Claude prompts that unlock Sonnet.
<role>, <task>, <constraints> tags Claude actually keys on. Plus 58 ready-to-fill Claude templates and a 200k-context chunker. Free, no API key.Top Claude prompts to steal today
Self-Refine — Generate → Critique → Revise (×3)
The single move that makes any output 2× better. Used in agentic systems.
Tree-of-Thoughts Decision Brancher
Explore 3 branches per decision, prune ruthlessly, beat one-shot answers.
Constitutional Self-Critique
Define principles up front, generate the answer, then audit it against each principle. Anthropic-style safety.
Multi-Persona Expert Council
Simulate 5 experts arguing — get a richer answer than any one expert could produce.
Adversarial Red-Team Audit
Generate the strongest attack on your idea before launching. Find vulnerabilities before users do.
Pre-Mortem Failure Analysis
Imagine the project failed 6 months from now. Work backwards. (Used at Amazon.)
How to write a great Claude prompt
The same patterns the FixAIPrompt linter checks for, condensed into actionable tips.
Use XML tags for sections
Claude was trained on XML-tagged sections. Wrap role, task, context, and output_format in tags and watch obedience to instructions jump.
Let Claude think before answering
Append "Think step by step before responding" — Claude is especially good at Chain-of-Thought reasoning when explicitly invited.
Use the 200k context window
Claude Sonnet/Opus has a 200k-token window. Paste entire codebases, long docs, full transcripts. Our chunker has Claude presets built-in.
Ask for citations from context
When using RAG with Claude, instruct it to quote the source text. Claude is unusually good at refusing to fabricate when asked to cite.
System prompt > user prompt for rules
Put hard constraints (never quote competitors, always respond in JSON) in Claude's system parameter, not the user message — much harder to jailbreak.
Use Multi-Persona Debate
Claude is exceptionally good at simulating multiple personas. Have it convene a 5-expert council to pressure-test important decisions.
Why use FixAIPrompt with Anthropic's Claude (Sonnet, Opus, Haiku)?
- Free, browser-only, no API key needed
- Auto-rewrites for Claude's preferred syntax
- 5-metric score (clarity / context / structure / specificity / risk)
- 58 prompt templates with interactive fill-in fields
- Detects 30+ secrets/PII before you paste into Claude
- Chunks long text to fit Claude's context window
Claude prompt FAQ
Answers to the most-asked questions about Anthropic's Claude (Sonnet, Opus, Haiku) prompting.
›Why does Claude prefer XML tags?
Anthropic trained Claude with heavy use of XML in its instruction data. Tagged sections like <role>, <task>, <context>, <output_format> reliably increase output quality and obedience compared to plain prose. Our Prompt Fixer renders every prompt in this XML style when 'Claude' is the selected target.
›What's the difference between a Claude prompt and a ChatGPT prompt?
Same content, different syntax. Claude responds best to XML-tagged sections; ChatGPT responds best to markdown headings; Gemini responds best to tight bullets. Our auto-fixer produces all three from a single input — click between the Claude/ChatGPT/Gemini tabs in the corrected-prompt preview.
›Are these Claude prompts free?
Yes. Every prompt template, the auto-rewriter, the chunker, and the secret detector are 100% free and run entirely in your browser. No accounts, no API keys.
›Do these work on Claude Sonnet and Opus?
Yes — the templates work on Claude Haiku, Sonnet, Sonnet 3.5, Opus, and any future Claude model. They use plain-English XML and rely on Anthropic's instruction-following.
›Can I use the Chunker with Claude's 200k context?
Yes. The Chunker has Claude Sonnet (200k), Claude Opus (1M), and Claude Haiku (200k) as built-in model presets — each pre-fills the recommended chunk size and overlap.
›What's the best Claude prompt for code?
Try the ReAct Agent Skeleton, Self-Refine Loop, or Constitutional Self-Critique templates. All three use XML tags and structured reasoning that Claude was explicitly trained on.