claude-specific · anthropic best practices

Claude Skills
Builder

Apply Anthropic's official Claude skill patterns in 4 steps — select a skill, define your task, configure settings, and get a production-ready prompt engineered for Claude's unique capabilities.

Step 1 of 4

Choose a Claude Skill

click a skill to continue

Step 2 of 4

Define Your Task

What do you want Claude to do?

Background context (optional)

Upload File Context (optional — .txt, .md, .csv, .pdf)

Drop file here or click to browse

PDF files must be less than 10 MB

Step 3 of 4

Configure Skill

Choose which XML sections to include. Each tag gives Claude a distinct semantic context to process separately.

<role> tag

Sets Claude's expert identity and perspective

<context> tag

Background information and situational data

<constraints> tag

Rules, limits, and what to avoid

<examples> tag

Few-shot examples for pattern matching

<output_format> tag

Specifies the structure of the response

Configure how deeply Claude should reason before answering. Deeper thinking = better results on complex tasks, slower response.

Thinking depth Medium

Challenge initial assumptions

Instruct Claude to question its first answer

Show reasoning steps

Include thinking process in the response

Configure how Claude flags uncertainty in its responses to reduce hallucinations.

Confidence markers [high/mid/low]

Append confidence level to each claim

Source verification flags

Flag claims that need external verification

Explicit knowledge cutoff notes

Note when information may be outdated

Optimize Claude's behavior for long document analysis across its 200k token context window.

Number of documents 3

Cross-reference sections

Cite specific documents and sections per claim

Detect contradictions

Flag when documents disagree on facts

Define the expert persona Claude will embody throughout the conversation.

Expert role / title

Reference background naturally

Draw on experience in every response

Never break character

No "as an AI" disclaimers

Configure the analytical framework for exploring multiple perspectives before synthesis.

Include strongest case FOR

Steelman the affirmative position

Include strongest case AGAINST

Steelman the opposing position

Include skeptic perspective

What a rigorous skeptic would challenge

What evidence would change analysis

Force Claude to identify falsifiable conditions

Final synthesis

Balanced conclusion across all perspectives

Configure voice matching for ghostwriting and brand content generation.

Number of writing samples to provide 3

Analyze vocabulary patterns

Word choice, register, and vocabulary level

Analyze sentence rhythm

Length patterns, pauses, and pacing

Analyze tonal register

Formal/casual, warm/clinical relationship to reader

Configure the scope of Claude's code architecture review.

Security vulnerabilities

Injection, auth flaws, data exposure, etc.

Architecture anti-patterns

Design flaws, coupling, abstraction issues

Performance bottlenecks

N+1 queries, memory leaks, blocking operations

Missing abstractions

Repeated patterns that should be extracted

Reference file + line numbers

Pin every finding to specific code location

Configure the self-critique loop — Claude generates, evaluates, then refines its own response.

Critique for clarity

Flag unclear or ambiguous sections

Critique for completeness

Identify missing information or coverage gaps

Critique for accuracy

Check claims for potential errors

Critique for actionability

Are recommendations specific and implementable?

Show [Draft] / [Critique] / [Revised] labels

Visible section labels in the response

Configure strict citation mode to eliminate hallucination in document-based research.

Quote source passages

Include direct quotes for every factual claim

Cite document name + section

Reference exactly where information comes from

Explicit gaps statement

State clearly when documents don't cover something

No outside knowledge

Block Claude from drawing on training data beyond docs

Step 4 of 4

Generate Skill Prompt

Your Claude skill prompt is ready to assemble. Click Generate to build it — then copy and paste directly into Claude.

ready to generate

skill prompt output

Select a skill and configure it,
then click Generate.