An agentic AI skill that transforms raw, vague prompts into professional-grade, context-rich instructions — like having a prompt engineer on your team.
The moment you invoke /enhance-prompt, it doesn't just rewrite your words. It steps back and scans your codebase to understand what you're actually building.
It pulls your tech stack, dependencies, file structure, patterns, and constraints — then injects that real context into a structured, professional prompt.
Every enhanced prompt includes a before/after quality score (out of 35), risk assessment, difficulty estimate, and a clean copy-ready version.
Install once. Works across Claude Code, Cursor, Codex, and OpenCode.
/plugin marketplace add VoDaiLocz/enhance-prompt
/plugin install enhance-prompt@enhance-prompt
In a Cursor Agent conversation:
/plugin-add enhance-prompt
Tell Codex:
Fetch and follow instructions from https://raw.githubusercontent.com/VoDaiLocz/Enhance-Prompt/main/.codex/INSTALL.md
Tell OpenCode:
Fetch and follow instructions from https://raw.githubusercontent.com/VoDaiLocz/Enhance-Prompt/main/.opencode/INSTALL.md
cp -r skills/enhance-prompt/ ~/.gemini/antigravity/skills/enhance-prompt/Start a new session and type:
/enhance-prompt Fix the login page
The agent will automatically invoke the skill, scan your codebase, and produce a professional enhanced prompt.
/enhance-prompt <your raw prompt>
Examples:
/enhance-prompt Fix the bug in auth
/enhance-prompt Add caching to the API
/enhance-prompt Refactor the user service
/enhance-prompt optimize the database
Raw prompt: Fix the bug in auth
Enhanced:
## Context
Next.js 14 App Router + TypeScript. Auth via `next-auth` v5 in `src/lib/auth.ts`.
Prisma adapter. Session breaks after form submit in `src/app/(auth)/login/page.tsx`.
## Task
1. Debug `signIn()` callback at `auth.config.ts:25`
2. Fix session persistence in `SessionProvider` wrapper
3. Verify redirect logic in `middleware.ts` callbacks
## Constraints
- ✅ Follow existing patterns in `src/middleware.ts`
- ✅ TDD: write failing test first
- ❌ Don't modify Prisma schema
## Verification
- [ ] Valid login → redirects to `/dashboard`
- [ ] Invalid login → shows error toast
- [ ] Session persists after page refreshScore: 6/35 → 34/35
- 7-Layer Enhancement — Clarity → Specificity → Context → Structure → Constraints → Output Format → Verification
- Auto-Category Detection — automatically identifies prompt intent (Bug Fix, Feature, Performance, etc.) and selects the best template
- Prompt Chain Decomposition — breaks complex multi-concern prompts into focused sequential sub-prompts
- Enhancement Iteration Mode — targeted refinement with tracked scoring deltas across up to 3 iterations
- HARD-GATE — codebase scan is mandatory, no exceptions
- Iron Law — No enhanced prompt without codebase context
- RED FLAG guards — per-layer checklist to catch incomplete enhancements
- L1 — Project overview (package.json, tsconfig, .env.example)
- L2 — Architecture scan (imports, naming, routing patterns)
- L2.5 — Monorepo detection (Turborepo, Nx, Lerna, PNPM workspaces)
- L2.5 — Framework detection (Next.js App/Pages Router, FastAPI, Django, Express, NestJS)
- L3 — Deep scan (function signatures, types, test patterns)
Every enhanced prompt includes:
| Section | Description |
|---|---|
| 🏷️ Detected Category | Auto-detected intent with confidence level |
| 📊 Enhancement Summary | Before/After score table (7 dimensions, /35) |
| 🔴 Original Prompt | The raw input |
| 🟢 Enhanced Prompt | Context + Task + Constraints + Output + Verification |
| ⏱️ Difficulty Estimate | Complexity, time, risk, files affected |
| High/Medium/Low risks with mitigation | |
| 🔀 Alternative Approaches | Options with pros/cons |
| 🔗 Prompt Chain | Decomposed sub-prompts for multi-intent inputs |
| 📋 Copy-Ready Prompt | Clean version, paste anywhere |
| # | Template | Use Case |
|---|---|---|
| 1 | 🐛 Bug Fix | Debugging with root cause investigation |
| 2 | ✨ New Feature | Adding functionality with TDD |
| 3 | ⚡ Performance | Optimization with measurable targets |
| 4 | 🔄 Refactor | Structural improvements, zero behavior change |
| 5 | 🔒 Security Audit | OWASP Top 10 scanning |
| 6 | 📝 Documentation | JSDoc, README, inline comments |
| 7 | 🧪 Testing | Unit, integration, coverage targets |
| 8 | 🎨 UI/Styling | Components, responsive, WCAG accessible |
| 9 | 🔌 API Integration | External services with retry/backoff |
| 10 | 🚀 DevOps | CI/CD, deployment, health checks |
| 11 | 🗃️ Database | Migrations, reversible, indexed |
| 12 | 🔍 Deep Scan | Root cause analysis, null/async/security bug hunting |
| 13 | 👀 Code Review | Security (OWASP), performance (O(n²), N+1), pattern compliance |
- Evidence over guessing — scan the codebase, don't assume the tech stack
- Auto-detect over manual selection — signal matching picks the right template
- One concern per prompt — multi-intent prompts get decomposed into focused chains
- Specificity beats length —
src/auth/login.ts:42> paragraph of description - Systematic over ad-hoc — 7-layer framework, not feel-good rewrites
- Prove improvement — scoring table shows the enhancement worked
- Iterate with precision — targeted refinement, not full rewrites
.claude-plugin/
├── plugin.json # Claude Code plugin manifest
└── marketplace.json # Claude Code marketplace config
.cursor-plugin/
└── plugin.json # Cursor plugin manifest
.codex/
└── INSTALL.md # Codex installation guide
.opencode/
└── INSTALL.md # OpenCode installation guide
commands/
└── enhance-prompt.md # /enhance-prompt slash command
skills/enhance-prompt/
├── SKILL.md # Main skill orchestrator
└── references/
├── prompt-patterns.md # 7-Layer Enhancement Framework
├── codebase-analysis.md # Multi-level codebase scanning
├── output-format.md # Structured output template
├── prompt-library.md # 6 core prompt templates
├── prompt-library-extended.md # 7 extended templates
├── auto-detection.md # Auto-category detection & template selection
├── prompt-chain.md # Prompt chain decomposition
└── iteration-mode.md # Enhancement iteration & refinement
RELEASE-NOTES.md
- Fork this repository
- Create a branch for your improvement
- Follow the skill format in
skills/enhance-prompt/SKILL.md - Submit a PR
MIT