Fork And Publish Skill

Fork and publish a skill with qvr

qvr lets you take an installed skill, eject it into your project to edit directly, then publish the result — either back to its origin or to a brand new fork you own — cutting tagged releases as you go. After each publish the lock entry flips back to consume mode, so you can iterate: edit → publish v0.2.0 → edit → publish v0.3.0. This skill drives that loop.

When to use this

• The user wants to tweak a skill's content and keep the change.
• They want to publish a skill back upstream, or fork it to a repo they control.
• They want to cut tagged releases and iterate version by version.

For installing/discovering skills use onboard-skills; for verifying provenance of what you publish use verify-skill-supply-chain.

Prerequisites

• The skill must be installed in a project (not only --global). Editing a

global skill in place would mutate a copy every project shares, so it's not supported. To change a global skill: qvr add into a project, edit and publish there, then re-add the published version with --global.

• A clean-ish working state helps; publish refuses a dirty eject dir unless you

pass --auto-commit.

Workflow

1. Eject the skill for editing

qvr edit promotes the symlinked skill into a real directory under the canonical agent target (the alphabetical-first installed target — e.g. .claude/skills// or .cursor/rules//, depending on what you installed into); any other installed target dirs become relative symlinks to it. The lock entry gains mode = 'edit'.

qvr edit my-skill
qvr edit my-skill --author "Jane Dev" --email jane@example.com   # initial-commit identity

It's idempotent — running it again after the first eject is a no-op.

2. Make and review changes

Edit the files in the ejected skill directory (the canonical agent target), then review:

qvr diff my-skill           # git diff in the skill worktree
qvr diff my-skill --stat
qvr status my-skill         # dirty state + git ahead/behind

3a. Publish back to the original upstream

qvr publish my-skill -m "Fix edge case" --tag v1.0.1
qvr publish my-skill --dry-run -m "..."     # validate and stage without pushing
qvr publish my-skill --auto-commit -m "..." # stage+commit a dirty eject dir first

3b. Fork to a new repo you own

Use --fork to retarget the push, and --migrate to rewrite the lock entry so future publishes track the fork:

qvr publish my-skill --fork https://github.com/me/my-skill.git --migrate --tag v0.1.0

--fork --migrate auto-registers the fork as a config registry named / (e.g. me/my-skill). To re-add the skill from it later, use that full name:

qvr add my-skill --registry me/my-skill

--layout controls the published repo shape: root (single-skill repo, the default for --fork) or nested (multi-skill registry under skills//, the default otherwise).

4. Iterate releases

Each publish flips the entry back to consume mode, so the loop repeats cleanly:

qvr edit my-skill
# ...changes...
qvr publish my-skill --tag v0.2.0
qvr edit my-skill
# ...changes...
qvr publish my-skill --tag v0.3.0

Consumers pick up a release with the version manager:

qvr outdated                  # which installs have newer upstream commits
qvr switch my-skill v0.3.0    # explicit pin
qvr switch my-skill --latest  # jump to the latest semver tag
qvr pull my-skill             # fast-forward the current ref (alias of switch --tip)

5. Add a new skill to a registry (greenfield path mode)

To publish a brand-new local skill directory into a multi-skill registry repo:

qvr publish ./my-new-skill --registry me/agent-skills -m "Add my-new-skill"

Gotchas

• Dirty eject dirs are refused. publish won't push uncommitted changes

unless you pass --auto-commit (which stages + commits first).

• --migrate only matters with --fork. Without it, the fork push is

one-off and the lock entry still tracks the original upstream.

• Registry name after a fork is the full /, not a short alias —

qvr add … --registry, registry list, and registry update all need it.

• Clean up edit ejects before teardown. A sync with a drifted/ejected entry

fails asking for --allow-drift; publish or revert the eject first (or re-consume the skill) before a cache clean + sync.

• Lockfile/HEAD mismatch is refused for integrity; only override with

--allow-lockfile-heal if you understand why HEAD diverged.

Troubleshooting

• "refusing dirty working directory" — commit the eject, or pass --auto-commit.
• Second release fails to resolve the commit — make sure each iteration goes

through qvr edit again; the entry must be back in edit mode to publish.

• qvr add --registry can't find the fork — use the full /

name that --fork --migrate registered, visible in qvr registry list.

• Want to undo an eject without publishing — re-consume the skill (re-add /

switch to a ref); confirm with qvr status and qvr sync.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: astra-sh
• Source: astra-sh/qvr
• License: MIT

Install and usage instructions live in the source repository linked above.