🧠 Claude Tips, Tricks & Hidden Features
A community-curated collection of tips, tricks, hidden features, prompt templates, and code snippets for getting the most out of Claude AI.
Claude is one of the most capable AI assistants available — but most people only scratch the surface of what it can do. This repo collects battle-tested tips, undocumented behaviors, clever prompting patterns, and reusable scripts from the community.
Whether you're a developer calling the API, a power user using Claude.ai, or a prompt engineer crafting complex workflows — there's something here for you.
Claude is trained to understand XML-style tags natively. Use them to separate context, instructions, examples, and data clearly.
<context>
You are an expert data analyst at a SaaS company.
</context>
<task>
Analyze the following churn data and identify the top 3 risk segments.
</task>
<data>
{{paste your data here}}
</data>
<format>
Return a markdown table with: Segment | Risk Score | Recommended Action
</format>
Why it works: Claude was trained with XML-structured prompts and processes tagged sections with higher fidelity than plain paragraphs.
Positive instructions alone often leave too much room. Add explicit negative constraints.
Summarize this article.
- Do NOT use bullet points
- Do NOT include your own opinions
- Do NOT exceed 3 sentences
Trigger chain-of-thought by asking Claude to reason first.
Before answering, think through this step by step.
Show your reasoning, then give your final answer.
Question: [your question]
This phrase consistently elicits more thorough, careful responses on complex topics compared to just asking the question directly.
Explain quantum entanglement to:
- A 10-year-old
- A physics undergrad
- A quantum computing researcher
Claude will adjust vocabulary, depth, and examples for each.
Hidden Features
Set a system prompt (or opening message) once and Claude will maintain consistent behavior throughout:
You are Alex, a brutally honest senior software engineer with 20 years of experience.
You give direct feedback without sugarcoating. You value simplicity over cleverness.
You speak in short, punchy sentences.
When Claude's response gets cut off, use this exact phrase:
Continue exactly where you left off, starting from: "[last few words]"
This prevents Claude from summarizing or restarting.
Claude will reliably output structured JSON when you provide the schema:
Respond ONLY with valid JSON. No explanation, no markdown, no preamble.
Schema:
{
"title": "string",
"summary": "string (max 100 words)",
"sentiment": "positive | neutral | negative",
"tags": ["string"]
}
Analyze the following text: {{text}}
Adding these phrases dramatically improves code quality:
"Production-ready code only""Include error handling""Add inline comments explaining non-obvious logic""Follow [language] best practices and idioms"
Before you begin, ask me the 3-5 most important clarifying questions
that would affect your approach. Wait for my answers before proceeding.
## About the user
- Name: [name]
- Role: Senior backend engineer
- Stack: Python, FastAPI, PostgreSQL, Redis
- Prefers: Concise answers, code examples over prose, no hand-holding
- Never: Use bullet points for explanations
When context is getting long, ask Claude to compress it:
Summarize everything we've discussed so far into a compact briefing
I can paste at the start of a new conversation to continue seamlessly.
Claude.ai Projects let you set a persistent system prompt and upload reference documents that persist across all chats in that project. Great for: codebases, style guides, research corpora.
Format your response as:
- A Notion-ready document with H1/H2/H3 headers
- A Slack message (no markdown, casual tone, < 300 chars)
- A formal memo with To/From/Date/Subject/Body
- An RFC document with Motivation, Proposal, Alternatives, and Drawbacks sections
"Answer in exactly 3 sentences.""Keep your response under 150 words.""Give me an exhaustive, comprehensive answer. Do not truncate."
Compare these 5 database options across: Performance, Cost, Ease of Use,
Scalability, and Community Support. Use a markdown table.
I want you to [complex task].
First, create a detailed plan with numbered steps.
Show me the plan and ask for my approval before executing anything.
When using Claude via API in agentic settings, orchestrate multiple specialized Claude instances:
# See /scripts/python/parallel_subagents.py for full implementation
orchestrator = ClaudeOrchestrator()
results = await orchestrator.run_parallel([
{"role": "researcher", "task": "Find competitor pricing"},
{"role": "analyst", "task": "Analyze our cost structure"},
{"role": "writer", "task": "Draft executive summary template"}
])| Action | Shortcut |
|---|---|
| New chat | Ctrl/Cmd + Shift + O |
| Focus input | Esc then type |
| Submit message | Enter |
| New line in message | Shift + Enter |
When Claude generates code, documents, or visualizations as Artifacts, you can:
- Ask Claude to modify specific parts without rewriting the whole thing
- Download the artifact directly
- Share a link to the artifact (Claude.ai Pro)
Claude.ai supports uploading PDFs, images, CSVs, code files, and more. You can:
- Upload an entire codebase (as a ZIP or individual files)
- Reference uploaded docs with "As described in the uploaded document..."
- Upload screenshots for UI debugging
Pre-built system prompts for common use cases. See /prompts/system-prompts/
| File | Description |
|---|---|
senior-engineer.md |
Code review & technical mentorship persona |
research-analyst.md |
Deep research and synthesis |
socratic-tutor.md |
Learn anything via Socratic dialogue |
editor.md |
Writing coach and copy editor |
product-manager.md |
PRD writing, user story generation |
devils-advocate.md |
Red team your ideas and decisions |
Single-shot prompts for specific tasks. See /prompts/task-prompts/
| File | Description |
|---|---|
summarize-paper.md |
Academic paper → structured summary |
code-review.md |
Thorough code review with scoring |
write-tests.md |
Generate comprehensive test suites |
explain-codebase.md |
Onboard to a new codebase fast |
rewrite-email.md |
Polish emails for any tone/audience |
sql-from-english.md |
Natural language → SQL queries |
Multi-step prompt sequences for complex workflows. See /prompts/chain-prompts/
| File | Description |
|---|---|
research-pipeline.md |
5-step deep research workflow |
blog-post-writer.md |
Outline → draft → edit → SEO |
system-design.md |
Requirements → architecture → tradeoffs |
| File | Description |
|---|---|
basic_chat.py |
Minimal API chat loop |
streaming_response.py |
Stream tokens as they generate |
structured_output.py |
Force JSON with Pydantic validation |
parallel_subagents.py |
Run multiple Claude calls in parallel |
conversation_manager.py |
Multi-turn conversation with context trimming |
batch_processor.py |
Process files/rows with Claude in bulk |
| File | Description |
|---|---|
basic_chat.js |
Node.js chat example |
stream_to_ui.js |
Stream response to browser UI |
tool_use.js |
Function/tool calling example |
| File | Description |
|---|---|
claude_ask.sh |
One-liner CLI wrapper |
summarize_file.sh |
Pipe any file to Claude for summary |
Contributions are what make this repo valuable. Please read CONTRIBUTING.md before submitting.
Quick contribution guide:
- Fork this repo
- Create a branch:
git checkout -b tip/your-tip-name - Add your content following the templates
- Open a Pull Request
Every tip, prompt, and script should include:
- A clear title and description
- Why it works (the reasoning)
- A concrete example
MIT License — see LICENSE for details.
If this helped you, please ⭐ the repo and share it with others!