Claude Public

Claude Code Portable Configuration

A template for syncing agent configuration — Claude Code settings, launchers, agents, skills, a statusline, dynamic workflows, and a Codex CLI target — across multiple Windows machines. It installs in one command, checks its own work, refuses to overwrite one machine's work with another's, and can put back anything a deploy changed. This is the public version of a personal setup — feel free to fork, adapt, and add your own skills, agents, and workflows.

Two scope notes up front:

• Windows-to-Windows only, by design. Secrets are DPAPI-encrypted, the launchers are PowerShell, and paths are resolved through Windows folder APIs. None of it travels to macOS/Linux.
• The Codex target means "Codex CLI launched through codex-sp". Codex profiles are CLI-flag-only (verified on 0.149.0: a profile key in config.toml is rejected as legacy), so direct codex launches, the IDE extension, and the desktop app never see the synced profile.

Repository Structure

• global/ — The Claude Code user target, deployed to %USERPROFILE%\.claude\
  • settings.json — Global settings, permissions, model, and statusline registration
  • CLAUDE.md — User-level instructions: what holds wherever you start a session. Also the source of the generated Codex instructions (see below)
  • statusline-command.ps1 — Statusline showing user@machine, directory and git branch, subscription usage for the current five-hour window, and context-window fullness
  • agents/ — Agent definitions (file-manager, for bulk file operations that would otherwise eat the main context; research-worker, the minimal-context worker used by deep-research-tiered)
  • skills//SKILL.md — Claude-only skills, one directory each (sync-config, deep-research-tiered)
• shared/skills/ — Skills both products consume, deployed to both ~/.claude/skills/ and ~/.agents/skills/. Empty in the public repo; the mechanism is what s