Glossary

Every term used in the course, in one place. ADK terms first, then external systems.

@-mention

The @ prefix in a Claude Code prompt that pulls a file or URL into the context window. Examples: @configs/core-nyc-01.cfg, @configs/ (whole directory), @https://... (remote URL). Claude reads the entire referenced content before answering.

Agent Development Kit (ADK)

The five-layer architecture of Claude Code: CLAUDE.md (memory), Skills (knowledge), Hooks (guardrails), Subagents (delegation), Plugins (distribution). Plus MCP on the side for external integration.

CLAUDE.md

A markdown file Claude Code reads automatically at the start of every session. Lives at ~/.claude/CLAUDE.md (global), ./CLAUDE.md (project root), or in subdirectories. More specific overrides more general. Use it for vendor stack, naming conventions, IP plans, and safety rules.

Headless mode

Running Claude Code non-interactively with the -p flag. No terminal UI, no permission prompts. Used in CI/CD pipelines for compliance scans and pre-merge config review. Always restrict tools to read-only in headless mode.

Hook

A deterministic shell command that fires on a specific Claude Code event. Not AI. Not probabilistic. Configured in .claude/settings.json under hooks. Events include PreToolUse, PostToolUse, SessionStart, PreCompact, PostCompact. Exit code 2 from a PreToolUse hook blocks the tool call.

MCP (Model Context Protocol)

The protocol Claude Code uses to talk to external systems. An MCP server is an adapter that translates Claude's tool calls into API requests against NetBox, Grafana, Postgres, GitHub, etc. Add via claude mcp add. Servers can be project-scoped (shared with team) or user-scoped (personal credentials).

Permission lists

Three lists in .claude/settings.json under permissions: allow (run without prompting), deny (never run), ask (require approval each time). The deny list is absolute — it blocks even after accidental approval. Use it for configure terminal, write memory, terraform apply.

Plugin

An installable bundle that includes CLAUDE.md, skills, hooks, subagents, and slash commands. Installed via /plugin install ./path or from a marketplace (/plugin marketplace add <url>). Used to distribute team-wide Claude Code configuration.

Sandbox mode

OS-level restriction that limits Claude Code to the project directory. Cannot access SSH keys, .env files, or anything outside the project tree. Activated with /sandbox. Used when analyzing untrusted input.

Skill

A directory under .claude/skills/{name}/ containing a SKILL.md with frontmatter (name, description, allowed-tools). Loaded on demand when Claude matches the task to the skill description. Can include reference scripts, templates, and an isolated context.md.

Slash command

A reusable prompt invoked with /{name}. Built-ins include /compact, /clear, /resume, /cost, /model, /export, /branch, /sandbox. Custom commands live in .claude/commands/{name}.md.

Subagent

A Claude Code agent with its own context window, separate from the main session. Defined in .claude/agents/{name}.md with frontmatter (name, description, allowed-tools, model). Used for parallel investigation across devices, sites, or domains. The main session sees only the subagent's results.

/compact

Slash command that summarizes the current conversation to free context window space. Run at 60% capacity. Past 85% Claude loses precision; past 90% responses become unreliable.

NetBox

An IPAM/DCIM source-of-truth tool. Connected to Claude Code via an MCP server, it lets Claude query device interfaces, IP allocations, cable paths, and circuit data without manual export.

BGP / OSPF / MD5 authentication

Routing protocols and a peer-authentication mechanism the course uses as examples. The bgp-troubleshoot skill demonstrates how to encode a BGP failure-mode decision tree as institutional knowledge.

Transceiver Rx/Tx power

Optical transceivers report receive (Rx) and transmit (Tx) power in dBm. Healthy Rx is typically above -14 dBm; falling Rx with healthy remote Tx indicates degraded optics on the local end. The course's flagship Section 12 example diagnoses a peering link failure from these values.

ServiceNow / change ticket

Change-management ticket. The course's CLAUDE.md example mandates that every production change references a ticket, includes pre/post validation, and documents rollback steps. Claude Code can draft these tickets from troubleshooting context.

Worktree

A git feature that creates a separate working directory for a different branch. Used in Section 8 to isolate parallel changes (BGP config, firewall rules, monitoring templates) so each lands as its own change ticket.