Glossary
Short definitions of the words used across this documentation. Each term links to the page that explains it in full.
Files and Build
PromptScript
A language and a compiler for AI coding agent configuration. You describe instructions, skills, agents, and tool integrations once, and the compiler writes the files each AI tool reads. See Getting Started.
.prs file
A PromptScript source file. A project usually keeps them in .promptscript/, with
project.prs as the entry file. One project can use many .prs files that import and
inherit each other. See Multi-File Organization.
promptscript.yaml
The project configuration file. It lists the entry file, the targets to compile, the registry, validation rules, and policies. See Configuration.
Block
A named section of a .prs file that starts with @, for example @identity,
@standards, @skills, or @mcpServers. Every block has a fixed shape. See
Block Shapes and the Language Reference.
Target
One AI tool that PromptScript compiles for, such as claude, github, cursor, or
codex. There are 50 built-in targets. See Supported Formatters.
Formatter
The part of the compiler that writes the files for one target. It maps each block to the native format of that tool. See Formatter Architecture.
Target version (mode)
How much a target writes. simple writes the main instruction file, multifile adds
rules, commands, or skills, and full writes the richest output, including agents and
integrations where the tool supports them. See Target Platforms.
Compile
Turning .prs sources into native files with prs compile. The output is deterministic,
so the same source always gives the same files. See CLI Reference.
Validate
Checking sources with prs validate without writing output: syntax, block shapes,
references, policies, and the security scan. See CLI Reference.
Syntax version
The syntax field in @meta. It tells the compiler which language features the file
uses. prs upgrade moves files to the newest version. See
Versions and Diagnostics.
Diagnostic code
A code such as PS005 or PS012 on a validation message. The code identifies the rule, so
you can search for it. See FAQ & Troubleshooting.
Reuse and Composition
Inheritance (@inherit)
One .prs file builds on another, for example a project on a team layer and the team
layer on an organization layer. A file has at most one @inherit. Parameters can be passed
to the parent. See Inheritance.
Import (@use)
Merges a fragment into the current file, like a mixin. A file can have many @use
lines, and an alias makes the fragment available to @extend. See
Composition and Precedence.
Parameters
Values a parent or fragment accepts, declared in @meta { params: ... } and passed as
@inherit @stacks/react-app(projectName: "Checkout"). See
Inheritance.
@extend
Changes an inherited or imported block in place: adds fields, appends to lists, or replaces values. See Merge and Replacement.
@override
Replaces a whole existing value instead of merging into it. See Merge and Replacement.
Overlay
A higher layer that uses @extend to customize something a lower layer defined, most
often a skill. See Skill Overlays.
Sealed property
A skill property that higher layers cannot override, set with sealed. See
Skill Overlays.
Markdown import
Using existing Markdown files, such as a SKILL.md, as PromptScript sources. See
Markdown Imports.
Agent Capabilities
Instructions
The rules an AI agent follows, written in blocks such as @identity, @context,
@standards, and @restrictions. See Agent Platform.
Skill
A reusable capability package: instructions plus optional references, scripts, assets, and input and output contracts. Targets that support skills get them as native skill files. See Skills and Resources.
Skill composition
Building a skill from other skills and shared parts instead of copying text. See Skill Composition.
Skill contract
The declared inputs and outputs of a skill. See Skill Contracts.
Agent (subagent)
A specialized assistant with its own instructions, tools, model, skills, and MCP access,
defined in @agents. See Agents.
MCP server
A Model Context Protocol server that gives an agent extra tools, defined in @mcpServers
and written to each target’s native MCP configuration. See
MCP Servers and Plugins.
Plugin
A bundle of skills, agents, hooks, and MCP servers that installs as one unit, defined in
@plugins. See MCP Servers and Plugins.
Hook
A command that runs on an AI tool event, for example after the agent edits a file. Defined
in @hooks. See Hooks and Workflows and AI Tool Hooks.
Workflow
A named, repeatable procedure defined in @workflows. See
Hooks and Workflows.
Shortcut
A user command, like a slash command, defined in @shortcuts. See the
Language Reference.
Guard
Rules in @guards that apply to matching files only, for example **/*.ts. Guards can
require other guards. See Guard Dependencies.
Model profile
A catalog entry that maps one model name to the name each target expects. See Model Catalog.
Teams and Organization
Registry
A shared collection of .prs files, usually a Git repository, that projects inherit
from. Create one with prs registry init and publish with prs registry publish. See
Build Your Registry.
Namespace
The @name prefix of a registry path, such as @company/base. See
Build Your Registry.
Lockfile (promptscript.lock)
Records the exact commit and hash of every remote import, so builds are reproducible.
Written by prs lock. See Build Your Registry.
Vendor directory
A local copy of all locked dependencies in .promptscript/vendor/, made by
prs vendor sync, for offline and air-gapped builds. See Build Your Registry.
Policy
A rule in promptscript.yaml that limits how layers may change each other, for example
which properties a project may override. See Policy Engine.
Security scan
Part of prs validate that looks for prompt injection, hidden encoded payloads, and
suspicious URLs in every source and import. See Security.