Claude Commands LibraryTransform Ideas into Precise Prompts

The Problem ​

You write a prompt. Claude guesses what you meant. You get something close but not quite right. You clarify. Claude tries again. Repeat.

This wastes time and tokens.

The Solution ​

Claude Commands Library ensures every prompt is perfect before execution:

# Before: Vague prompt leads to wrong implementation
"Add login to my app"

# After: Phase 0 asks what you actually need
/prompt-technical Add login to my app

# Claude asks:
# - Which authentication method? (JWT / OAuth / Session)
# - Which database stores users?
# - What UI framework for the login form?
# - Should I add password reset functionality?

# Result: Precise implementation matching your exact needs

Quick Start ​

1. Install ​

# Download and run installer
Invoke-WebRequest -Uri "https://raw.githubusercontent.com/Tadzesi/claude-ideas/main/install-claude-commands.ps1" -OutFile "install.ps1"
.\install.ps1

2. Use Commands ​

# Perfect any prompt (2 seconds)
/prompt Fix the authentication bug

# Technical analysis with agent assistance (5-30 seconds)
/prompt-technical Implement caching with Redis

# Deep multi-agent research (1-3 minutes)
/prompt-research Perform security audit of payment system

# Save your work
/session-end

3. Track Progress ​

Install the Enhanced Statusline to see real-time metrics:

■ my-project | ⎇ main | ████████░░ 45% | ● 27k/155k | ▶ 89k/15k | ◆ 3.2s

Command Overview ​

How It Works ​

Phase 0: Prompt Perfection ​

Every command runs your prompt through validation:

1. Recall - Load known facts from project profile
2. Detect - Language, type, and intent
3. Check - 6 completeness criteria, pre-filling from memory
4. Ask - Only truly unknown information
5. Structure - Transform into executable format
6. Approve - Wait for your confirmation

Hybrid Intelligence ​

Tasks are scored for complexity (0-50+):

• Simple (0-4): Fast inline validation
• Moderate (5-9): Optional agent assistance
• Complex (10-19): Automatic agent exploration
• Research (20+): Multi-agent deep analysis

Continuous Learning ​

The system tracks:

• Successful transformations
• Your modification patterns
• Common missing information
• Agent effectiveness

After 3+ occurrences, it suggests smart defaults.

What's New ​

v4.6 - Superprompting: Anti-Hallucination Contract ​

Every skill now has a built-in contract that prevents Claude from inventing facts:

HARD-GATE blocks — pre-flight checklist in every skill before any output:

- [ ] project-profile.md read this session
- [ ] No version numbers copied from templates — all from read files
- [ ] No file paths invented — all verified with Read or Glob

NEVER rules — explicit domain-specific prohibitions (e.g. /prompt-dotnet will never state a NuGet version without reading .csproj).

Chain-of-Thought REASONING block — Step 0.1 now requires:

REASONING
Prompt type: Task (Feature) — because input contains "add" + specific component name
Facts from project-profile.md: Stack is Node.js/VitePress, kebab-case conventions
Cannot determine: which specific file to edit

Core library v2.0 with Mermaid flowchart and Grounding Protocol.

v4.5 - Universal Skills ​

• /session-start and /session-end removed — replaced by Claude Code auto-memory
• /deploy and /new-stack are now universal — all server config read from personal-profile.md
• /prompt-dotnet and /prompt-react scan fallbacks added (monorepo support)
• Very High complexity threshold corrected: 20+ (was 15+)

v4.4 - .NET + React Project-Aware Skills ​

Two new project-aware skills that scan your codebase before asking any questions:

/prompt-dotnet - Reads .csproj, Program.cs, appsettings.json, Docker files:

PROJECT SCAN COMPLETE
Framework: net10.0
Architecture: Minimal API
Auth: JWT
ORM: EF Core | DB: PostgreSQL — ConnectionStrings:Default
Docker: yes — stack: myapp
Packages: FluentValidation, Serilog, MediatR

/prompt-react - Reads package.json, vite.config.ts, tsconfig.json:

PROJECT SCAN COMPLETE
React: 19.0.0 | TypeScript: yes, strict: yes
Router: React Router v6
State: Zustand
Data fetching: TanStack Query
Base path: /appname/

Both skills apply best practices for the detected stack automatically — no repeated questions.

v4.3 - Skills Format & Project-Aware Commands ​

Skills Format - Commands now use the native Claude Code Skills format:

---
name: prompt
description: Transform any prompt into an unambiguous, executable format...
argument-hint: "[your prompt or task description]"
---

Claude automatically suggests the right skill based on description. No more manual /prompt when Claude already knows what you need.

Project-Aware Startup - Every command loads project context first:

CONTEXT LOADED FROM PROJECT PROFILE
Project: My App v2.1 | Stack: Node.js + PostgreSQL
Platform: macOS | Branch: feature/auth
Recent work: Added JWT middleware (yesterday)

What else do you need for this task?

Zero Repeated Questions - The populated project-profile.md eliminates asking the same questions across sessions. Tech stack, preferences, structure - all pre-filled automatically.

v4.2 - Memory Recall & AI Fluency ​

Memory Recall - Commands remember your project between sessions:

Session 1: "What framework?" → "Express + TypeScript"
Session 2: Context: Express + TypeScript (from project profile) ✓

AI Fluency - Full 4Ds Framework integration:

• Delegation - Quick Delegation Check in all commands
• Description - 9 criteria completeness check
• Discernment - Post-Execution Evaluation with feedback loop
• Diligence - Responsibility reminders in Approval Gate

v4.1 - Skill Reflection ​

/reflect

• Signal detection (corrections, successes, edge cases)
• Priority-coded change proposals (HIGH/MED/LOW)
• Direct skill file modifications with approval

v4.0 - Predictive Intelligence ​

• Journey stage detection (exploring → implementing → debugging)
• Proactive warnings before you make mistakes
• Domain risk analysis (security, payment, performance)

v3.0 - Multi-Agent Research ​

• 2-5 specialized agents in parallel
• Iterative refinement with gap detection
• Comprehensive reports with file:line citations

Ready to Start? ​

Install Now → | Browse Commands → | Understand Architecture →