JSPM

@promptg/cli

0.9.2
  • ESM via JSPM
  • ES Module Entrypoint
  • Export Map
  • Keywords
  • License
  • Repository URL
  • TypeScript Types
  • README
  • Created
  • Published
  • Downloads 11
  • Score
    100M100P100Q87435F
  • License Apache-2.0

Prompts as code: versioned, shareable, standard.

Package Exports

  • @promptg/cli
  • @promptg/cli/dist/cli.js

This package does not declare an exports field, so the exports above have been automatically detected and optimized by JSPM instead. If any package subpath is missing, it is recommended to post an issue to the original package (@promptg/cli) to support the "exports" field. If that is not possible, create a JSPM override to customize the exports field for this package.

Readme

PromptG CLI

license npm version CI

npm install -g @promptg/cli

What is this

Less command line gymnastics. Version prompts, render with variables, insert code/files, pipe to any LLM.

# Before
cat prompt.txt | sed "s/{{lang}}/Python/g; s/{{code}}/$(cat app.py)/g" | ollama run mistral

# With promptg:
promptg get code-review --var lang=Python --var code@app.py | ollama run mistral

Who is this for?

You don't need this for a few prompts you paste into ChatGPT.

You might want this if you're managing lots of prompts, scripting prompts, inserting file contents, or keeping a team on the same prompt versions.

Getting Started

# Store once
echo "Refactor this {{language}} code for better {{focus}}. Suggest improvements in readability, performance, and maintainability. Output refactored code and explanations." | promptg prompt save code-refactor

# Or edit prompts in default editor
promptg prompt new code-refactor

# Use everywhere
promptg get code-refactor --var language=Python --var focus=Security | llm

# Version in Git, share with your team, run in CI
git add .promptg/ && git commit -m "Add code refactor prompt"

Examples

Reuse prompts globally without scattered text files

# Save once
echo "Write a clear PR description (summary, rationale, risks, testing) for this diff:\n\n{{diff}}" | promptg prompt save pr-desc
echo "Write an imperative commit message for this diff:\n\n{{diff}}" | promptg prompt save commit-msg

# Reuse forever (no temp prompt files)
git diff --staged > /tmp/diff.txt
promptg get pr-desc --var diff@/tmp/diff.txt | llm > PR.md
promptg get commit-msg --var diff@/tmp/diff.txt | llm | head -n 1

You run the same prompt with different inputs

# Compare outputs across models
for model in llama3.1 mistral deepseek; do
  promptg get review --var code@app.py | ollama run $model > "review-$model.txt"
done

# Run it on every file in a directory
for file in docs/*.md; do
  promptg get summarize --var content@"$file" | llm > "${file%.md}-summary.txt"
done

# Translate to multiple languages
for lang in Spanish French German Italian; do
  promptg get translate --var lang=$lang --var doc@README.md | llm > "README.$lang.md"
done

CI/CD Integration

# .github/workflows/pr-review.yml
- run: npm install -g @promptg/cli
- run: |
    git diff origin/main > diff.txt
    promptg get code-review --var diff@diff.txt | llm > review.md
    gh pr comment -F review.md

Your team needs the same prompts

# You: create a project store
promptg init

# Save your prompts and prompt templates, then commit to the project repo

git add .promptg/ && git commit -m "Add project prompts"

# New dev runs:
git clone your-repo
cd your-repo
promptg list

# They see:
# - code-review: Your team's code review standard
# - commit-msg: Commit message format
# - pr-summary: PR description template

Prompts live in .promptg/ and get committed with your code.

You share prompts with your team or users

# Build a pack (versioned collection)
promptg pack build dev-essentials --pack-version 1.0.0

# Distribute via URL or Git
promptg pack install https://promptg.io/dl/packs/promptg-pack-dev-essentials.json
promptg pack install ./promptg-pack-dev-essentials.json

Packs are versioned bundles of prompts and templates you can share via URL or file.

See PromptG Starter Packs for examples.

How it works

Prompts are JSON files

Lets us store structured data like default template values. You don't edit JSON, everything can be done in plain text through the CLI.

{
  "schemaVersion": "1",
  "kind": "prompt",
  "name": "review",
  "content": "Review this {{language}} code for security and performance."
}

PromptG renders them to plain text:

promptg get review --var language=TypeScript
# Output: Review this TypeScript code for security and performance.

Pipe the output anywhere. PromptG doesn't care which LLM you use.

Packs distribute collections

{
  "kind": "pack",
  "name": "team-essentials",
  "version": "1.2.0",
  "prompts": [...]
}

One command installs everything:

promptg pack install ./promptg-pack-team-essentials.json

The .promptg/ folder

promptg init  # Creates this structure
.promptg/
--- prompts/          # Your prompt instances
--- templates/        # Reusable blueprints
--- packs/            # Distribution bundles

Commit it to Git. New team members get the prompts when they clone the repo.

Quick reference

# Create & save
promptg init
promptg prompt new hello              # Create in $EDITOR
echo "content" | promptg prompt save hi

# Use
promptg get hello                     # Render to stdout
promptg get review --var lang=Go      # With variables
promptg get review --var code@app.go  # From file
promptg get review | llm              # Pipe to LLM

# Discover
promptg prompt list
promptg prompt show code-review
promptg template list --tag security

# Share
promptg pack build my-pack --pack-version 1.0.0
promptg pack install ./path/to/pack.json

# CI/CD
promptg validate  # Validate all files

Full docs: docs/CLI.md

Design principles

PromptG sticks to basics: it outputs plain text you can pipe anywhere (like cat but for prompts), works with any LLM without lock-in, and stores everything as git-friendly JSON files in .promptg/ so you can version, branch, or PR them like code. If the tool vanishes, your data is just portable files with an open spec, no drama.

FAQ

Can I still use ChatGPT/Claude/my-favorite-tool?

Yes. PromptG outputs plain text. Pipe it wherever you want:

promptg get review | pbcopy  # Copy to clipboard, paste in ChatGPT
promptg get review | ollama run mistral  # Use llm CLI
promptg get review | claude -p # Use Claude CLI

Do I need to learn JSON?

No. Editing JSON sucks - CLI lets you work in plain text. Prompt files are in JSON to store extra metadata.. (e.g. default template values)

echo "your prompt text" | promptg prompt save name
promptg prompt edit name  # Opens in your $EDITOR

You can hand-edit JSON if you want. You don't have to.

What if I want to stop using PromptG?

You have JSON files. Read them with cat, parse with jq, convert with a script. The spec is public. Migration to anything else takes minutes.

Does this work offline?

Yes. PromptG is a local CLI. No network, no accounts, no API keys. Your prompts live on your disk.

Documentation

Contributing

See CONTRIBUTING.md

License

Apache-2.0. See LICENSE