Better methods.
Not more files.

One reviewed catalog of your team's AI skills. Approve exact bytes, know where every version landed, and see when a copy drifted.

Join the beta

Free during the beta. Self-hosted, standard library only, one SQLite file. Read the source.

The problem

Everyone keeps a private copy.

Nobody knows which version a teammate runs.

An edit that helps one person silently regresses another.

How it works

Make a change, rehearse it, then see where it lands.

Every change walks through the same four steps, so the catalog, not the copy on disk, is the source of truth.

Edit

Draft in a safe place. Production stays on the last approved version until you say otherwise.

Test

Checks run against the exact draft: package shape, required sections, no secrets, no machine paths, plus your team's own rules.

Release

A different maintainer approves an immutable, digest-stamped version. Release to canary, then the team. Roll back by pointing the channel at the previous version.

Adoption

status and sync show every copy as current, outdated, modified, missing or deprecated, and record what they found. Installed is not the same as loaded; a loaded receipt comes from your tool's hook.

Sixty seconds

The whole loop, from a terminal.

# one machine to start; add --server later
$ skillcurrent init --team northstar --owner maya
$ export SKILLCURRENT_TEAM=northstar SKILLCURRENT_USER=maya
$ skillcurrent members add noah --role maintainer
$ skillcurrent members add sam --role contributor

# 01 edit · 02 test
$ skillcurrent --as sam skills new ./refund-handoff/SKILL.md
$ skillcurrent --as sam skills checks refund-handoff
refund-handoff: all 7 checks passed
$ skillcurrent --as sam skills submit refund-handoff --note "baseline"

# 03 release · a different maintainer approves the bytes
$ skillcurrent --as noah skills approve refund-handoff --release production
Approved refund-handoff 1.0.0 (authored by sam); sha256 2fffa7f67a44…

# 04 adoption
$ skillcurrent install refund-handoff
$ skillcurrent status
refund-handoff  claude-code  production  1.0.0  1.0.0  current

The beta ask

We're onboarding 5 to 10 teams that use two or more AI coding tools.

You get the tool and direct support while it's rough. We get to watch two things:

whether you keep review switched on, and
whether sync ever catches a bad copy.

You run it on your own machines. Once a week you send us the counts from skillcurrent beta-report: numbers only, no skill content, no names. No logos, no case studies, no numbers we haven't seen. If the beta shows the workflow doesn't earn its keep, we'll say so.