/prompt

The foundational command for prompt perfection. Project-aware, fast, and effective.

What's New in v4.0 (Anti-Hallucination Contract)

• HARD-GATE — pre-flight checklist verifies project-profile.md was read before any output
• NEVER rules — explicit prohibitions: no invented file paths, no hardcoded versions
• Chain-of-Thought REASONING block — every Step 0.1 requires grounding facts to source files
• STARTUP Section — loads project-profile.md, sessions.md, prompt-patterns.md first
• Zero repeat questions — never asks about tech stack, platform, or preferences again

Overview

Aspect	Details
Speed	~2 seconds
Version	v4.0
Format	.claude/skills/prompt/SKILL.md
Complexity	Simple
Agent	None
Best For	Quick fixes, simple questions, single-file changes

Usage

/prompt [your prompt]

Claude also auto-invokes this skill when it detects you want to refine or clarify a task.

What Happens at Startup

Before any analysis, /prompt loads your project context:

CONTEXT LOADED FROM PROJECT PROFILE

Project: Your App v2.1
Stack: Node.js, Express, PostgreSQL
Platform: macOS, bash
Preferences: TypeScript strict, 2-space indent, no auto-commit
Recent: Added JWT auth middleware (from sessions.md)
Patterns: "authentication" → auto-include security checklist

What I still need for this task: [only truly missing items]

This means you never need to explain your stack, platform, or preferences again.

Examples

Bug Fix (Project-Aware)

/prompt Fix the null pointer error in UserService.getById()

With project context loaded, output:

CONTEXT LOADED
Project: Your App | Stack: Node.js + TypeScript + PostgreSQL

Phase 0 Analysis:
- Type: Bug Fix
- Pre-filled: TypeScript, Express, your DB schema (from profile)
- Missing: What error message are you seeing?

Question: What does the error say exactly?

No questions about stack, framework, or project structure — Claude already knows.

Simple Question

/prompt How do I add environment variables in Next.js?

Output:

CONTEXT LOADED
Project: My App | Stack: Next.js + TypeScript

Phase 0: Complete (100%)
[Claude proceeds with answer, using your specific Next.js version from profile]

Code Change

/prompt Add input validation to the email field in RegisterForm.tsx

Output:

CONTEXT LOADED
Project: My App | Stack: React + TypeScript | Pattern: use Zod for validation

Phase 0: Complete (90%)
Note: Following existing Zod validation pattern from LoginForm.tsx (from profile)

Proceeding with email validation using Zod, consistent with your codebase.

How It Works

STARTUP: Load Project Context

Always runs first, before any analysis:

1. Read .claude/memory/project-profile.md — full project facts
2. Read last 3 sessions from .claude/memory/sessions.md — recent work
3. Read .claude/memory/prompt-patterns.md — learned patterns

Display brief "CONTEXT LOADED" summary, then proceed.

Step 0.1: Initial Analysis

Detects:

• Language: English, Slovak, etc.
• Type: Task, Question, Bug Fix, Feature, Refactoring, Config, Docs
• Intent: What you're trying to achieve
• Scope: This project or another project?

Step 0.2: Project-Aware Completeness Check

Uses loaded project context to pre-fill criteria. Only checks for task-specific unknowns:

Already known from project profile (never asked again):

• Environment, platform, OS
• Tech stack and framework versions
• Code conventions and style preferences
• Project file structure

Checked per task (may be missing):

Criterion	Question
Goal	What exactly do you want to achieve?
Scope	Which specific file or component?
Requirements	Specific needs beyond defaults?
Expected Result	How will you verify success?

Step 0.3: Targeted Clarification

Only asks for genuinely missing information:

What I already know: [from project profile]
What I need for this task:

1. [Critical] Which specific file? (detected: UserService.ts, UserController.ts)
2. [Important] What should happen when user is null? Return null or throw?

When multiple approaches exist, presents options with a recommendation:

APPROACHES:

Option 1: Return null (simple, caller handles)
  Pros: Non-throwing, flexible  Cons: Caller must always check

Option 2: Throw UserNotFoundError (explicit)  ⭐ RECOMMENDED
  Pros: Clear contract, consistent with ErrorHandler  Cons: Caller must catch

Select: 1, 2, or describe your preference

Step 0.4-0.5: Perfect Prompt Output

PERFECTED PROMPT

Goal: Fix null pointer in UserService.getById() with proper error handling

Context (from project profile):
- Stack: Node.js + TypeScript + PostgreSQL/Prisma
- Pattern: Throws typed errors (ErrorHandler.ts)
- Conventions: async/await, strict null checks

Scope:
- File: src/services/UserService.ts (line 42)
- Pattern to follow: ProductService.getById() (already throws NotFoundError)

Requirements:
1. Add null check after database query
2. Throw UserNotFoundError with descriptive message
3. Follow existing pattern in ProductService

Expected Result: No more null pointer. Callers receive typed error.
Verify: Run existing tests in __tests__/UserService.test.ts

Changes Made:
- Project context auto-filled from project-profile.md
- Pattern reference added from profile

Step 0.6: Approval Gate

PERFECTED PROMPT READY

Type: Bug Fix | Scope: UserService.ts | Confidence: High
Project context: Auto-loaded from profile

Approve?
- y / yes  - Execute
- n / no   - Cancel
- modify [description] - Adjust
- explain [part]       - Explain a decision
- options              - Show alternatives

Skills Format (v4.6)

The /prompt skill uses native Claude Code Skills format:

Location: .claude/skills/prompt/SKILL.md

Frontmatter:

---
name: prompt
description: Transform any prompt into an unambiguous, executable format.
             Use when the user wants to refine or perfect a prompt...
argument-hint: "[your prompt or task description]"
---

The description field allows Claude to suggest this skill automatically when relevant. You can still invoke it directly with /prompt.

Old .claude/commands/prompt.md continues to work unchanged.

When to Use /prompt

Good For

• Single-file bug fixes
• Simple questions
• Quick code changes (1-2 files)
• Adding comments or documentation
• Small refactoring tasks
• Any task where you have clear scope

Not Ideal For

Need	Better Command
Multi-file changes	/prompt-hybrid
Pattern detection across codebase	/prompt-hybrid
Architecture decisions	/prompt-technical
Deep analysis or research	/prompt-research
Writing articles or documentation	/prompt-article

Comparison

Feature	/prompt	/prompt-hybrid
Speed	~2s	2-30s
Project context	✅ Full	✅ Full
Agent support	No	Yes
Caching	No	Yes
Complexity detection	Basic	Advanced
Pattern matching	From profile	Active agent scan
Multi-file context	From profile	Full exploration

AI Fluency Features

Quick Delegation Check (Step 0.11)

Before processing, validates task appropriateness:

Delegation Check:
- Task: Suitable for AI ✓
- Capability match: Code analysis (excellent) ✓
- Responsibility: User remains responsible ✓

Post-Execution Evaluation (Step 0.7)

After completion, prompts for feedback:

Quick Evaluation

How was this output?
- good    — Accurate, appropriate ✅
- partial — Mostly good, minor adjustments ⚠️
- wrong   — Significant issues ❌
- explain — Show me your reasoning 🔍

Diligence Reminder

In approval gate: you remain responsible for verifying AI output before use.

Tips

Be Specific About Scope

# Less effective
/prompt Fix the bug

# More effective
/prompt Fix the TypeError on line 42 of UserService.ts where user.email is undefined

Reference Existing Patterns

/prompt Add validation to handleSubmit in ContactForm.tsx, similar to LoginForm.tsx

Claude will find the pattern in your profile or from the mentioned file.

State Expected Behavior

/prompt The submit button in ContactForm should show a loading spinner
        but currently does nothing when clicked. Fix it.

Version History

Version	Date	Changes
v4.0	Apr 2026	Anti-Hallucination Contract: HARD-GATE, NEVER rules, REASONING block
v3.0	Mar 2026	Skills format, STARTUP section, project-aware, zero repeat questions
v2.1	Jan 2026	AI Fluency: Delegation Check, Post-Execution Evaluation, Diligence
v2.0	Dec 2024	Migrated to unified library system
v1.0	Oct 2024	Initial standalone implementation

Related

• /prompt-hybrid - When you need agent assistance
• /prompt-technical - For implementation planning
• /reflect - Improve skills after a session
• Skills Format - How the new format works
• Phase 0 - The validation flow