context-not-control — Автоматизация контекста и доступа для AI-агентов

--- name: context-not-control description: Enable "Context, not Control" workflow - clarify requirements through multi-turn dialogue, reduce rework, and execute with appropriate permission levels. Use when users want AI to take more autonomy, need help clarifying vague requirements, or want to establish trust-based collaboration patterns. Supports three permission levels (Master/Collaborative/Assistant) and automatic context management. --- # Context, not Control A skill that transforms how you work with AI - from micromanaging every step to providing context and letting AI make decisions. Inspired by the "Context, not Control" philosophy from the OpenClaw community. ## Core Philosophy **Traditional approach**: You tell AI exactly what to do, step by step. **This approach**: You tell AI what you want to achieve, AI figures out how. The key insight: AI works best when you give it rich context about your goals, constraints, and preferences - then trust it to execute within appropriate boundaries. ## When to Use This Skill - Starting a new project with vague requirements - Want to reduce back-and-forth and rework - Need AI to take more initiative and make decisions - Want to establish clear permission boundaries - Transitioning from "micromanaging AI" to "trusting AI" ## Quick Start ### 1. Initialize Your Context Run the initialization script to set up your project context and permission level: ```bash python scripts/init_context.py ``` This creates: - `PROJECT.md` - Your project context (goals, constraints, preferences) - `PERMISSION_CONFIG.yaml` - Your permission boundaries ### 2. Set Your Permission Level Choose one of three levels: **Level 1 - Master Mode** (Full autonomy) - AI makes all technical decisions - Only confirms: spending money, public messages, deleting databases - Best for: High trust, high risk tolerance **Level 2 - Collaborative Mode** (Balanced, recommended) - AI executes most tasks autonomously - Confirms: money, public messages, important deletions, system changes - Best for: Most users, balanced control **Level 3 - Assistant Mode** (High control) - AI provides suggestions and code - Confirms: All operations before execution - Best for: New users, low risk tolerance, learning mode ### 3. Start with Requirements Instead of detailed specifications, start with what you want: ``` "I need a team chat tool" ``` AI will ask clarifying questions: - Who is this for? - What's the core use case? - Any similar products to reference? - Technical constraints? - Time/budget limits? ### 4. Iterate and Execute AI clarifies → You answer → AI confirms understanding → You approve → AI executes All clarified requirements are saved to `PROJECT.md` for future reference. ## How It Works ### Requirement Clarification Framework When you provide a vague requirement, AI uses a structured approach: 1. **Understand the domain** - What type of project is this? 2. **Identify the user** - Who will use this? 3. **Clarify the goal** - What problem does this solve? 4. **Establish constraints** - Technical, time, budget limits? 5. **Set success criteria** - What does "done" look like? 6. **Confirm understanding** - Repeat back what you heard See `references/clarification-framework.md` for detailed question templates. ### Permission System The skill automatically checks permissions before executing operations: ```python # Example: AI wants to delete a file if permission_check('delete_file', user_permission_level): # Ask user for confirmation else: # Execute directly ``` Customize your red/yellow/green lines in `PERMISSION_CONFIG.yaml`. ### Context Management All clarified requirements are automatically saved to `PROJECT.md`: - Project goals and constraints - Technical stack decisions - Success criteria - Permission level - Iteration history This context is loaded in future conversations, eliminating repeated questions. ## Permission Levels in Detail ### Level 1: Master Mode **Philosophy**: Maximum autonomy, minimum interruption **AI can do without asking**: - Write, test, and deploy code - Install dependencies and tools - Modify configurations - Create/update files - Make architectural decisions - Research and learn new technologies **AI must confirm**: - Spending money (API calls, services, domains) - Sending public messages (emails, tweets, posts) - Deleting databases or critical data - Restarting production services **Best for**: Experienced users who trust AI and can handle mistakes ### Level 2: Collaborative Mode (Default) **Philosophy**: Trust but verify on important operations **AI can do without asking**: - Write and test code - Create/update files - Research and documentation - Install development dependencies - Run tests and checks **AI must confirm**: - Spending money - Sending any external messages - Deleting important files/data - Modifying system configurations - Restarting services - Installing system-level packages **Best for**: Most users, balanced approach ### Level 3: Assistant Mode **Philosophy**: AI suggests, you decide **AI can do without asking**: - Provide suggestions and explanations - Show code examples - Research information **AI must confirm**: - All file operations - All code execution - All installations - All external calls **Best for**: New users, learning mode, high-stakes environments ## Examples See `references/examples.md` for detailed examples including: - Building a chat application from vague requirements - Migrating a legacy system with unclear scope - Creating automation tools with evolving needs See `assets/EXAMPLE_DIALOG.md` for sample conversations. ## Customization ### Custom Permission Rules Edit `PERMISSION_CONFIG.yaml` to define your own boundaries: ```yaml permission_level: 2 custom_red_lines: - deploy_to_production - modify_database_schema - send_customer_emails custom_yellow_lines: - install_npm_packages - modify_env_files # Everything else is green (no confirmation needed) ``` ### Project Templates Create custom templates in `assets/` for recurring project types: - `PROJECT_TEMPLATE_WEBAPP.md` - `PROJECT_TEMPLATE_API.md` - `PROJECT_TEMPLATE_AUTOMATION.md` ## Troubleshooting See `references/troubleshooting.md` for common issues: - AI asking too many questions - AI not asking enough questions - Permission checks too restrictive/loose - Context not being saved properly ## Scripts Reference ### `init_context.py` Initialize project context and permission config ```bash python scripts/init_context.py [--project-name NAME] [--permission-level 1|2|3] ``` ### `clarify_requirement.py` Run requirement clarification dialogue ```bash python scripts/clarify_requirement.py "I need a chat app" ``` ### `permission_check.py` Check if an operation requires confirmation ```bash python scripts/permission_check.py --action delete_file --level 2 ``` ### `update_context.py` Update project context with new information ```bash python scripts/update_context.py --add-goal "Support 1000 concurrent users" ``` ## Philosophy: Three Modes of AI Usage ### Mode 1: Paintbrush (Micromanagement) - You specify every detail - AI is a tool that executes exactly what you say - Upper limit: Your expertise ### Mode 2: Employee (Delegation) - You assign tasks with some guidance - AI follows your preferred patterns - Still requires oversight ### Mode 3: Master (Autonomy) - You set goals and constraints - AI makes decisions and executes - You review outcomes, not process This skill helps you transition from Mode 1 → Mode 3 at your own pace. ## Credits Inspired by the "Context, not Control" philosophy discussed in the OpenClaw community, particularly the experiences shared by contributors who achieved remarkable results by trusting AI with more autonomy. ## Version 1.0.0 - Initial release