https://github.com/joseph0926/prompt-smith
prompt quality management for Claude Code
https://github.com/joseph0926/prompt-smith
claude-code claude-plugin claude-skills
Last synced: 28 days ago
JSON representation
prompt quality management for Claude Code
- Host: GitHub
- URL: https://github.com/joseph0926/prompt-smith
- Owner: joseph0926
- License: mit
- Created: 2026-01-02T00:35:39.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2026-01-10T10:34:19.000Z (about 1 month ago)
- Last Synced: 2026-01-10T22:19:26.155Z (about 1 month ago)
- Topics: claude-code, claude-plugin, claude-skills
- Language: JavaScript
- Homepage:
- Size: 656 KB
- Stars: 4
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- Changelog: CHANGELOG.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
- Code of conduct: CODE_OF_CONDUCT.md
- Codeowners: CODEOWNERS
- Security: SECURITY.md
Awesome Lists containing this project
README
# prompt-smith
> Prompt quality management for Claude Code
> **[한국어 문서 (Korean)](README.ko.md)**
[](https://opensource.org/licenses/MIT)
[](https://github.com/joseph0926/prompt-smith/releases)
**Core Features**:
- 8-Point Quality Check (LINT / BUILD / Review / Intercept)
- Eval Mode: A/B testing with test datasets
- Long Context: 200K token optimization
- Token Management: Cost estimation & context validation
- Structured Outputs: 100% JSON validity
## When to Use Which Mode?
| Situation | Recommended | Command |
| ----------------------------- | ----------- | ---------------------------------------------------------------- |
| "Is this prompt good enough?" | LINT | `/ps:lint ` |
| "Make this prompt better" | Review | `/ps:r ` |
| "Just fix it and run" | Intercept | `/ps:a ` |
| "Design a new prompt" | BUILD | `/ps:build ` |
| "Compare prompt variants" | Eval | See [Eval Mode](skills/prompt-smith/playbooks/eval/eval-mode.md) |
## Quick Start
### Commands at a Glance
| Command | Description | Best For |
| ------------------- | --------------------------------------------------------- | ----------------- |
| `/ps:r ` | **Review Mode** - Show improvements, await approval | Daily use (safe) |
| `/ps:lint ` | **LINT Mode** - Full diagnosis with test cases | Debugging prompts |
| `/ps:a ` | **Intercept Mode** - Auto-improve and execute immediately | Quick tasks |
| `/ps:build ` | **BUILD Mode** - Design from requirements | New prompts |
| `/ps:help [topic]` | **Help** - Show usage guide and command reference | Getting started |
**Quick Example**:
```
/ps:r Write a function to parse JSON
```
Shows score (3/10 → 8/10), changes, asks "Proceed? (y/n/e)"
---
### Option 1: Plugin Install (slash commands enabled)
1. In VS Code, type `/plugin` to open plugin terminal
2. Press `Tab` to navigate to "Add Marketplace"
3. Enter: `joseph0926/prompt-smith`
4. Press `Tab` to navigate to "Install Plugin"
5. Select `ps@prompt-smith`
### Option 2: Local Plugin (development)
```bash
git clone https://github.com/joseph0926/prompt-smith
claude --plugin-dir ./prompt-smith
```
### Option 3: Skill Only (natural language triggers)
```bash
git clone https://github.com/joseph0926/prompt-smith
cp -r prompt-smith/skills/prompt-smith ~/.claude/skills/
```
> **Note**: This method only enables natural language triggers (`use prompt-smith -r`), not slash commands.
### Usage
**Slash Commands (recommended)**:
```
/ps:r your prompt here # Review Mode (single line)
/ps:r your prompt here
with multiple lines # Review Mode (multiline)
/ps:a your prompt here # Intercept Mode
/ps:lint your prompt # LINT Mode
/ps:build requirements # BUILD Mode
```
> **Note**: Triple backticks (```) are optional. Plain text and multiline input are fully supported.
**Natural Language (alternative)**:
```
use prompt-smith -r Your prompt here
```
## Real Examples
### LINT Mode (Diagnose)
```
/ps:lint Analyze user feedback
```
**Output**: Score 3/10 → Top 3 issues → Improved prompt → 5 test cases
### Review Mode (Safe Improvement)
```
/ps:r Write a function to parse JSON
```
**Output**:
```
+----------------------------------------------------------+
| Score: 3/10 -> 8/10 (+5) |
+----------------------------------------------------------+
Changes:
- [+] Added: Software engineer role
- [+] Added: Error handling examples
- [~] Modified: Specified return format
Proceed? (y/n/e)
```
### Intercept Mode (Auto-fix)
```
/ps:a Summarize this document
```
**Output**: Auto-improved if +2 points possible → Executes immediately
## What's New in v3.x
### v3.3.1 (Latest)
| Feature | Description |
| ------------------------- | ------------------------------------------------------------------- |
| **Examples & TOCs** | Expanded examples and added TOCs to long references |
| **Structured Outputs** | Stronger guidance for schema mismatch and JSON mode caution |
### v3.3.0
| Feature | Description |
| ------------------------- | ---------------------------------------------------- |
| **ps:a Token Efficiency** | Minimal output: `[Prompt Smith] 활성화됨 (X→Y점)` |
| **ps:r Intent Capture** | AskUserQuestion for format/detail/constraints/goals |
### v3.2.0
| Feature | Description |
| ---------------------- | --------------------------------------------- |
| **Eval Mode** | A/B prompt comparison with test datasets |
| **Long Context** | 200K token optimization (chunking, placement) |
| **Token Management** | Cost estimation, context window validation |
| **Structured Outputs** | 100% JSON validity techniques |
### v3.0.0 - v3.1.0
| Feature | Description |
| ----------------------- | -------------------------------------------------- |
| **Progressive Loading** | Skills load on-demand |
| **Hooks** | Auto-LINT on prompt submit (optional) |
| **Agents** | 3 specialized: optimizer, reviewer, test-generator |
| **GitHub Actions** | CI/CD integration |
See [CHANGELOG.md](CHANGELOG.md) for full release notes.
## Modes
| Mode | Slash Command | Natural Language | Description |
| --------- | ------------------- | ----------------------- | --------------------------------- |
| Review | `/ps:r ` | `use prompt-smith -r` | Show improvements, await approval |
| Intercept | `/ps:a ` | `use prompt-smith -a` | Auto-improve and execute |
| LINT | `/ps:lint ` | `lint this prompt` | Diagnose existing prompts |
| BUILD | `/ps:build ` | `build a prompt for...` | Design from requirements |
## 8-Point Quality Check
| # | Dimension | Score | Note |
| --- | ---------------- | ----- | ------------ |
| 1 | ROLE | 0-2 | Base |
| 2 | CONTEXT | 0-2 | Base |
| 3 | INSTRUCTION | 0-2 | Base |
| 4 | EXAMPLE | 0-2 | Base |
| 5 | FORMAT | 0-2 | Base |
| 6 | SUCCESS_CRITERIA | 0-2 | Base (v2.7+) |
| 7 | STATE_TRACKING | 0-2 | Extended |
| 8 | TOOL_USAGE | 0-2 | Extended |
**Base Score**: Dimensions 1-6 (max 12 → normalized to 10)
**Extended Score**: All 8 dimensions (normalized)
## Documentation
### Getting Started
- [Quick Start](skills/prompt-smith/onboarding/quick-start.md)
- [First LINT](skills/prompt-smith/onboarding/first-lint.md)
- [First BUILD](skills/prompt-smith/onboarding/first-build.md)
### Mode Guides
- [LINT Mode](skills/prompt-smith/playbooks/lint/full-lint.md)
- [BUILD Mode](skills/prompt-smith/playbooks/build/build-mode.md)
- [Review Mode](skills/prompt-smith/playbooks/intercept/review-mode.md)
- [Intercept Mode](skills/prompt-smith/playbooks/intercept/intercept-mode.md)
- [Eval Mode](skills/prompt-smith/playbooks/eval/eval-mode.md)
### Advanced (v3.2.0)
- [Long Context Optimization](skills/prompt-smith/references/long-context-optimization.md)
- [Token Management](skills/prompt-smith/references/token-management.md)
- [Structured Outputs](skills/prompt-smith/references/structured-outputs.md)
- [Prompt Chaining](skills/prompt-smith/playbooks/prompt-chaining.md)
### Reference
- [8-Point Quality Checklist](skills/prompt-smith/references/quality-checklist.md)
- [Anti-Patterns](skills/prompt-smith/references/anti-patterns.md)
- [Technique Priority](skills/prompt-smith/references/technique-priority.md) (Anthropic recommended)
- [Claude 4.x Best Practices](skills/prompt-smith/references/claude-4x-best-practices.md)
## Troubleshooting
### Slash commands not working
1. Verify plugin: In VS Code, type `/plugin` → check installed list
2. Restart VS Code or reload Claude Code extension
3. Try reinstalling: Remove plugin, then reinstall
### Score seems wrong
- Extended items (STATE_TRACKING, TOOL_USAGE) only apply to multi-step or tool-using prompts
- If N/A, they're excluded from score calculation
## Platform
- **Claude Code** (CLI / VS Code extension)
## Contributing
We welcome contributions! See [CONTRIBUTING.md](CONTRIBUTING.md) for guidelines.
- [Report a bug](https://github.com/joseph0926/prompt-smith/issues)
- [Security issues](SECURITY.md)
## License
MIT License - see [LICENSE](LICENSE)