Rune is not just a collection of isolated prompts, but a Skill OS designed to address the biggest problem facing AI today: Consistency and Context Persistence in projects. 1. Design Philosophy: From "Isolated Skills" to "Skill Mesh" Most skill sets (like Claude's 500+ skills) operate in a "Hub-and-Spoke" or "Pipeline" fashion. If one link fails, the entire process stops. Rune breaks this with its Mesh architecture: 55 skills are networked bidirectionally. If Skill A fails, it automatically reroutes through Skill B or C to achieve the final goal. Advantage: Extremely high resilience. 2. The 5-Layer Architecture The highly hierarchical Rune system prevents AI from being overwhelmed with information: L0: Router (Meta-enforcement): Directs all actions. This is the "brain" controlling the flow. L1: Orchestrators (Workflows): Contains powerful workflows such as /rune cook (build features), /rune rescue (refactor bad code), /rune launch (deploy & marketing). L2: Workflow Hubs: Specialized skill clusters (Research, Development, Quality, Rescue). L3: Utilities: Atomic tools - such as hallucination-guard, sequential-thinking, sats (static security scanning). L4: Extension Packs: Extension packages for each domain (UI, Backend, Trading, AI/ML). 3. "Killer" Features Bro should pay special attention to these because they directly elevate the project: Phase-Aware Execution (Multi-session): This is the "lifesaver" for large projects. Rune breaks down the task into phases. If the AI session is context-filled (freezes), it saves the state to the .rune/plan-*.md file. In the next session, it reads this file and continues running from that phase without needing to re-explain from the beginning. Cognitive Bias Detection: Before making architectural decisions, Rune runs 12 bias checks (such as anchoring, status-quo). It forces the AI to think in multiple dimensions (second-order thinking). Amateur-Proof Plans: When you use skill plans, it creates extremely detailed phase files with: data flow, code contracts, and failure scenarios. The goal is to ensure even a weak model like Haiku can code correctly. Cost Intelligence: Rune automatically selects the model for each task (Haiku for bug scanning, Sonnet for coding, Opus for architecture). This saves up to 75% on costs compared to using the most powerful model for everything. 4. Practical Workflow for Antigravity :: npx @rune-kit/rune init --platform antigravity This command compiles 55 prompts into a configuration file that is fully compatible with Antigravity. Use scenarios: Want to build a new feature: /rune cook "Add whale wallet analysis module" -> Rune will automatically run from codebase survey -> planning -> writing tests -> writing code -> verifying. Want to clean up old code: /rune rescue -> It will measure the code's "Health score", write "Safeguard tests", and then proceed with "Surgery" (refactoring) each module one by one. To check security: Sentinel will automatically scan secrets, OWASP patterns, and dependency vulnerabilities. 5. Overall Assessment (Forensics Tagging) [CORE_LOGIC]: Mesh and Phase-persistence systems. [API_INTEGRATED]: Deep integration with MCP servers and search/browser tools from Antigravity/Cursor/Windsurf. [UI_SHELL]: Unified interface via Markdown files (everything is in the .rune/ folder).

Rune — Less skills. Deeper connections.

A mesh ecosystem for AI coding assistants.

A mesh ecosystem for AI coding assistants.