Glossary

Short meanings of the words used in the zn-gate docs. Where a page explains a word in more detail, the word links to it. Not sure where to begin? Start with Install.

Your setup

• Coding agent: a program such as Claude Code, Codex CLI, Cursor or OpenCode that reads your files and runs commands for you. See Install.
• Terminal: the window where you type commands. Run zn-gate's setup commands in your own terminal, not through an agent.
• Node.js: the program that runs zn-gate's npm commands. npx comes with it. You can get it at nodejs.org.
• MCP server: a tool your agent can call. MCP is the standard way agents connect to tools. zn-gate can also run as one. See MCP clients and tools.
• Plugin: the zn-gate add-on for Claude Code. It sets up zn-gate's hooks and its MCP servers in one step. See Claude Code.
• Keychain: the password store of your operating system. zn-gate can keep an optional API key there. See Connect.

Secrets

• Secret: a value that gives access to something, such as an API key, a token or a password.
• .env file: a file in a project that holds secrets and other settings for your app.
• Vault: an encrypted store on your machine where zn-gate keeps the real values of your secrets. See Scrubber and secrets vault.
• Marker: a stand-in for a secret, such as __ZN_STRIPE_KEY__. Your agent sees the marker. The real value stays in the vault. See Scrubber and secrets vault.
• Redact: replace a secret in some text with a marker or a label, so nobody who reads the text sees the value. See Secret redaction.
• Shadow (env shadow): move the secret values of a .env file into the vault and leave markers in their place. See Scrubber and secrets vault.
• Scrubber: the part of zn-gate that replaces secrets with markers in what your agent reads, runs and sends. Values stay in the vault. See Scrubber and secrets vault.

Checks and protection

• Hook: a small check that your agent runs before or after it reads a file, runs a command or gets your prompt. zn-gate uses hooks to block or redact. See How zn-gate works.
• Proxy (shield): a layer that sits between your agent and an MCP server and checks what passes through. See Protect MCP servers.
• Shadow mode (--shadow): a way to roll zn-gate out in which decisions are logged but nothing is blocked or changed. Not the same as env shadow above. See How zn-gate works.
• Prompt injection: text that tries to make an agent ignore your instructions, for example hidden inside a file or a web page. See Threat model.
• Classifier: a model that sorts text into groups, such as safe or an injection attempt.
• Galvanize: zn's prompt-injection checks. They run locally as rules, and the open Galvanize 2 classifier is available as an optional second opinion. See Galvanize 2.
• Cloud check: an optional second check, by the Galvanize 2 classifier at usezn.com, of the text sent to zn-gate's analyze_prompt tool. It is off until you turn it on. See What leaves your machine.

Commands and records

• Self-test (zn-gate test): runs zn-gate's built-in examples.
• Reach (zn-gate reach): a read-only report of what your coding agents can reach on this machine. It shows names, never values. See Reach.
• Audit (zn-gate audit): reads the agent configuration files of a repository, without running anything, and reports what would run. See Audit.
• Scan (zn-gate scan): looks through the prompts, Markdown, skills and config files of a folder for leaked secrets and prompt-injection traps. See Scan.
• Evidence (the local audit log): a record of each shield decision on your machine. Each entry is chained to the one before it, so changes can be detected. See Local audit log.