PromptG CLI

Prompts as code. Versioned, shareable, standard.

PromptG CLI is a tool for managing AI prompts as versioned JSON files, allowing easy storage, sharing, and rendering to plain text for use with any LLM.

promptg get code-review | llm

Quickstart (60 seconds)

Requirements: Node.js 20.12+

# Install
npm install -g @promptg/cli

# Create a project store (.promptg/)
promptg init

Create a prompt (stdin -> stored JSON):

echo "Review this code for correctness, security, and performance." | promptg prompt save code-review

Render it (stdout -> pipe anywhere):

promptg get code-review | llm

CLI (scriptable, pipe-friendly)

The CLI is designed for automation: render prompts to stdout, inject runtime context via variables, and pipe into any runner.

# With variables
promptg get code-review --var language=Python --var focus=security | llm

# Load multiline values from a file (diffs/logs/etc.)
git diff > pr-diff.txt
promptg get pr-review --var diff@pr-diff.txt | llm

How It Works

• Store prompts/templates as JSON in a user store (~/.promptg/) or a project store (.promptg/).
• Render prompts with promptg get <name> (plain text to stdout).
• Templates are not rendered; instantiate them into prompts with promptg prompt new <prompt-name> --from-template <template>.
• Share across repos/teams by committing .promptg/ or distributing packs.

Core Concepts

• Prompt: a ready-to-run, concrete prompt. It has final-ish text (may still include variables) and is meant to be rendered and used directly via promptg get <name>.
• Template: a reusable blueprint designed to be instantiated into prompts. It's the portable, shareable source-of-truth for a workflow (PR review, release notes, issue triage), and can include defaults + optional interactive metadata.
• Pack: a versioned bundle of prompts/templates for sharing across repos.

Blueprints -> Instances

The intended workflow is:

• Install or author templates (blueprints) once.
• Create prompts (instances) from templates for a specific repo/team/task.
• Dispose/regenerate prompts freely as your context changes; keep templates stable.

Why templates require metadata:

• Catalog separation: tools can separate a "Template Catalog" (browse blueprints) from "My Prompts" (execute instances), reducing noise.
• Catalog quality: required displayName/description ensures useful browsable views.

Quick examples:

promptg prompt new my-pr-review --from-template pr-review --var diff@changes.txt
promptg get my-pr-review | llm
echo "Review code" | promptg prompt save code-review         # Save prompt
promptg get code-review | llm                               # Execute prompt

Ecosystem

• You are in promptg/cli: the reference implementation (CLI). CLI docs live in docs/.
• Format/spec: https://github.com/promptg/spec (canonical spec, schemas, implementer guide). PromptG documents use schemaVersion: "1".
• Conceptual model: https://github.com/promptg/spec/blob/main/spec/promptg-spec.md#2-core-concepts-and-rationale-informative
• Website: https://promptg.io (landing page, hosted schemas, pack mirror).

Routing:

• CLI bugs/features: https://github.com/promptg/cli/issues
• Spec/schema issues: https://github.com/promptg/spec/issues

One Prompt Source. Everywhere You Work.

PromptG stores prompts/templates as JSON in git and renders plain text you can use in LLM runners, CI, and editor extensions -- you choose.

• Stop Slack prompts: the same prompt works for devs and CI.
• No prompt drift: changes are reviewed, versioned, and reproducible.
• Templates + variables: one standard prompt, many contexts.
• Tool-agnostic: swap models/tools without rewriting prompts.

Documentation

• CLI Reference
• Format Specification (canonical spec)
• Vendored Schemas (offline copy)

Contributing

See CONTRIBUTING.md.

Support

• Bug reports and feature requests: https://github.com/promptg/cli/issues
• Format/spec questions: https://github.com/promptg/spec

Release Workflow

• Changes land via PRs to main.
• Releases are tagged from main as vX.Y.Z and published to npm as @promptg/cli.

License

Apache-2.0