fabric

Applies curated, community-maintained prompt patterns to user input through a command-line interface. Fabric maintains a versioned library of tested prompts (stored as markdown files with embedded instructions) that users invoke by name, passing stdin or file content as context. The CLI resolves pattern names to prompt templates, injects user input, and routes to configured LLM backends (OpenAI, Anthropic, Ollama, etc.), returning structured or unstructured output based on pattern definition.

Provides a unified CLI interface that abstracts away differences between multiple LLM providers (OpenAI, Anthropic, Ollama, local models, etc.). Fabric detects or accepts a configured backend, translates prompt patterns into provider-specific API calls (handling token limits, model-specific parameters, and response formats), and normalizes output regardless of backend. This allows users to swap providers without rewriting patterns or CLI commands.

Provides CLI commands to list, search, and describe available prompt patterns in the local or remote pattern library. Fabric scans the patterns directory (typically ~/.fabric/patterns or a cloned Git repository), parses pattern metadata (name, description, tags), and presents them via commands like `fabric --list` or `fabric --search <keyword>`. Users can inspect pattern definitions before applying them, reducing trial-and-error.

Integrates deeply with Unix pipes and shell redirection, accepting input via stdin, file arguments, or clipboard, and outputting results to stdout for further processing. Fabric is designed as a filter in a shell pipeline, allowing users to chain multiple patterns or combine fabric with other CLI tools (grep, sed, jq, etc.) without intermediate files. This enables workflows like `cat document.txt | fabric --pattern summarize | fabric --pattern extract-entities | jq`.

Supports embedding variables or placeholders in prompt patterns that are substituted at runtime based on user input, environment variables, or pattern arguments. Patterns can define required or optional parameters (e.g., `{{LANGUAGE}}`, `{{TONE}}`) that users provide via CLI flags or environment variables, allowing a single pattern to be customized for different contexts without duplication. Fabric parses pattern files for template syntax and performs substitution before sending to the LLM.

Manages a local Git repository of prompt patterns, allowing users to clone the official fabric patterns library, pull updates, and optionally fork or create custom patterns. Fabric provides commands to initialize, update, and manage the patterns directory, treating it as a version-controlled artifact. Users can pin specific pattern versions, create local overrides, or contribute patterns back to the community via Git workflows.

Supports applying a single prompt pattern to multiple input files or documents in sequence, with options for parallel execution or sequential processing. Fabric can iterate over a directory of files, apply a pattern to each, and aggregate or save results. This is typically achieved via shell loops or xargs integration, but fabric may provide built-in batch commands to simplify common scenarios like 'summarize all PDFs in a directory' or 'extract entities from all logs'.

Provides a configuration system (typically YAML or JSON files) where users specify default LLM provider, API keys, model preferences, and other settings. Fabric reads configuration from standard locations (e.g., ~/.fabric/config.yml) and allows per-command overrides via CLI flags. Configuration supports multiple provider profiles, enabling users to switch between OpenAI, Anthropic, Ollama, etc. without editing files each time.

Allows users to specify output format (plain text, JSON, markdown, code) and optionally apply post-processing transformations (e.g., pretty-printing, filtering, extraction). Fabric may support format flags like `--format json` or `--format markdown`, and can pipe output through external tools (jq, sed, etc.) for further transformation. Some patterns may define their own output format expectations.

Provides workflows for users to create, test, and contribute new patterns back to the fabric community via Git pull requests. Fabric may include documentation or templates for pattern creation, and the GitHub repository accepts community contributions. This enables a decentralized pattern library where users can discover, use, and improve patterns collaboratively.