18 system prompts, task prompts and agent configs for coding, writing, email, support and research. Fill in the {placeholders}, copy the result, and jump to the model that ranks best for the task.
A free system prompt library with 18 prompts for AI agents and assistants: system prompts, task prompts, OpenClaw agent configs and model routing rules. Filter by task, search the full text, fill in {placeholders} inline and copy the filled prompt. Each prompt shows a token estimate and links to the best model for its task.
A system prompt for coding agents that keeps answers short: changed code only, one-line explanations, root cause first.
You are a senior software engineer. Be concise and direct.
Rules:
- Show only changed code, not entire files
- Use comments like "// ... existing code ..." to skip unchanged sections
- Explain decisions in 1-2 sentences max
- If multiple approaches exist, pick the best one and explain why briefly
- Always consider edge cases and error handling
- Write {language} by default unless told otherwise
- Follow existing code patterns and conventions in the project
When debugging:
- State the root cause first
- Show the fix
- Explain why it works in one line
Never output code you haven't verified mentally. Never apologize. Never repeat the question back.Was this tool helpful?
Five parts, each one line or a short list. These come from the customer support agent in the library.
You are the support assistant for {company}, which makes {product}.
Tells the model whose voice to use and what it knows.
Answer customer questions about {product} using only the help articles and account details you're given.
Says what it may draw on, so it stops at the edge of what you gave it.
No more than 120 words per reply unless you're giving steps.
Fixes the shape of every answer, which is what your code or your reader depends on.
Never guess at prices, dates or policies.
The lines it must never cross, written as plain instructions.
Hand off: say a person will follow up and give {escalation_contact}.
Without this, a model fills gaps with confident guesses.
Choose a task or search the full text, then open a prompt to see all of it.
Type values for placeholders like {language} or {company}. The preview updates as you type.
Copy the filled prompt into ChatGPT, Claude, an API call or your agent, then check which model ranks best for that task.
Each goes in a different place. A system prompt is sent with every call and a task prompt only once, so they cost differently too.
| System prompt | Task prompt | Agent config | |
|---|---|---|---|
| Where it goes | Custom instructions, a project, the API system message | The chat box or the user message | Your agent's config file |
| Sent | With every request | Once, for one job | Read at startup |
| Holds | Role, scope, format, rules | Context, instructions, length | Models, fallbacks, limits |
| Token cost | Adds up with every call | Paid once | Not sent to the model |
A system prompt is only half the setup. See which model ranks best for the job on intelligence, cost and speed.
18 prompts for AI agents and assistants: system prompts for coding, code review, SQL, writing, email triage and customer support, task prompts for blog posts, research and meeting notes, three OpenClaw agent configs and a set of model routing rules. You can filter them by task, search the full text and copy any of them in one click.
A system prompt is the standing instruction a model gets before any user message. It sets the role, the rules, the tone and the output format for a whole conversation or app. In ChatGPT it goes in custom instructions or a GPT's instructions, in Claude in a project's instructions, and in an API call in the system message.
Pick a prompt and type into the fields above the preview. Each field matches a placeholder such as {language} or {company}, and the preview updates as you type, with filled values in green and empty ones in orange. The copy button copies the filled version. Anything you leave empty stays as a {placeholder} you can fill in later.
A good system prompt names the role, says what's in and out of scope, fixes the output format and length, and lists the rules the model must never break, such as not guessing prices. It also says what to do when the model doesn't know. Keep it as short as those parts allow, because it's sent with every request.
The prompts here run from about 100 to 320 tokens, which is enough for a role, rules and a format. Every token is billed on every request, so a 400-token system prompt on a model at $2 per million input tokens adds $0.80 per 1,000 calls. Every prompt in the list shows an estimate of its token count.
Each prompt links to the matching Best model page, which ranks current models on intelligence, cost per 1,000 tasks and speed for coding, writing, email, research, documents or agents. The prompts are plain text and work with GPT, Claude, Gemini, DeepSeek and other chat models, though stronger models follow long rule lists more reliably.
They're OpenClaw configuration files at three budgets: free and budget models, a balanced developer setup on GPT-6 Sol, and a premium setup led by Claude Opus 5.5. Each sets a primary model, fallbacks, an image model and subagent limits. To build your own, use the Agent Config Generator.
Yes. Copy them, change them and use them in your own apps, agents and chats. There's no signup, and the variables you type stay in your browser. For a prompt that isn't here, the AI Prompt Generator writes one from a description.
Everything from model picks to CSV converters, with no account needed.