ctrl+shft: An Intelligent Development Infrastructure for AI Programming Agents

ctrl+shft is a complete configuration management system for AI programming agents. It addresses core pain points like context drift and configuration inconsistencies of AI agents in multi-device environments through unified instruction synchronization, skill management, secure credential protection, and autonomous workflows. The project is maintained by arndvs, hosted on GitHub, and licensed under MIT.

With the popularity of AI programming agents like Claude Code and GitHub Copilot, developers face issues such as context degradation (repeated answers, detail loss), cross-device configuration inconsistencies, sensitive information leaks, and resource waste from loading irrelevant rules. As a complete infrastructure, ctrl+shft solves these pain points through structured management of instructions, skills, rules, and secure credentials.

ctrl+shft uses a two-layer architecture: ctrl (structural layer, managing instructions, skills, rules, secure credentials, and context) and shft (autonomous loop layer, executing task queues). The core philosophy is "single repository, multi-end synchronization": clone the repository to ~/dotfiles, then establish symbolic links via bootstrap.sh, and git pull automatically updates configurations across all devices.

Progressive Context Loading: Identifies tech stacks via detect-context.sh, loads in four hierarchical levels: T1 (global), T2 (context-gated), T3 (path-gated), T4 (skill-triggered), to avoid interference from irrelevant rules.

Three-Layer Secure Credential Protection: .env.agent (non-sensitive), .env.secrets (sensitive isolation), AFK temporary tokens (expire in 1 hour); scripts inject sensitive information securely, and rejection rules prevent command leaks.

Skill System: Built-in skills like grill-me (requirements understanding), write-a-prd (PRD writing), architect (solution planning), etc., which can be chained into workflows, with breakpoints persisted to the working directory.

Sub-Agents: Independent system prompts, tool restrictions, and model preferences; preset model variants like researcher (Sonnet/Opus/Haiku), code-reviewer, security-auditor.

Path-Limited Rules: Files in the rules directory limit their scope via the paths field (e.g., test-conventions applies to test files), avoiding bloated global rules.

Quick Start: After forking the repository, clone it to ~/dotfiles and execute bootstrap.sh (an idempotent cross-platform script that creates symbolic links, configures the shell, initializes secrets, etc.); AFK mode requires configuring a GitHub App to generate short-term tokens.

ctrl+shft is suitable for scenarios like multi-device development (consistent experience), team collaboration (shared norms), automated workflows (automatic Issue handling), security compliance (layered credentials), skill accumulation (reusable experience modules), etc., providing a solid infrastructure for AI-assisted development.

ctrl+shft represents a new paradigm for AI programming agent management, treating agents as development partners that need careful configuration. Through systematic instruction management, intelligent context loading, strict security control, and a rich skill ecosystem, it solves current pain points and lays the foundation for future complex AI collaboration.