Moderate (3–4 variables), Advanced (5 or more variables). It is a rough guide to setup effort — a Simple pattern can be copied and used immediately, while an Advanced pattern benefits from using the fill form to ensure all variables are correctly populated.

Yes. Use the "↓ Export visible" button in the toolbar to download all currently filtered patterns as a single JSON file. You can filter by category or model first, then export just that subset. Each exported entry includes all fields: id, version, category, promptText, variables, exampleValues, outputShape, limitations, notes, and models.

Use the Model filter chips below the category filter. Select GPT-4o, Claude 3.5, or Gemini 1.5 to show only patterns tested and tagged for that model. All patterns currently in the library work with all three, but the filter will become more meaningful as model-specific patterns are added in future updates.

Yes — these professional categories were added in the S102 update. SEO includes meta title/description generation, content briefs, and JSON-LD schema markup. HR & Recruiting includes job description writing and interview question generation. Sales & Outreach includes cold email and objection response playbooks. Legal Review includes contract clause risk flagging and privacy policy section drafting. All entries include the standard limitations disclosure.