The right skills in every repo — and any drift in plain sight.
Teams from $8/seat/month. No seat minimum. 14-day trial, no card.
Right now, every engineer is running their own setup.
Three people on your team used roughly the same agent skill last quarter; two of them don't know about each other's version. Your senior engineer has a setup that actually works — nobody else can see what's in it. New hires ask whoever happens to be around and copy whatever they're handed.
The skills are fine. There's just no shared place they live.
Drift isn't a one-time mistake. It's the default.
Every agent works from a slightly different copy of “how we do things,” so the output diverges. Your codebase starts to read like it was written by five different teams — because, through five different agents, it was.
Curate one library, distributed across your org — scoped to what each repo needs.
A team library on SkillRepo is the one approved set of agent skills for your whole team. You decide what goes in it; it's the one place every engineer's agent draws from. Add skills from the public catalog, publish your own private ones, or mix both in a single library — vetted best-practice skills alongside the internal ones that encode how your team actually works.
Curating is a deliberate act, not a free-for-all. Adding, removing, and approving skills runs through the people you've given that permission — so the library reflects a decision about what your team standardizes on, not whoever happened to push last.
The same approved versions in every repo — not just the same library.
Keeping a team in sync used to mean every machine pulling the latest published version. That closes the obvious gap and leaves two open: a publisher's merge reaches every repo before anyone on your side decides, and you can't see which repos are actually running what.
Skillsets closes both. Your lead pins an approved version in the library, and that version is what every repo receives until you approve the next one — your own team's skills still advance on publish; external updates wait for review. Group skills into named skillsets — engineering, marketing, finance — and each repo declares the one it runs in a one-line file checked into git, reviewed like any other change.
Approving a skill isn't the same as trusting it.
A skill runs inside your agents with real access to your tools, so curating the library is only half the job. The other half is governing it: a signal before you approve a skill, a record of every change that shaped it, and a way to take a bad version back when one slips through. Grading shows you where to look, the audit log shows you what happened and who did it, and recall takes a bad version out of service across your repos. You're not asked to trust the library — you're given the evidence to check it, and the control to act.
Graded, so you're not curating blind.
Curating a library means vouching for what's in it. 36.8% of public skills contain a security flaw (Snyk, Feb 2026) — you don't want to approve those by accident.
Every skill on SkillRepo gets a two-layer analysis on publish — static checks plus an AI pass across safety, quality, clarity, and completeness — and ships with an A–F grade and the findings behind it. You curate with the score in view instead of guessing.
Grades are signals, not a certification. They tell you where to look; they don't replace reading a skill before you run it.
An audit log, because someone will ask.
When someone asks what your agents were allowed to run — a security review, a customer's vendor questionnaire, an incident — “we think it was current” isn't an answer. SkillRepo records the changes that shape the library to an account audit log: skills added and removed, publishes and visibility changes, member and role changes, GitHub integration syncs, and API key creation and revocation — each with who did it and when.
It's the paper trail for what your team standardized on and when it changed — so curation and compliance are the same motion, not two separate projects.
When a bad version gets through, take it back.
Grading and review catch most bad versions, but approval isn't proof — one can still turn out wrong after you've signed off, and by then it's synced across your repos. Recall pulls that exact version out of service: it stops reaching your organization's repos and agent connections right away, and repos pinned to it fall back to the last good version on their next sync — or, if the recalled version was the only one, the skill stops being served until you restore it.
Recall shows exactly who received the version while it was live — the developers, across the repos, that pulled it. Every recall is written to your organization's history and attributed to whoever made the call, and it's reversible if it was a false alarm. Routine updates go through review; recalls don't wait for one.
What every seat gets.
A seat isn't access to a list of links. It's the same curated library, delivered as real files into the tools your developers already use.
Into every dev environment
The library syncs as native skill files into Claude Code, Cursor, Windsurf, VS Code + Copilot, Gemini CLI, Codex CLI, and Cline — landing where each tool already looks for them. No per-editor wiring.
Current without the reminders
On most dev environments a session-start hook refreshes the library automatically every session — nothing to run, nothing to remember. The handful without a hook stay current with an occasional skillrepo update.
One source, every agent
Curate once and every engineer's agent runs the same vetted skills — whichever tool they prefer. The library is the standard; the editor is just where it shows up.
Questions teams ask.
- How do you prevent skill drift across a team?
- Your lead approves versions in the library, and each repo declares the skillset it runs. On every sync a repo receives exactly the approved versions of exactly the skills in its skillset — and you get a per-repo record of what was delivered, so you can see which repos are on the approved set and which have drifted.
- Is the grade a certification?
- No. Grades are signals — a two-layer analysis (static checks plus an AI pass) scored A–F with the findings attached. They tell you where to look. Always review a skill yourself before your team runs it.
- Which dev environments are supported?
- Claude Code, Cursor, Windsurf, VS Code + Copilot, Gemini CLI, Codex CLI, and Cline. The library syncs as native skill files into each one — no per-editor setup.
The right skills in every repo — at the versions you approve.
Curate once, scope per repo, approve the versions — and every repo runs exactly its approved set, across every dev environment, with a view of which comply.
Teams from $8/seat/month. No seat minimum. 14-day trial, no card.
Weighing the alternatives? See SkillRepo vs git and SkillRepo vs skills.sh.