Juanyque has released agent-brain, a personal operating model designed to unify configuration and memory for AI coding agents like Claude Code, OpenCode, Codex, and Antigravity CLI. The project treats the user's notes directoryβ€”whether an Obsidian vault or a simple folderβ€”as the single source of truth, decoupling agent intelligence from specific runtime installations. This approach aims to solve the fragmentation problem where agent context is trapped in local config files like CLAUDE.md or AGENTS.md.

Version-Controlled Runtime Configuration

The core innovation of agent-brain is its use of Git to track runtime config and memory. Instead of scattering settings across ~/.claude, ~/.config/opencode, or ~/.codex, the system stores these files in a git-tracked 'brain' directory. The installer then symlinks these files into the respective runtime directories. This ensures that agent configuration and accumulated memory travel with the user across different machines, providing a consistent experience regardless of the hardware.

Session Lifecycle and Optional Skills

Beyond static configuration, agent-brain introduces a structured session lifecycle managed by the 'brain' skill. This includes daily notes, session notes, and consolidation routines. The system is built on a state machine that transitions the brain from 'virgin' to 'attached' and 'maintenance' states. Additionally, users can opt-in to a 'boyscout' skill, which identifies improvement opportunities during work sessions and routes them to a remediation workflow or backlog, fostering continuous project hygiene.

Multi-Runtime Support and Installation

The project supports a wide array of runtimes, including Claude Code, OpenCode, Codex CLI, and Antigravity CLI. Installation is handled via a bootstrap script that clones the repository to ~/.local/share/agent-brain. The orchestrator performs a dry-run by default, allowing users to review the plan before applying changes. It includes safeguards for existing setups, such as backing up current _COMMON entries and relinking stale templates, ensuring that the migration process does not break existing agent configurations.

Key Takeaways

  • Agent-brain decouples agent memory from specific runtime installations using symlinks.
  • Configuration and memory are version-controlled via Git, enabling cross-machine portability.
  • Supports Claude Code, OpenCode, Codex CLI, and Antigravity CLI.
  • Includes a 'boyscout' skill for automated improvement spotting and backlog management.

The Bottom Line

This is a solid infrastructure play for power users drowning in fragmented agent configs. By treating memory as a git-tracked artifact, agent-brain offers a pragmatic solution to agent context portability that feels like a necessary evolution for serious AI-assisted development.