Fabric — AI Prompt Patterns for Everything

What it is

Fabric is a collection of over 100 AI prompt patterns designed for real-world tasks. Each pattern is a carefully crafted system prompt that instructs an LLM to perform a specific task: summarize articles, extract key wisdom, analyze code quality, write essays, create presentations, rate content, and more. Fabric runs as a CLI tool that pipes content through these patterns.

This tool is for anyone who uses LLMs regularly and wants proven, reusable prompt templates instead of writing ad-hoc prompts for common tasks.

How it saves time or tokens

Fabric patterns are optimized through community iteration. Each pattern includes specific output formatting instructions, reducing the back-and-forth of getting the LLM to produce useful output. The CLI interface pipes content directly, eliminating copy-paste workflows. Chaining patterns enables complex workflows like 'extract key points, then create a summary, then generate social posts.'

How to use

1. Install Fabric CLI.
2. Configure your LLM API key.
3. Pipe content through patterns.
4. Chain patterns for multi-step workflows.

# Install Fabric
go install github.com/danielmiessler/fabric@latest

# Or via pip
pip install fabric-ai

# Summarize an article
curl -s https://example.com/article | fabric --pattern extract_wisdom

# Analyze code
cat main.py | fabric --pattern analyze_code

# Create a presentation outline
cat research.md | fabric --pattern create_presentation

# Chain patterns
cat article.md | fabric --pattern extract_wisdom | fabric --pattern create_summary

Example

Using the extract_wisdom pattern:

$ echo 'Long article about AI safety...' | fabric --pattern extract_wisdom

IDEAS

• AI alignment research needs more funding
• Interpretability tools are improving rapidly
• Red-teaming is becoming standard practice

INSIGHTS

• The gap between AI capabilities and safety research is narrowing
• Open-source safety tools lower the barrier to responsible development

QUOTES

• 'Safety is not a feature, it is a foundation'

RECOMMENDATIONS

• Invest in interpretability before scaling further
• Mandate red-teaming for models above certain capability thresholds

Related on TokRepo

• Prompt library — More curated prompts
• AI coding tools — Developer-focused AI tools

Common pitfalls

• Pattern quality varies. Some community patterns are well-tested while others are rough drafts. Test patterns before relying on them.
• Patterns are optimized for specific models. A pattern tuned for GPT-4 may produce different results with Claude or Gemini.
• Long input content may exceed model context limits when combined with the pattern's system prompt.
• The CLI requires API keys configured. Each invocation makes API calls, so costs accumulate with heavy use.
• Chaining patterns multiplies API calls and token usage. Use chaining selectively for high-value workflows.