Caliber open-sources AI context CLI

This feels less like a prompt helper and more like a compiler for AI setup. The traction is the tell: teams are hungry for anything that replaces hand-maintained context files.

• –Repo-first generation should age better than hand-written rules because the codebase stays the source of truth.
• –Cross-tool support matters because teams are already mixing Claude Code, Cursor, and OpenAI Codex rather than standardizing on one assistant.
• –MCP auto-discovery is the strongest wedge; tool wiring is often the hidden tax that keeps teams from using richer agent workflows.
• –The `score`/`refresh`/`undo` loop lowers adoption risk, and open source should let framework-specific skills accumulate quickly.
• –The main risk is output quality: if Caliber becomes noisy or wrong after a refactor, devs will stop trusting the generated files.