Skip to content

Target Platforms

PromptScript includes 50 coding agent targets and two prompt targets. All coding agent targets receive project instructions. Rich formatters additionally emit native skills, agents, commands, MCP configuration, hooks, workflows, or plugin manifests when their platform supports those concepts.

Platform Families

Rich Native Formatters

Hand-written formatters model platform-specific capabilities:

  • Claude Code
  • GitHub Copilot
  • Cursor
  • Factory AI
  • Codex
  • Gemini CLI
  • OpenCode
  • Antigravity
  • Grok

AGENTS.md Targets

Targets using the shared AGENTS.md instruction contract:

  • Aider
  • Amazon Q
  • Warp
  • Zed
  • Jules
  • Devin
  • Kimi
  • Mimo
  • Deep Agents
  • ForgeCode
  • Hermes Agent

Markdown Instruction Targets

Targets using platform-specific instruction paths with shared rendering:

  • Windsurf
  • Cline
  • Roo Code
  • Continue
  • Augment
  • Goose
  • Kilo Code
  • Amp
  • Trae
  • Junie
  • Kiro CLI
  • Cortex
  • Crush
  • Command Code
  • Kode
  • MCPJam
  • Mistral Vibe
  • Mux
  • OpenHands
  • Pi
  • Qoder
  • Qwen Code
  • Zencoder
  • Neovate
  • Pochi
  • Adal
  • iFlow
  • OpenClaw
  • CodeBuddy
  • GitLab Duo

Configure Targets

Simple configuration:

id: multi-target-project
syntax: '1.4.0'
targets:
- github
- claude
- cursor
- factory
- codex

Feature-rich configuration:

id: rich-target-project
syntax: '1.4.0'
targets:
- github:
version: full
- claude:
version: full
- cursor:
version: full
- factory:
version: full
rulesMode: split
- codex:
version: full
maxThreads: 6
maxDepth: 2
- gemini:
version: full

Output Modes

Most rich formatters expose three modes:

ModeTypical output
simpleMain instruction file
multifileMain file plus rules, commands, prompts, or skills
fullRichest target-native output, including agents and integrations where supported

Additional modes:

  • Cursor: modern, multifile, legacy, agents-md, full
  • Antigravity: simple, frontmatter, agents-md
  • Factory: rulesMode: monolith|split
  • AGENTS.md targets: optional experimental frontmatter

Capability Model

Target support is explicit and varies by native platform contract:

CapabilityTypical target output
InstructionsMain Markdown or rules file
Scoped rulesNested rule or instruction files
SkillsNative SKILL.md directories or inlined guidance
AgentsNative Markdown, TOML, or droid files
CommandsPrompt, command, or shortcut files
MCP serversNative JSON or TOML configuration
HooksNative settings or hook configuration
WorkflowsNative workflow files
PluginsNative plugin manifests where implemented

Use the formatter matrix for exact output paths and capability support.

Scoped Monorepo Output

Build profiles place target output under package or application directories:

builds:
backend:
entry: .promptscript/backend.prs
output: services/backend
targets:
- factory
- codex:
agentsFile: AGENTS.override.md
frontend:
entry: .promptscript/frontend.prs
output: apps/frontend
targets:
- cursor:
version: full

Compile one profile or all profiles:

Terminal window
prs compile --build backend
prs compile --all-builds

Capability Portability

Portable PromptScript source does not force every platform into one schema. Instead:

  1. Shared concepts remain platform-neutral.
  2. Validators check field shape and target options.
  3. Formatters emit native files for supported capabilities.
  4. Unsupported target-specific behavior reports an actionable compatibility warning.
  5. Main instruction output remains available across all targets.

Lifecycle hooks have an exhaustive 50-target capability registry. Native project hooks are currently emitted for Claude, Codex, Cursor, Factory, Gemini, GitHub, Grok, and Windsurf. Plugin-only, custom-agent-scoped, unsupported, and incompatible output modes report PS4002 rather than silently omitting enabled hooks. See the hook capability matrix.

Review generated output and compatibility warnings when adopting a new target or target version.

Hermes Agent

Hermes Agent discovers project-local AGENTS.md files as workspace instructions. See the official context-files documentation.

PromptScript capabilityHermes outputContract
Supported instruction blocksAGENTS.md@identity, @context, @standards, @knowledge, @restrictions, and @examples render into the single file
simple, multifile, fullAGENTS.mdAll aliases intentionally emit the same file
SkillsNoneNo verified project-local Hermes skill output path; see the official skills documentation
Agents, commands, and workflowsNoneNo verified project-local native contract
Prompt definitions (@prompts)NoneNot instructions for a coding agent, so the block is skipped without a warning
Scoped rules (@guards) and local filesNoneNo verified project-local native contract
Hooks, MCP servers, pluginsNoneNo verified project-local native contract; PS4002 compatibility warnings include source locations

PromptScript does not create .hermes.md, native skill directories, agent files, command files, workflow files, prompt files, hook files, MCP configuration, or plugin manifests. Unsupported blocks other than @prompts are omitted with actionable non-fatal compatibility warnings. Move required guidance into supported AGENTS.md instruction blocks.

GitLab Duo

GitLab Duo reads the root AGENTS.md file (GA in GitLab 18.8) and Agent Skills from skills/<name>/SKILL.md at the repository root (GitLab 18.10+). The GitLab UI surfaces the instructions from GitLab 18.11. See the official AGENTS.md documentation.

PromptScript capabilityGitLab Duo outputContract
Supported instruction blocksAGENTS.mdShared markdown rendering into the single root file
simple and multifileAGENTS.mdSingle file; skills stay full-mode-only
fullAGENTS.mdPlus skills/<name>/SKILL.md for each skill
Agents, commands, workflows, and promptsNoneNo verified project-local native contract
Scoped rules (@guards) and local filesNoneNo verified project-local native contract
Hooks, MCP servers, pluginsNoneNo verified project-local native contract; PS4002 compatibility warnings include source locations

Prompt Targets

Two further targets are not coding agent targets. They compile the prompts of an application. See Application Prompts.

TargetOutputContent
bundleprompts/<key>.json and prompts/index.jsonOne content-hashed JSON bundle per @prompts entry
prompt-textprompts/<key>.md and prompts/README.mdA Markdown preview of each composed system message

Prompt targets are opt-in. No default selection includes them: not compile(), not compileAll, not the browser compiler, and not the prs init target picker. Name them in targets to use them. They write nothing for a program without @prompts entries, and they need syntax 1.6.0. They have one version, v1, and no skills, agents, commands, hooks, or other resources.