Agent Platform

PromptScript is an agent platform configuration language. A single .prs source defines the instructions, capabilities, integrations, and automation used by AI coding agents across a repository or organization.

The compiler translates that source into native files for 50 coding agent targets. Each target receives the formats and capabilities it supports, such as instruction files, native skills, custom agents, MCP configuration, lifecycle hooks, workflows, and plugins.

Platform Model

flowchart LR
  Source["PromptScript source"] --> Resolve["Resolve inheritance and imports"]
  Resolve --> Validate["Validate language and capabilities"]
  Validate --> Compile["Compile target-native output"]
  Compile --> Instructions["Instructions and policies"]
  Compile --> Skills["Skills and resources"]
  Compile --> Agents["Agents and tools"]
  Compile --> Integrations["MCP and plugins"]
  Compile --> Automation["Hooks and workflows"]

Platform capability	PromptScript source	Result
Instructions and policy	@identity, @context, @standards, @restrictions, @guards	Native instruction and scoped rule files
Reusable capabilities	@skills	Native skills, bundled resources, scripts, and contracts
Delegated specialists	@agents	Native subagent or custom-agent definitions
User commands	@shortcuts	Native prompts, commands, or documented shortcuts
Tool integrations	@mcpServers	Target-native MCP server configuration
Capability bundles	@plugins	Target-native plugin manifests where supported
Lifecycle automation	@hooks	Target-native hook configuration
Repeatable procedures	@workflows	Native workflow files where supported
Multi-package delivery	builds in promptscript.yaml	Scoped output for packages and applications

Complete Platform Definition

@meta {
  id: "checkout-service"
  syntax: "1.5.0"
  tags: ["payments", "typescript"]
}

@identity {
  """
  You are working on a payment service.
  Preserve transaction integrity and auditability.
  """
}

@standards {
  code: ["Use strict TypeScript", "Write tests for business rules"]
}

@skills {
  security-review: {
    description: "Review payment changes for security risks"
    allowedTools: ["Read", "Grep", "Bash"]
    content: "Inspect authentication, authorization, secrets, and payment data handling."
  }
}

@mcpServers {
  issue-tracker: {
    transport: "stdio"
    command: ["node", "./tools/issues.mjs"]
  }
}

@agents {
  reviewer: {
    description: "Review changes before merge"
    tools: ["Read", "Grep", "Glob", "Bash"]
    skills: ["security-review"]
    mcpServers: ["issue-tracker"]
    content: "Review changed code, tests, and operational impact."
  }
}

@hooks {
  validate-changes: {
    event: "post-tool-use"
    matcher: "Edit|Write"
    command: ["pnpm", "run", "typecheck"]
  }
}

@workflows {
  release: {
    description: "Validate and prepare a release"
    content: "Run quality gates, summarize changes, and prepare release metadata."
  }
}

@plugins {
  payment-engineering: {
    description: "Payment engineering capability bundle"
    version: "1.0.0"
    skills: ["security-review"]
    hooks: ["validate-changes"]
    mcpServers: ["issue-tracker"]
  }
}

Target support varies because each AI platform exposes a different native contract. PromptScript preserves one source model while formatters map supported capabilities to target-native output. Use the target platform matrix to choose versions and capabilities.

Explore Features

• Agents - specialized subagents, models, tools, skills, and MCP access
• Skills and Resources - portable capability packages with references and scripts
• MCP and Plugins - tool servers and reusable capability bundles
• Hooks and Workflows - lifecycle automation and repeatable procedures
• Target Platforms - 50 coding agent targets, two prompt targets, and output families

Compile One or Many Projects

Compile configured targets:

prs compile

Use named builds when one repository hosts several packages or apps and each needs its own agent configuration. Every build has its own entry file, output directory, and target list, so packages/api and packages/web each get files scoped to their subtree. Define them in promptscript.yaml (see the Configuration Reference):

builds:
  api:
    entry: .promptscript/packages/api.prs
    output: packages/api
    targets:
      - factory
      - codex
  web:
    entry: .promptscript/packages/web.prs
    output: packages/web
    targets:
      - cursor:
          version: full

prs compile --all-builds

Enterprise Control Plane

PromptScript applies the same lifecycle to every agent capability:

1. Store source in Git.
2. Resolve organization, team, and project layers.
3. Validate syntax, references, policies, and target options.
4. Compile deterministic target-native output.
5. Review generated changes in pull requests.
6. Enforce prs validate --strict and compilation checks in CI.

See Enterprise Setup, Security, Policy Engine, and CI/CD Integration for the pipeline that enforces it.