Install
Claude Code with short commands such as /real-audit:
git clone https://github.com/giraybatiturk/skills.git "$HOME/Developer/skills"
bash "$HOME/Developer/skills/scripts/install-local.sh" "$HOME/.claude/skills"
Codex and other agents, or an editable copy:
npx skills@latest add giraybatiturk/skills -g -a codex -s '*' -y
Skip the clone line if the repository already exists, then run the installer. Do not enable the plugin at the same time. Open a new session after installing.
Get oriented
Start with the smallest workflow that matches the work. Real Audit progressively loads Product, Module, Design, Monetization, Security, Quality or Performance only when relevant.
- route
/real-startchoose the smallest workflow for unclear or mixed work - plan
/real-planshape user stories, requirements, architecture and acceptance criteria - audit
/real-auditassess the relevant product, UX, security, quality and performance modes - research
/real-researchresolve current external facts from primary sources
Natural-language selection depends on the client and context. Use the explicit command when routing must be deterministic.
The skill set
Product workflows
Route unclear product work: where to start, nereden başlayalım, or mixed planning, audit and research requests. Select the smallest Real workflow; skip routing for simple edits or already-approved implementation.
Plan new features and product changes before implementation: user stories, requirements, architecture and acceptance criteria. Use for planla or an unsettled idea; reuse existing approval and skip renewed planning for approved edits.
Audit existing code, modules, screens or products. Use for review, denetle, audit, missing behavior, UX quality, readiness or performance checks. Report evidence-backed defects without designing their remedy; apply fixes only when separately authorized. Covers produced visual deliverables (catalogue, deck, poster, artboard) in Deliverable mode.
Engineering
Research questions using primary sources: araştır, official documentation, API behavior, competitors or provider limits. Separate verified facts from inference and record sources; do not implement the researched change.
All four can be selected automatically from natural language or invoked directly. Automatic selection is client- and context-dependent.
Source:
github.com/giraybatiturk/skills
opens in a new tab · v0.9.0 · 2026-09-21