nv:ops

Agent infrastructure: safety, measurement, scaling.

npx skills add johnnichev/nv-ops -g -y

3 skills for running AI coding agents safely at scale:

Skill	Command	What it does
nv:guard	/nv-guard	Guardrails: permission tiers, destructive action blocking, secret remediation, auto-checkpoints, graduated trust (L0-L3), audit trails. "Make all mistakes recoverable."
nv:eval	/nv-eval	Agent evaluation: config quality scoring (A-F grading), Quality Flywheel, before/after measurement, compound failure math.
nv:team	/nv-team	Multi-agent orchestration: hub-spoke topology, worktree isolation, sequential merging, validation as control plane.

Why You Need This

Real AI agent incidents:

• Replit wiped a production database
• Claude ran rm -rf ~/
• Cursor deleted 70 files in Plan Mode
• Google Antigravity deleted an entire D: drive

The fix isn't "prevent all mistakes." It's "make all mistakes recoverable."

nv:ops sets up the infrastructure that makes agent autonomy safe: bootstrap templates for permissions/hooks, auto-checkpoints before every edit, destructive action blocking, secret remediation, graduated trust levels.

Research-Backed

Finding	Source
85% per-step accuracy = 20% success over 10 steps (compound failure)	OpenAI Agents docs
Anthropic's C compiler: 16 agents, 2,000 sessions, 100K lines of code	Anthropic engineering blog
Mesh topology breaks at 6-8 agents — use hub-spoke for more	Multi-agent research
Validation is the control plane (not the agent, not the human — the tests)	Anthropic C compiler

Install

npx skills add johnnichev/nv-ops -g -y

Installs all 3 skills globally. Works with Claude Code, Cursor, GitHub Copilot, Windsurf, Aider, Gemini CLI, and 9+ other tools.

Use

/nv-guard     # Set up safety infrastructure (do this first)
/nv-eval      # Score and improve agent configs
/nv-team      # Coordinate multiple agents on parallel tasks

See Also

• nv:context — Set up context engineering for your repo (the foundation)
• nv:dev — Plan, test, debug workflow
• nv:design — Professional web design with AI

License

MIT

---

Built by johnnichev. For engineers who ship.