Glossary

Plain-English meanings of the words Prismor and its dashboard use. Each entry links to the page that covers it in depth. New to Prismor? Start with Getting started.

Agent

An AI coding assistant that can take actions on your computer, such as Claude Code, Codex, Cursor or Copilot. Prismor watches what agents do. Each agent registers itself the first time it runs in a project Prismor protects. See Dashboard: Agents tab.

Allow / warn / block

The three basic outcomes of a check. Allow means the action runs. Warn means it runs, but a finding is recorded so you can see it. Block means it is stopped before it runs, and the agent is told why. A rule only blocks in enforce mode. See Rule actions.

Audit trail

A tamper-evident record of every action an agent took through Prismor, kept in ~/.prismor/audit/. Each record is linked to the one before it and signed, so an edited or deleted record shows up when you run prismor trail verify. See Signed audit trail.

Cloak / secret cloaking

A way to keep real passwords and API keys out of the agent's view. You store a secret once with prismor cloak add <name>, and the agent only sees a placeholder such as @@SECRET:name@@. The real value is filled in when a command runs and scrubbed from the output. See Sweep and Cloak.

Egress

Network traffic leaving your machine, such as an agent fetching a web page or sending data to an API. Prismor's egress policy is a list of hosts agents may or may not contact. See Network isolation.

Enforce mode

The setting where Prismor stops actions that match a blocking rule, before they run. Compare observe mode. See Modes.

Extension

Anything that adds instructions or code to an agent from outside: a skill, a plugin, a hook command or an MCP server. Prismor keeps a list of them and flags new or changed ones until someone reviews them. See Extension ledger.

Finding

A record that a rule matched an action. A finding has a severity (critical, high, medium or low) and does not always mean something was blocked: in observe mode, or for warn-only rules, the action still ran. See Dashboard.

Hook

A small command an agent runs at set moments, which lets Prismor see and judge what the agent is about to do. prismor setup or prismor install-hooks installs them. The main moments, using Claude Code's names:

• PreToolUse: just before a tool call runs. This is where Prismor can block.
• PostToolUse: just after a tool call, with its result.
• UserPromptSubmit: when you send the agent a message.
• SessionStart: when a new agent session begins.

Other agents use their own names for the same moments. See Governance surfaces.

Inherit (global)

An agent setting on the dashboard's Agents page. inherit (global) means the agent follows the mode set by policy (organization, project or global) instead of its own observe or enforce override. See Dashboard: Agents tab.

MCP server

A separate program that gives an agent extra tools, for example to talk to GitHub or a database. MCP stands for Model Context Protocol. Prismor can sit in front of MCP servers and check each call. See MCP gateway.

Mode

A named security setup that sets many options at once. Prismor ships three: dev-safe, trusted-workspace and regulated-airgap. "Mode" is also used for the simpler choice between observe and enforce. See Modes.

Observe mode

The setting where Prismor checks and records every action and reports anything risky, but blocks nothing. A safe way to start. Compare enforce mode. See Modes.

Pause

Temporarily stops Prismor from blocking, while it keeps checking and recording. prismor pause lasts 24 hours by default; prismor resume ends it early. See Pause and resume.

Pinned rule

A rule shown with a lock on the Policy Control page. It belongs to Prismor's core safety floor or its self-protection rules, so you cannot turn it off for a project. See Making exceptions.

Plugin

A bundle that adds features to an agent, installed with the agent's own plugin command. Prismor lists plugins as one kind of extension. See Extension ledger.

Policy

The full set of rules and settings Prismor applies. Your project's version is the file .prismor/policy.yaml. It sits on top of Prismor's defaults, and an organization's policy can sit on top of both. See Policy layers and exemptions.

Risk score

A number from 0 to 100 for a session. It goes up with each finding, more for serious ones, and stops at 100. Use it to sort sessions and look at the riskiest first. See Dashboard.

Rule

One check in the policy, for example "do not read .env files". A rule has an id, a severity and an action (block, warn and so on). See Policy engine.

Rules fired

A dashboard count of every time a rule matched an action in the selected period, whether or not it blocked anything. One action can fire several rules. See Dashboard.

Scope / scoped agent / clear scope

At the start of a session, Prismor reads your first message and builds a small set of extra rules for that task only. This is the session's scope, and the feature is called the scoped agent. Clear scope (on the Sessions page, or prismor scope clear <id>) removes those session rules; your project and global policy still apply. See Scoped agent.

Session

One run of an agent, from start to finish. Prismor records every action in a session, so you can replay it later with prismor session <id> or on the Sessions page. See Dashboard.

Skill

A file of instructions, usually named SKILL.md, that an agent reads and follows. Because it can tell an agent what to do, Prismor tracks skills and flags new or changed ones. See Skill scanner and Extension ledger.

Supply chain

The packages and libraries your project installs from places like npm and PyPI. A bad package can run harmful code when installed. Prismor checks installs before they run. See Supply chain.

Tool call

A single action an agent takes, such as running a shell command, reading a file, writing a file or fetching a web page. Prismor checks each tool call. See Prismor.

Tool tag

A label that says what a tool does, such as untrusted_content (it reads outside input) or critical_action (it sends, publishes or deletes). Tag rules then forbid risky combinations, such as reading untrusted content and then acting on it. See Tool tags.

Turn

In a session timeline, one turn is one message you sent and everything the agent did in response, until your next message. See Dashboard.

Workspace

A project folder Prismor protects. Each one is registered when you install hooks in it, so the dashboard can show all of them together. A workspace can be personal or, on an enrolled machine, org-managed. See Dashboard: where the data lives and Connecting to the platform.