People build great skills for Claude and keep them to themselves. SkillKeeper collects them, merges the duplicates into one best version, reads every line, and puts the result in everyone’s Claude by itself. The same files work in ChatGPT, Perplexity and the rest.
Send a skill from inside Claude, by upload, or by pasting instructions. It lands as a private draft.
Share with the team
write-proposalprivate draft
noteLine 14 names a client. Remove it before sharing?
Share with teamA lead is told
2Share
Share it with the team when it is ready. A lead is told.
3 similar skills → 1
write-proposalMaya
proposal-v2Jon
client-pitchPriya
→
write-proposalStructure · Maya Pricing · Jon Tone · Priya
3Review and merge
Similar skills are grouped. Claude drafts one best version, with every section’s origin shown.
Review
write-proposalsubmitted
ApproveSend back with a note
4Approve
Approved, or sent back with a note.
Publish v1.3.015 s
✓Checked
✓Validated with Claude’s plugin validator
✓Written to your private GitHub
5Publish
One click writes the library to your company’s private GitHub. Nobody opens GitHub.
Everyone’s Claude · 2 min later
MMayawrite-proposalv1.3.0
JJonwrite-proposalv1.3.0
PPriyawrite-proposalv1.3.0
6In everyone’s Claude
New and changed skills reach everyone’s Claude within minutes, by themselves.
Your own files
Write it once, review it once
A skill is a folder with a SKILL.md file inside. Anthropic released that format as an open standard, Agent Skills. SkillKeeper publishes your library in exactly that format, as plain files in your company’s own private repository. It survives a change of vendor, plan or admin.
A skill is text an AI obeys. Before one is shared or sent, the check reads every file as text: hidden text, instructions aimed at the model, secrets, links that reach out, and a reading by Claude. On every plan. A skill that passes is checked, with a date.
IT MEANS
IT DOES NOT MEAN
Every file of this exact version was read as text, by the rules and by Claude, on this date
That the skill is safe in every situation
A person decides on what the check flags
That nothing can be found. Rules can be fooled, and so can a model that reads
It was checked on its own
That it is harmless in combination with other skills
A finding is a reason to look
That a finding proves bad intent
Every check ends in one of three outcomes:passnotesrefused
One partner set the whole thing up himself in twenty minutes: GitHub, Claude, the first skill pushed, shared, approved, published, three colleagues invited.
A CLIMATE FUND
The team imported its existing plugin repository and published five skills. They were in their Claude within minutes, with nobody clicking anything.
~15 s
for a publish
~20 s
for a merge draft by Claude
537
automated tests run on every change
Every
publish is validated with Claude’s own plugin validator before it goes out
Who it is for
Any company where Claude is in daily use
The lead
A head of operations, chief of staff, team lead or founder at any company of five to two hundred people where Claude is in daily use. You want the team’s best way of doing things captured once and used by everyone.
Everyone else on the team
Analysts, consultants, operators, account managers. You use Claude’s app all day and build good skills without calling them that. One sentence in Claude sends one to the library: “push this skill to SkillKeeper”.
Black Matter VC builds your starter library from your company’s own public material and walks the lead through the first week. Day one is not an empty page. Free for the first teams.
No. SkillKeeper creates and maintains a private repository for you. You never open it.
Which Claude plan?
Any paid one. On Team and Enterprise an owner adds the library once for everyone; on Pro and Max each person adds it once.
Can we use the skills outside Claude?
Yes. The library is the open Agent Skills format. ChatGPT, Codex, Perplexity Computer, Cursor, Copilot, Gemini CLI and others read it; today you upload or point them at the files yourself.
Who reads our skills?
Your leads, in the app. The check reads them as text, and Claude reads them once as part of the check. Nothing is used to train anything. Details on the Security page.
Is a checked skill safe?
No tool can promise that. Checked means every byte was read by the rules and by Claude on a given date, and a person decides on what the check flags.
One library for the skills your team writes for Claude.