ftitos-claude-code

ftitos-claude-code

A Claude Code configuration that runs a project from first commit to deployed and monitored — with only 3 human checkpoints.

---

What it does

Claude Code ships with a good base, but no structure around how it plans, reviews, tests, or ships. This project provides that structure: 21 specialist agents, 67 skills, 15 behavioral rules, and a lifecycle hook system — all installed into ~/.claude/ with one command.

Start a feature with /project:sprint validate. Claude validates scope, writes a plan, gets your approval (Gate 1), builds autonomously, runs 7 parallel code reviewers, executes tests, asks you to do UAT (Gate 2), ships, then confirms the push (Gate 3). The three gates are the only manual steps.

Built for developers who want Claude Code to behave like a disciplined engineering team, not a chat assistant.

---

Quick Start

# Clone
git clone https://github.com/nassimbf/ftitos-claude-code.git
cd ftitos-claude-code

# Install (copies agents, skills, rules, hooks, commands into ~/.claude/)
./install.sh

# Verify everything installed correctly
npm run doctor

Expected output from doctor:

[OK] 21 agents loaded
[OK] 67 skills loaded
[OK] 15 rules loaded
[OK] 14 hooks registered
[OK] 11 commands available
All checks passed.

Requirements: Node.js 18+, Claude Code CLI. Zero external npm dependencies.

---

How it works

flowchart LR
    V[VALIDATE] -->|auto| P[PLAN]
    P -->|"Gate 1: approve plan"| B[BUILD]
    B -->|auto| R[REVIEW]
    R -->|auto| T[TEST]
    T -->|"Gate 2: UAT"| S[SHIP]
    S -->|"Gate 3: confirm push"| M[MONITOR]

Three gates. Everything between them runs without intervention.

Gate	Trigger	Your action
Gate 1	PLAN.md generated with scope, tasks, and risks	Review, type approve
Gate 2	Build complete, review cleared, tests pass	Test the product, type approved
Gate 3	Release packaged, changelog generated	Confirm push, type ship

Start: /project:sprint validate — Check status: /project:status

See guides/architecture.md for the full system design.

---

Key features

• Review Army — 7 parallel specialist reviewers (security, performance, migrations, API contracts, testing, maintainability, UX) dispatched based on what changed in the diff. Each has a focused checklist and a confidence threshold.
• Review Council — when a specialist flags a CRITICAL finding, two independent reviewer agents validate it before it can block ship. Neither sees the other's verdict (anti-anchoring). Two dismissals downgrade the finding; a split escalates to you.
• GateGuard — blocks Edit and Write calls on any file that has not been Read in the current session. Prevents modifying code you have not looked at.
• 4-brain memory — four MCP engines routed via /brain <query>: GBrain (people/companies), Graphify (concepts/knowledge), GitNexus (code structure/blast radius), Engram (session memory/cross-session recall).
• CARL decision log — every architectural decision is recorded with domain scoping and recall keywords, queryable across sessions, promotable to permanent rules.
• Anti-slop enforcement — 17-item behavioral rule that blocks generic AI output patterns (gradient backgrounds, copy clichés, meaningless abstractions) before they reach output.
• Context survival — auto-checkpoint before compaction, auto-recovery after. Session context survives claude --compact.
• 21 specialist agents — architect, security-reviewer, tdd-guide, debugger, refactor-cleaner, anti-patterns detector, and 15 others. Each has a focused role and dispatch conditions.
• 67 skills — covering TDD, architecture patterns, agentic engineering, deployment, context management, design intelligence, frontend design (impeccable, taste-skill), plan writing/execution (superpowers), and language-specific conventions.
• 15 behavioral rules — coding style, git workflow, testing discipline, security pre-commit checklist, agent usage, and more. All enforced via hooks and rule files.
• Design system templates — brand-inspired DESIGN.md files (Stripe, Vercel, Notion, Claude) plus skillui CLI for extracting design systems from any URL.

---

Documentation