agent-skill-sync

What it does

• A/B/C/D migration taxonomy per skill — portable, missing deps, needs rewrite, platform-private — with the reason and the missing dependency named on every line
• Six well-known roots auto-discovered (Codex skills + marketplace plugin cache, Claude Code, Copilot, Gemini, Cursor, OpenCode), extended via skillsync.toml
• skillsync index generates INDEX.md + inventory.json so a hand-written tool-reuse skill stops rotting; check diffs upstream changes against the last snapshot
• Safe sync by construction: dry-run plans, forced updates back up first, nothing deletes, secrets and virtualenvs are excluded from copies

When to use it

Developers running Claude Code alongside Codex, Cursor or OpenCode whose skills have multiplied into inconsistent, half-stale copies across home directories — and who want one inventory and a migration plan before copying anything.

Getting started

pipx install agent-skill-sync

skillsync status                 # what do I have, and in what state
skillsync scan                   # one line per skill: category, sync state, missing deps
skillsync scan --category B      # only the ones needing dependencies
skillsync index --out ./output --lang zh
skillsync check                  # diff against the last snapshot
skillsync sync --dry-run         # always plan first
skillsync sync --categories A    # copy portable skills only

Things to know

The default sync target is ~/.workbuddy/skills — a different destination must be set in skillsync.toml before any real sync. Classification is heuristic and deliberately conservative, and category D (platform-private runtime) never migrates; treat the labels as a first pass, not a verdict. Marketplace plugin caches are reported with qualified plugin:name so collisions stay visible. The frontmatter parser implements exactly the YAML subset skill files use — unusual constructs become warnings, not crashes. Python 3.11 or newer. MIT.