Everything you need to manage
AI skills at scale

Whether you work from the terminal, a desktop app, or the browser - these are the features that make Agentver the only way to manage AI agents & skills.

Git-native versioning

Real Git under the hood

Not a database pretending to be version control. Every skill lives in a real Git repository with real branches, real tags, and real commit history.

  • Semantic versioning with agentver version
  • Draft branches for experimentation
  • Full commit history with agentver log
  • Unified diffs with agentver diff
  • Cryptographic integrity verification on every file
  • Reliable, consistent installations every time
version control
$ agentver log deploy-checker
a1b2c3d v2.0.0 Add error retry logic
e4f5g6h v1.2.0 Improve validation rules
i7j8k9l v1.1.0 Add timeout configuration
m0n1o2p v1.0.0 Initial release
 
$ agentver diff deploy-checker
--- a/SKILL.md
+++ b/SKILL.md
@@ -12,3 +12,5 @@
+ retry_count: 3
+ retry_delay: 1000
$
agent detection
$ agentver scan
scanning for AI agents...
 
✓ claude-code ~/.claude/
✓ cursor ~/.cursor/rules/
✓ windsurf ~/.windsurf/rules/
✓ copilot ~/.github/
 
4 agents detected. skills installed to all.
$
Universal agent support

43 agents and counting

Agentver auto-detects which AI agents you have installed and automatically connects skills to each one. One skill works everywhere - zero manual configuration.

  • Automatic agent detection with agentver scan
  • Smart file placement - one source, everywhere you need it
  • 5 package types: Skills, Configs, Plugins, Scripts, Prompts
  • Config translation between agent formats
  • Alias system for agent ID variants
See all 43 agents →
Built-in security

28 rules. Zero external deps.

Every install is scanned automatically. The built-in security scanner checks for dangerous commands, data exfiltration, obfuscated code, suspicious URLs, credential exposure, prompt injection, and remote code execution.

Dangerous commands
Data exfiltration
Obfuscated code
Suspicious URLs
Credential exposure
Prompt injection
Remote code execution
security audit
$ agentver audit ./my-skills/
scanning 12 files across 3 skills...
 
✓ deploy-checker PASS (0 issues)
✓ code-reviewer PASS (0 issues)
⚠ untrusted-skill WARN (2 medium)
line 14: base64 decode detected
line 31: suspicious URL pattern
 
2/3 passed, 1 warning. no critical issues.
$
Team collaboration

Collaborate like you already know how

If you know Git and GitHub, you already know how to collaborate on Agentver. Proposals are like pull requests. Forks work the way you expect. Role-based permissions keep things safe.

Organisations

Create teams with custom namespaces. Every org gets its own skills repository backed by Git.

Permissions

Customisable permission levels so the right people have the right access. Secure tokens for integrations.

Change proposals

Propose changes to skills with diffs. Review workflow with approve, request changes, and comments.

Fork and sync

Fork any skill to your namespace. Track upstream changes. Merge updates when you are ready.

Collections

Curate skill sets for onboarding, projects, or teams. Share with your organisation or publicly.

Audit logging

Every action is logged. Track who installed what, when proposals were merged, and all team activity.

Import gateway

Bring your skills from anywhere

Connect your existing sources. GitHub repos, GitLab projects, Bitbucket repositories, Google Drive folders, Microsoft OneDrive. Everything lands in Git — versioned and integrity-checked.

GH
GitHub
GL
GitLab
BB
Bitbucket
GD
Google Drive
OD
OneDrive
Developer experience

Powerful when you need it

21 CLI commands with automation support. A full API for custom integrations. Reliable, consistent installations every time.

CI/CD integration
$ agentver install --json | jq .installed
["deploy-checker@2.0.0", "code-review@1.3.0"]
 
$ agentver audit --json | jq .verdict
"PASS"
$
lockfile
$ cat .agentver/lockfile.json | jq .packages[0]
{
"name": "deploy-checker",
"version": "2.0.0",
"integrity": "sha256-a1b2c3...",
"resolved": "github.com/acme/..."
}
$