Today, AI assistants (coding agents, web agents, etc.) rely on specialized assets like system instructions and multi-file "skills". These assets are often hardcoded into applications or scattered across repositories without version control.
This leads to several issues:
- 🗑️ Domain clutter: Assets are not part of the domain, but they are included in the project.
- 🔎 Discoverability: Hard to find assets for a specific purpose.
- 🔗 Sharing: Hard to share assets with other projects.
- 📋 Duplication: Assets are often duplicated across repositories.
- 📉 Version Drift: Different developers or tools using outdated versions of an instruction.
- 🕵️ No Traceability: Losing track of why a specific instruction was changed.
- 🏝️ Platform Silos: A "skill" built for one assistant is hard to use in another.
⚠️ Security Risks: Using unverified assets without integrity checks.
ARCA solves these problems by treating agentic assets as first-class versioned dependencies, much like NPM, Go Modules, or Maven.
- 📌 Deterministic Resolution: Use specific versions (SemVer) or commit SHAs to ensure your agent's behavior is predictable and reproducible.
- 🔀 Platform Independence: ARCA is not tied to one IDE. Its "Projection" system allows it to bridge versioned assets into terminal CLIs, VS Code, Cursor, or even mobile apps.
- 🛡️ Security by Default: Content-addressable hashing (SHA-256) ensures that what you run is exactly what was published.
- ⚡ Performance: The Go-based core is designed for speed, ensuring that syncing assets happens in milliseconds, even in complex workspaces.
- 🌐 Decentralized: No central "Store" required. Use your existing Git infrastructure or local file shares.