Prompt Optimizer

/prompt-optimizer — Prompt Analysis Pipeline

Six-phase advisory skill that takes a user's draft prompt or task description and produces an optimized version with the right skills, agents, and workflow.

Advisory only — this skill never executes the task. It produces an optimized prompt the user can then run.

Usage

/prompt-optimizer "Add authentication to my SvelteKit app"
/prompt-optimizer                # Analyze the last user message
/prompt-optimizer --quick        # Short version for experienced users

The Six Phases

Phase 1: Detect Project Context

Gather project metadata without reading the full codebase:

# Quick project detection
cat CLAUDE.md 2>/dev/null | head -30
cat package.json 2>/dev/null | node -e "const p=require('/dev/stdin'); console.log(p.name, Object.keys(p.dependencies||{}).slice(0,10))"
ls -d src/ app/ lib/ pages/ routes/ components/ 2>/dev/null
git remote -v 2>/dev/null | head -1

Output: Tech stack, framework, project type, repo context.

Phase 2: Detect User Intent

Parse the prompt to identify:

• Primary goal: What does the user want to achieve?
• Implicit requirements: What's assumed but not stated?
• Scope boundaries: What should NOT change?
• Success criteria: How will we know it's done?

Phase 3: Assess Scope

Classify the task:

Phase 4: Match Hangar Components

Scan available skills and agents for the best match:

# List available skills and their descriptions
for skill in core/skills/*/SKILL.md; do
  head -5 "$skill" | grep "description:"
done

# List available agents
for agent in core/agents/*.md; do
  head -5 "$agent" | grep "description:"
done

Match criteria:

• Skill trigger keywords vs. task keywords
• Agent capabilities vs. task requirements
• Stack-specific skills if applicable
• Context modes (dev/research/review) if applicable

Phase 5: Identify Missing Context

What information would improve the prompt?

Phase 6: Generate Optimized Prompt

Produce two versions:

Full Version (Copy-Paste Ready)

Includes all context, constraints, skill references, and step-by-step approach. Ready to paste as a new prompt.

Quick Version (Experienced Users)

Key instructions only, assumes familiarity with Hangar tools.

Output Format

## Prompt Optimization Report

### Project Context
- **Framework:** SvelteKit 2 + Svelte 5
- **Stack:** Drizzle ORM + PostgreSQL
- **Context mode:** dev

### Intent Analysis
- **Goal:** Add user authentication
- **Implicit:** Session management, login/signup pages, protected routes
- **Scope:** Auth module only, don't touch existing routes
- **Success:** Users can register, login, access protected content

### Scope: LARGE (6+ files, architectural change)
**Recommended:** Use planner agent first, then TDD with tdd-guide

### Matched Components
| Component | Type | Relevance |
|-----------|------|-----------|
| Auth stack | Stack | High — has auth patterns for bcryptjs + sessions |
| security-scan | Skill | Medium — run after implementation |
| verification-loop | Skill | High — run before PR |
| tdd-guide | Agent | High — TDD for auth is critical |
| security-reviewer | Agent | High — auth needs security review |

### Missing Context (ask these first)
1. Custom auth or external provider (OAuth, Auth.js)?
2. Email verification required?
3. Role-based access control needed?

### Optimized Prompt (Full)

> I need to add custom authentication (bcryptjs + sessions) to my SvelteKit app.
>
> **Requirements:**
> - User registration with email + password
> - Login/logout with secure session cookies
> - Protected routes via server-side hooks
> - Password hashing with bcryptjs (12 rounds)
>
> **Approach:**
> 1. Use the Auth stack patterns from Hangar
> 2. Start with planner agent for implementation plan
> 3. Follow TDD (tdd-guide) for all auth logic
> 4. Run security-reviewer when done
> 5. Run /verify before PR
>
> **Constraints:**
> - No external auth providers
> - DSGVO-compliant (no tracking cookies)
> - Existing routes must not break

### Optimized Prompt (Quick)

> Add custom auth (bcryptjs + sessions) to SvelteKit.
> Use Auth stack patterns. TDD with tdd-guide. Security review after.

When to Use

• Unclear tasks — When you don't know which skills to use
• Large tasks — When the approach matters as much as the implementation
• New to Hangar — When learning which tools are available
• Complex prompts — When the task has many implicit requirements

What This Skill is NOT

• Not a prompt rewriter for AI conversations in general
• Not a prompt injection defense tool
• Not an auto-executor — it ONLY advises

Inspired by ECC's prompt-optimizer with Hangar's component-matching approach.