Lumosylva Skills

A collection of Claude Code plugins providing structured expert programming skill workflows.

Design Philosophy

AI coding tools tend to be overly eager—making changes without explicit permission and assuming they know better than the user. Lumosylva Skills enforces mandatory mode declarations and strict responsibility boundaries to ensure AI always does the right thing at the right stage, preventing unauthorized modifications and destructive changes.

Plugin List

Plugin	Description	Path
project	Structured expert programming skills with five collaborative modes for rigorous software development workflows	plugins/project

Workflow Overview

Research → Innovate → Plan → Execute → Review
  ^                              |
  └───── Rollback on issues ─────┘

Each new session starts in Research mode by default. Switch to the next phase using explicit signal commands.

Five Collaborative Modes

Mode	Command	Purpose	Core Capabilities
Research	/project-research	Information gathering and deep understanding	Read files, analyze architecture, identify technical debt
Innovate	/project-innovate	Brainstorm potential approaches	Explore multiple solutions, weigh pros and cons, document findings
Plan	/project-plan	Create detailed technical specifications	File paths, function signatures, checklist generation
Execute	/project-execute	Accurately implement the approved plan	Follow checklist item by item, track progress
Review	/project-review	Verify implementation matches the plan	Line-by-line comparison, flag deviations, prepare commits

Installation

Install from GitHub

# Claude Code
claude plugin install Lumosylva/lumosylva-skills

# MiMoCode
mimo plugin install Lumosylva/lumosylva-skills

Manual Installation

Copy the plugin folders from the plugins/ directory into your skills directory:

• Claude Code: .claude/skills/
• MiMoCode: .mimocode/skills/

Core Protocol

• Mode Declaration: Every response must begin with [MODE: XXX] declaring the current mode
• Responsibility Isolation: Each mode may only perform actions allowed at that stage
• Rollback Mechanism: When issues are discovered during execution, automatically return to Plan mode for readjustment
• Mode Switching: Must be triggered by explicit signals (e.g., /project-execute); never switch on your own

Placeholders

Placeholder	Description	Example
TASK_FILE_NAME	Task file name	2025-01-14_1
TASK_IDENTIFIER	Short task identifier	fix-cache-bug
MAIN_BRANCH	Main branch name	main
YOLO_MODE	Confirmation mode	Ask / On / Off

Yolo Mode

• Ask: Prompt the user for confirmation before each step
• On: Execute all steps automatically (high risk)
• Off (default): Require user confirmation for every significant step

Contributing

Issues and pull requests are welcome.

License

MIT License - see LICENSE