Agent-agnostic Agent Skills (SKILL.md format) for a controlled, chunked development workflow. They work in Claude Code, Cursor and any other agent that reads SKILL.md.
| Skill | Invocation | Description |
|---|---|---|
| merge‑parent‑branch | User | Merge the parent branch into the current one: sync both with the remote, resolve conflicts so neither side's fix or feature is lost, then commit and push after one approval. |
| ship | User | Grill → plan → small chunks, each built by a subagent, that you commit with one click or ship commits at the autonomy you pick (chunk, review, auto) → independent review, through the repo's own verify skill when it has one. Worktrees and parallel chunks for independent work. State and history live in ~/.agents/ship/, never in the repo. |
| web‑ready‑audit | User | Audit a website or web app before launch, on any stack: what is missing or misconfigured, as Blocker, Should fix and Nice to have findings. Report only, never edits. |
With the skills CLI, globally, for Claude Code and Cursor:
npx skills add xpaczka/skills -s <name> -g -a claude-code -a cursor
npx skills update -g # pull the latest versionOr as a Claude Code plugin, with every skill namespaced (/skills:ship; the bare /ship still works unless a project defines its own ship):
/plugin marketplace add xpaczka/skills
/plugin install skills@xpaczkaship keeps its state outside the repo, in ~/.agents/ship/. To stop Claude Code from asking on every call, allow exactly that folder and ship's locate script in ~/.claude/settings.json (use your absolute home path where shown):
"permissions": {
"additionalDirectories": ["/Users/<you>/.agents/ship"],
"allow": [
"Read(~/.agents/ship/**)",
"Edit(~/.agents/ship/**)",
"Bash(/Users/<you>/.claude/skills/ship/scripts/locate.sh)"
]
}The dev commands and the repo's rules for agents and authors live in AGENTS.md, with topic docs in .agents/.