skill-sync

To sync Claude skills, run the Python CLI via uv to execute a dry-run plan that detects endpoint drift, then confirm the apply action to mirror new and updated skills across canonical git, Codex, and Claude Code. This ensures all runtimes stay consistent without manual file copying.

Endpoint drift happens when user-created Claude Desktop skills are manually copied or edited independently across canonical git, Codex, and Claude Code runtimes. Running a status check lists each endpoint with skill and file counts to quickly spot where skills have diverged.

Yes, you can run a status check to list each endpoint including canonical, Codex, Claude Code, and Claude Desktop. This visibility operation displays skill counts, file counts, and the full skill list for each target so you can spot drift before applying any sync operations.

You can safely publish a new Claude skill by running an idempotent sync with a dry-run preview first. After reviewing the planned changes, you confirm the apply step to mirror the skill to canonical git, Codex, and Claude Code while propagating any necessary deletions.

No, the targeted removal workflow removes a specified skill from the canonical git repo and downstream targets like Codex and Claude Code, but it explicitly does not delete the skill from Claude Desktop. The removal process uses a dry-run plan followed by a confirmed apply step.

When a conflict is created during synchronization, the operational guardrails surface the conflict clearly and skip the canonical commit. You can use inspection flags like --dry-run and --no-commit to review the state before attempting to resolve the divergence and reapply the sync.