10 New Skills Audited: From SDLC Mastery to VM Control
Our latest audit batch covers everything from development methodology to infrastructure management. One skill scored a perfect 9.0/10. Another poses significant risk.
Every week, we audit OpenClaw skills and publish our findings. This batch of 10 skills covers a wide spectrum: productivity tools, infrastructure automation, messaging utilities, and even a baby tracker.
The Standouts
sdlc-skill
Software Development Lifecycle methodology
A documentation-only methodology skill that enforces disciplined software development practices. No network access. No file system writes. Just pure process guidance for Agile workflows.
What impressed us:
- ✓ Zero external dependencies
- ✓ No credential handling
- ✓ Read-only methodology enforcement
- ✓ Clear, documented processes
clawd-notify-skill
Cross-session messaging
Simple cross-session messaging for OpenClaw. It does one thing well: notify other sessions. The attack surface is minimal because the functionality is intentionally constrained. No external APIs, no credential storage.
lobster-jobs
Cron-to-workflow converter
This is clever: it transforms OpenClaw cron jobs into Lobster workflows with approval gates. The medium risk rating comes from its ability to execute jobs, but the approval mechanism adds a valuable safety layer. This is how you build automation with guardrails.
The Infrastructure Stack
Three skills target DevOps workflows, each with medium risk ratings that reflect their power.
vercel-skill
Vercel deployments
CLI wrapper for Vercel. Risk comes from deployment capabilities—this can push code to production.
supabase-skill
Database management
Handles Supabase credentials and can execute SQL. Read-only defaults help security.
ansible-skill
Infrastructure automation
Multi-body Yjs coordination. Powerful when needed, dangerous if misconfigured.
The High-Risk Outlier
kvm-manager-skill
KVM/libvirt VM management
This skill provides comprehensive KVM/libvirt virtual machine management: start, stop, create, and delete VMs. It requires sudo access.
The risk is straightforward: full VM lifecycle control means a compromised agent could destroy infrastructure. The 4.0/10 score reflects this power combined with the lack of additional safety mechanisms.
✓ When to use it
Homelabs, development environments, non-production infrastructure
✗ When to avoid it
Production systems, shared hosting, anywhere VM destruction would be catastrophic
The Productivity Suite
Email and cloud storage integrations with medium risk profiles.
email-send-skill
SMTP integration via nodemailer. The risk here is credential storage—SMTP passwords need to be handled carefully. Proper environment variable isolation is essential.
dropbox-skill
OAuth-based Dropbox integration with automatic token refresh. The medium risk comes from cloud storage access and the OAuth flow complexity.
The Curveball
huckleberry-skill
Baby tracker
A baby tracker that uses reverse-engineered API access to Huckleberry's service. We're not kidding—this exists, and it's surprisingly well-built. The risk comes from the unofficial API usage and credential handling, but the functionality is focused and well-scoped.
Score Distribution
| Score | Risk Level | Skills |
|---|---|---|
| 9.0 | 🟢 Low | sdlc-skill |
| 8.0 | 🟢 Low | clawd-notify-skill |
| 7.0 | 🟡 Medium | lobster-jobs |
| 6.5 | 🟡 Medium | email-send-skill, supabase-skill, ansible-skill, dropbox-skill |
| 6.0 | 🟡 Medium | vercel-skill, huckleberry-skill |
| 4.0 | 🔴 High | kvm-manager-skill |
Key Takeaways
Documentation-only skills are the gold standard
The sdlc-skill proves that powerful functionality doesn't require risky permissions. When a skill can do its job without network access or credential handling, it should.
Approval gates matter
The lobster-jobs skill demonstrates how to add safety to powerful automation. If your skill can execute actions, consider adding human checkpoints.
Sudo requirements need scrutiny
The kvm-manager-skill is well-built, but its sudo requirement makes it inherently high-risk. Users should understand exactly what they're granting before installing.
Full Audit Reports
Every skill audited has a detailed report available:
Browse all 85+ verified skills at secureskills.io/skills