Set the rules once. Every publish, install and tool grant follows them, across every team and repository in your company.
Governance
Rules every publish has to pass.
Decide what a skill needs before it reaches anyone. A policy applies to a project, a team or the whole company, and every exception is recorded with a reason.
Required reviewers for each team
A minimum eval lift over the version it replaces
Admin approval for any new tool grant
Separate rules for each project, team and the company
Publish policyPlatform14 skills
Require review2 approvals from Platform reviewers
Minimum lift+10% over the version it replaces
Passing evalA PASS receipt on the exact bytes published
New tool grantsAn admin approves any tool a skill adds
ChangelogEvery version says what changed
4 checks run on every publish to PlatformEdited by Mira Reyes, Sep 12
Security checks
Every skill and tool is scanned before it installs.
Each new version is checked for secrets, network calls, shell commands that reach outside the working tree, tool grants and dependencies. A flagged version is held until someone with the right role clears it.
Keys and tokens in any file of the skill
Network calls checked against your allow list
Shell commands that touch production or leave the working tree
Tool grants and dependency advisories, for skills and MCP tools
db-snapshot0.9.21 flag
Secrets
No keys or tokens in 7 files
Network
api.github.com, on your allow list
Shell
Reads a production database URL from the environment
Publish a skill to one project, one team or the whole company. Engineers outside its scope never see it, and widening the scope goes through the same policy as a publish.
Scopes for a project, a team or the company
Teams and members come from your identity provider
Widening a scope needs approval
Engineers only see the skills that reach them
Scopesmigration-guard 3.0.0
Available to
Platform48 engineers
Engineers outside Platform never see it.
Audit trail
A record your auditors can read.
Every publish, install, grant, approval, eval and membership change is recorded with who did it, when, and what changed. Filter it in the dashboard, export it, or stream it to your SIEM.
Who, what and when for every change
SKILL.md diffs and eval receipts attached to each publish
Export as CSV or JSON
Streaming to your SIEM
Audit logAcme
MRMira Reyes approved deploy-check 2.3.0Publish
JLJon Lee ran an eval on deploy-check 2.3.0+18%
AKAda Kim granted Bash to db-snapshotGrant
SCIMSCIM removed dana from AcmeMember
TSTom Sato installed migration-guard 3.0.0Install
MRMira Reyes published flag-cleanup 1.2.0Publish
SCIMSCIM added priya to MobileMember
JLJon Lee revoked Network from slack-notifyGrant
Streaming to your SIEMKept for 7 years
Everything else a rollout needs.
SSO and SCIM
Sign in with Okta, Microsoft Entra ID or Google Workspace. Engineers are added and removed from your identity provider.
Any git host
Keep the team repository on GitHub Enterprise, GitLab or Bitbucket, in the cloud or self-managed.
Roles and permissions
Decide who can publish, approve, grant tools and change policy, separately from repository admin.
Hosted or self-hosted
Run the dashboard and registry in our cloud or inside your own network.
Adoption analytics
See which skills each team has installed, which versions they are on, and where an update is still rolling out.
Signed builds for IT
A signed, notarized desktop app your IT team can deploy through MDM.
Sandboxed evals
Evals run in isolated containers, away from engineers’ files and credentials.
Priority support
A named engineer, a shared Slack channel and agreed response times.
Your skills stay in your repository.
The open source tool and the self-hosted edition run with no Terum server between your engineers and your git host. Every network call is documented.
Stays with you
Skills, versions and receiptsIn a private repository your company owns
Who can read and change themYour git host’s permissions and your identity provider
Eval transcripts and sandboxesOn the machine that ran the eval
SettingsIn each engineer’s home folder, readable only by them
Leaves your machines, and where it goes
Git fetch and pushYour git host
Eval and judge sessionsAnthropic, through each engineer’s own Claude Code login
A daily release checkgithub.com, for the list of release tags only
No telemetry, analytics or crash reportingApache-2.0, so your security team can read every lineDownloads verified against GitHub build provenance
Yes. terum-skills stays free and open source under Apache-2.0. The enterprise edition adds governance, identity and support on top of it.
How is the enterprise edition priced?
Pricing depends on the size of your engineering team and how you deploy. Book a call and we’ll walk you through it.
What does a design partner commit to?
Using the enterprise edition with at least one team and telling us what works. We meet on a schedule that suits you.
Security and data
Where does our data live?
Skills, versions and receipts live in a git repository you own. With the self-hosted edition the dashboard runs inside your network as well. Evals reach Anthropic through each engineer’s own Claude Code login.
Who pays for the model usage in evals?
Evals run on each engineer’s own Claude Code login, so model usage stays on the plan your company already has with Anthropic.
Rollout
What does IT need to roll it out?
The signed desktop app deployed through your MDM, single sign-on with your identity provider, and access to the git host where the team repository lives.
Which git hosts are supported?
GitHub, GitHub Enterprise, GitLab and Bitbucket, in the cloud or self-managed.
Which agents does it work with?
Claude Code, Codex and Cursor, the agents your engineers already use.
Can we keep what our team already has?
Yes. The enterprise edition reads the same team repository as the open source tool, so skills, versions and receipts carry over.
Works with the agents your team already uses
Claude Code
Codex
Cursor
Bring Terum to your company.
Book a call and we’ll show you the enterprise edition on your own skills.