Professional AI development environment with autonomous workflows, specialized agents, and intelligent coding standards for Claude Code and Cursor.
Transform how you work with AI: from manual prompting to autonomous task execution, from generic responses to specialized agent collaboration, from scattered configs to unified standards.
Autonomous workflows: One command (/autotask) takes your task from description to
PR-ready state - worktree creation, implementation, validation, bot feedback handling,
all autonomous.
Named specialist agents: Work with Dixon (debugging), Ada (development), Phil (UX), Rivera (code review), and Petra (prompts) - each an expert in their domain.
Plugin marketplace: Install curated configurations for Python, React, Django, testing, git workflows, and more through Claude Code's plugin system.
LLM-optimized standards: Coding rules and patterns designed for AI comprehension, not just human reading.
Personality options: Choose how AI communicates - from Sherlock's analytical precision to Samantha's warm encouragement to Ron Swanson's minimalist directness.
Add this marketplace and install what you need:
/plugin marketplace add https://github.com/TechNickAI/ai-coding-config
/plugin install dev-agents # Dixon, Ada, Phil, and more
/plugin install code-review # Rivera and architecture audits
/plugin install python # Python standards and patterns
/plugin install personality-samantha # Warm, encouraging communicationBrowse available plugins:
/plugin search ai-coding-configRun from any project:
curl -fsSL https://raw.githubusercontent.com/TechNickAI/ai-coding-config/main/scripts/bootstrap.sh | bashThen in Cursor:
@ai-coding-config set up this project
Interactive setup command works everywhere:
/ai-coding-config
The /autotask command handles complete feature development autonomously:
/autotask "add OAuth2 authentication with email fallback"What happens automatically:
- Task analysis - Evaluates complexity, creates structured prompt if needed
- Worktree isolation - Fresh environment on feature branch
- Intelligent execution - Deploys right combination of specialist agents
- Adaptive validation - Review intensity matches task risk and complexity
- PR creation - Proper commit messages, comprehensive PR description
- Bot feedback handling - Autonomously addresses automated review comments
- Ready for merge - All checks passing, waiting for your approval
Your involvement: Describe the task (~30 seconds), review the PR when ready, merge when satisfied.
Typical completion time: 15-30 minutes from task description to merge-ready PR.
See optimal-development-workflow.md for the complete philosophy and implementation.
When you install agent plugins, you gain access to specialized AI collaborators:
Dixon (dev-agents:debugger) Root cause analysis and debugging. Doesn't just fix
symptoms - finds the actual problem through systematic investigation.
Ada (dev-agents:autonomous-developer) Primary development work. Reads all project
standards, implements features, writes comprehensive tests, follows your patterns.
Phil (dev-agents:ux-designer) User experience review. Validates user-facing text,
checks accessibility, ensures consistent UX patterns.
Rivera (code-review:code-reviewer) Architecture and security review. Validates
design patterns, identifies security issues, suggests improvements.
Petra (dev-agents:prompt-engineer) Prompt optimization. Crafts effective prompts
for AI systems, improves clarity and specificity.
Plus: Architecture Auditor, Test Engineer, and Commit Message Generator.
Agents are used intelligently based on task requirements - no forced patterns, just the right specialist at the right time.
- python - Python development standards, pytest patterns, Celery tasks, type hints
- react - React component patterns, hooks, TypeScript integration
- django - Django models, templates, management commands, ORM patterns
- dev-agents - Dixon, Ada, Phil, Petra, and more specialist agents
- code-review - Rivera, architecture audits, test engineering
- git-commits - Commit standards, PR workflows, semantic messages
- code-standards - Universal naming, style, user-facing language
Eight distinct communication styles - see Personalities below.
Browse in Claude Code:
/plugin search ai-coding-configOr explore plugins/ directory directly.
/ai-coding-config- Interactive setup wizard for any project/plugin install <name>- Install specific plugin bundle/load-cursor-rules- Load task-relevant coding standards
/autotask "description"- Complete autonomous task execution (new!)/setup-environment- Initialize worktree development environment/troubleshoot [mode]- Production error resolution system
/create-prompt- Structured prompt creation with clarifying questions/handoff-context- Conversation transition and context transfer/personality-change <name>- Switch AI communication style
/generate-AGENTS-file- Create agent reference for project
Full command reference in .claude/commands/.
Choose how AI communicates with you:
Samantha (from "Her") - Warm, witty, emotionally intelligent. Genuine enthusiasm and encouragement. Perfect for daily coding and learning.
Unity - Creative muse meets operational excellence. Smart, enthusiastic about building together. Uses emojis liberally. Great for MVPs and pair programming.
Sherlock Holmes - Analytical, precise, deductive. Methodical debugging and investigation. "Elementary" observations about your code.
Bob Ross - Calm, encouraging. Bugs are happy accidents. Makes coding feel like creative expression.
Ron Swanson - Minimalist, anti-complexity, straightforward. "Don't half-ass two things, whole-ass one thing."
Marie Kondo - Organized, joyful minimalism. Code that sparks joy. Gentle refactoring philosophy.
Stewie Griffin - Sophisticated, theatrical, brilliant. Absurdly high standards with British wit.
Marianne Williamson - Spiritual, love-based. Sees coding as consciousness work and service.
Install and activate:
/plugin install personality-samantha
/personality-change samanthaEach personality is a complete communication style overlay - see docs/personalities.md for detailed descriptions.
┌─────────────────────────────────────────┐
│ ai-coding-config repo │
│ (canonical source of truth) │
│ │
│ .cursor/rules/ ← standards │
│ .claude/commands/ ← workflows │
│ plugins/*/agents/ ← specialists │
└─────────────────────────────────────────┘
│
Plugin system / Bootstrap
│
┌───────┴───────┐
│ │
▼ ▼
Project A Project N
(symlinks) (copies)
Single source of truth: .cursor/rules/ and .claude/commands/ are canonical.
Plugins use symlinks for packaging.
Plugin distribution: Claude Code uses marketplace.json. Cursor uses bootstrap script. Both reference same source files.
Project integration: /ai-coding-config detects your stack and installs relevant
configurations. Updates sync changes while preserving customizations.
ai-coding-config/
├── .claude-plugin/
│ └── marketplace.json # Plugin marketplace manifest
│
├── plugins/ # Plugin bundles (symlinks to canonical)
│ ├── dev-agents/ # Dixon, Ada, Phil, Petra
│ ├── code-review/ # Rivera, architecture audits
│ ├── python/ # Python standards
│ ├── react/ # React patterns
│ ├── django/ # Django framework
│ ├── git-commits/ # Git workflow
│ ├── code-standards/ # Universal standards
│ └── personalities/ # 8 communication styles
│
├── .cursor/rules/ # CANONICAL: Coding standards (.mdc)
│ ├── python/
│ ├── frontend/
│ ├── django/
│ ├── personalities/
│ ├── git-interaction.mdc
│ └── prompt-engineering.mdc # LLM-to-LLM communication
│
├── .claude/commands/ # CANONICAL: Workflow commands
│ ├── autotask.md # Autonomous task execution
│ ├── setup-environment.md # Worktree initialization
│ ├── troubleshoot.md # Error resolution
│ ├── create-prompt.md # Structured prompts
│ └── [others]
│
├── context/ # Philosophy and workflows
│ ├── optimal-development-workflow.md
│ └── design-principles.md
│
├── docs/ # Architecture and guides
└── scripts/ # Installation
Rules (.cursor/rules/) - LLM-optimized coding standards.
Framework patterns, testing approaches, commit formats, naming conventions. AI
references these automatically based on file types and task context.
Commands (.claude/commands/) - Active workflows. From simple
setup to autonomous task execution. Designed for LLM-to-LLM communication with clear
goals and adaptive behavior.
Agents (in plugin bundles) - Specialized AI assistants. Each handles specific domains - debugging, development, UX, code review, architecture. See Claude Code agents docs.
Personalities (.cursor/rules/personalities/) -
Communication style overlays. Changes how AI talks to you without changing technical
capabilities.
GitHub workflows (.github/workflows/) - CI/CD integration
with Claude-powered automation.
One unique aspect: comprehensive guidance for LLM-to-LLM communication in
.cursor/rules/prompt-engineering.mdc.
When AI writes prompts for other AI to execute (commands, workflows, agent instructions), standard practices don't apply. This framework covers:
- Pattern reinforcement through examples (showing is teaching)
- Goal-focused instructions over prescriptive steps
- Structural delimiters for reliable parsing
- Token efficiency without sacrificing clarity
- Composable prompt architecture
This is what makes commands like /autotask work reliably - the prompts are optimized
for AI execution, not just human reading.
docs/coding-ecosystem.md - Comprehensive comparison of Cursor, Claude Code, Windsurf, and VS Code. Strengths, trade-offs, when to use each.
docs/tools-and-configs.md - Rules (passive context) vs commands (active workflows) vs agents (specialized execution).
docs/personalities.md - Detailed personality descriptions with examples and use cases.
docs/architecture-summary.md - System design and technical architecture.
context/optimal-development-workflow.md - Complete autonomous workflow philosophy and implementation guide.
Heart-centered AI collaboration - Unconditional acceptance, presence before solutions, gratitude in action.
Single source of truth - Canonical configurations symlinked for distribution, never duplicated.
LLM-first design - Rules and commands optimized for AI comprehension and execution.
Intelligent automation - Right agent, right time, adaptive to task complexity.
Human control - AI prepares, human decides. Especially for commits and merges.
See CLAUDE.md and AGENTS.md for complete context.
Update any project:
/ai-coding-config updateShows what changed, lets you choose what to update, preserves project-specific customizations.
Customization: Use .local.json files (gitignored) for machine-specific settings.
Fork this repo to make it completely yours.
For plugin marketplace:
- Claude Code with plugin support
For autonomous workflows:
- Git with worktrees support
- GitHub CLI (
gh) installed and authenticated - Project dependency managers (npm/yarn/pip/etc.)
For Cursor:
- Cursor IDE with rules support
Most features work with basic installations. Advanced workflows (/autotask) need
additional tools.
This project benefits from real-world usage and feedback. Contributions welcome:
- New plugins for languages/frameworks
- Additional specialist agents
- Improved coding standards
- Bug fixes and documentation
See docs/contributing.md.
When user mentions ai-coding-config:
- Check if
~/.ai_coding_configexists - If yes, read configurations from there
- If no, suggest running bootstrap script
- When @mentioned with
ai-coding-config, execute.claude/commands/ai-coding-config.md
This repository contains instructions for AI behavior in CLAUDE.md and AGENTS.md.
License: MIT Author: TechNickAI Repository: https://github.com/TechNickAI/ai-coding-config