What problem does it solve? Setting up and maintaining a personal Claude Code environment involves many fragile pieces: native settings.json, CLAUDE.md instructions, read-only agent roles, hook registrations, private task-state directories, and MCP server integrations. Manual edits risk clobbering personal configuration, duplicating hook registrations, or silently breaking idempotency. This Skill performs explicit, patch-only configuration and recovery while preserving every personal value. ## Core Features & Use Cases - Patch-only reconciliation: Applies narrow managed-block updates to CLAUDE.md and settings.json while preserving personal content, formatting, ordering, and values, with byte-level drift detection before every write. - Read-only convergence checker: A redacted structural checker reports each surface as Aligned, Not aligned, or Blocked without printing secrets or modifying files. - Atomic missing-settings recovery: Creates a missing settings.json exclusively via a private hard-link publish that never replaces existing or concurrently created files. - Optional profiles and integrations: Explicitly selected trusted-local permissions, delegation policy, private Task Implementer storage, and user-scope MCP servers (Context7, Playwright, Terraform, MarkItDown, Microsoft Docs, GitHub, OpenAI docs). - Use Case: You installed Claude Code on a new machine and want the repository's three read-only roles, shared hooks, and managed instructions installed without losing your existing model choice, deny rules, or custom preferences. Invoke the skill to inspect, reconcile, and verify convergence. ## Quick Start Ask the agent to use config-claude to inspect your Claude Code setup and reconcile the native instructions, roles, and hooks while preserving your personal settings.
