Imported from lance0/nbox (SKILL.md). Install upstream with npx skills add lance0/nbox. Copyright stays with the author.

nbox — NetBox from the shell

nbox is a CLI / TUI / MCP server for NetBox (DCIM + IPAM). Use it to answer questions about network inventory and addressing without clicking through the NetBox web UI. Reads are the default; seven safe-write commands (interface description, device status, IP/prefix/ip-range reserve, tag add/remove) are available behind --allow-writes + confirmation.

When to use this skill

Reach for nbox when the user wants to:

• look up a device, interface, IP, prefix, VLAN, site, rack, circuit, provider, aggregate, ASN, IP range, tenant, contact, VM, cluster, VRF, or route target;
• search NetBox across object types in one query (e.g. "find anything matching edge01");
• find a free IP or prefix, check prefix utilization, or trace a cable path;
• pull NetBox data as JSON/CSV for a script, audit, or report.

Prerequisites

Assumes nbox is installed and pointed at a NetBox instance. Verify with nbox status (reports connectivity + the NetBox version). If it fails with exit code 3 (auth) or a missing-config error, the user still needs to install nbox and configure a profile + token — see the README.

Core usage

Always pass a subcommand and request machine-readable output:

nbox --no-tui device edge01 -o json            # one device, full detail
nbox --no-tui interface edge01 swp25 -o json   # an interface + its cable path
nbox --no-tui ip 10.0.0.1 -o json              # an IP → most-specific parent prefix
nbox --no-tui prefix 10.0.0.0/24 -o json       # a prefix (children, utilization)
nbox --no-tui search edge01 -o json            # cross-object search (ranked, deduped)
nbox --no-tui next-ip 10.0.0.0/24 -o json      # next free address in a prefix
nbox --no-tui raw GET dcim/devices/?limit=1    # escape hatch (path with or without /api/)

Output flags:

• --no-tui — guarantee a non-interactive run (any invocation that would launch the TUI exits with a usage error instead of blocking on a terminal). Always pass this.
• -o json / --json — JSON to stdout; --raw for one compact line.
• -o csv — CSV for list/search results (single objects are rejected — use --json).
• --fields a,b,c — keep only those top-level fields; --envelope wraps as { "schema_version": 1, "data": … } for stable parsing.

stdout carries only the requested data; logs/diagnostics/errors go to stderr. Exit codes are stable: 3 auth, 4 not-found, 5 ambiguous. Full command + flag reference: AGENTS.md.

Read domains

The read surface is split into focused skill files — reach for the one that matches the question:

• Search — ranked, deduped cross-kind nbox search: the entry point when the kind or exact reference isn't known yet; the one-scope-filter-at-a-time rule and the search→detail-lookup chain
• IPAM read — nbox ip / prefix / vlan / ip-range / aggregate / vrf / route-target, prefix utilization and the tree, the read-only next-ip / next-prefix previews, and --vrf disambiguation
• Device context — nbox device / interface (the cable-path A↔Z trace) / mac reverse-resolve, plus rack / site context
• MCP server — nbox serve as the MCP server (read-only by default; local stdio writes use --local-writes; shared HTTP/OIDC writes use --allow-writes + caller nbox:write + [serve.vault]; HTTP local/profile-token writes are rejected): stdio vs HTTP+OIDC, the read tools and resources, the prompts catalog, and --print-config

Safe writes

nbox can modify NetBox through seven plan-first, dry-run/confirm-gated write commands (ADR-0001). Reads stay the default — a write never happens without explicit opt-in. See the focused skill files:

• Safe writes — the universal dry-run / confirm / audit lifecycle shared by all seven write commands
• IPAM allocate — ip reserve, prefix reserve, ip-range reserve (POST to available-ips / available-prefixes)
• Tag writes — tag add, tag remove (PATCH the tags array on any object kind)
• PATCH writes — interface set description, device set status (minimal PATCH with ETag/If-Match concurrency)

The recommended agent pattern is a two-step dry-run-then-apply:

nbox --no-tui <cmd> ... --dry-run --json          # preview the plan
nbox --no-tui <cmd> ... --allow-writes --confirm --json  # apply