Author primitives

A package’s .apm/ directory contains primitives. Each primitive type is an opinion about how a piece of agent context should be authored once and projected onto every supported target (Copilot, Claude, Cursor, OpenCode, Codex, Gemini, Windsurf).

Primitive types

Primitive	One-liner	Page
Skills	Self-contained capability bundles with SKILL.md + scripts + assets	Skills
Prompts	Reusable prompt templates with frontmatter	Prompts
Instructions	Long-lived behavior rules (style guides, conventions)	Instructions and agents
Agents	Personas with explicit scope, tools, and triggers	Instructions and agents
Hooks	Event handlers fired by the runtime (pre-commit, on-tool-use, …)	Hooks and commands
Commands	Slash-command shortcuts the developer types into the agent UI	Hooks and commands
MCP servers	Tool-server declarations consumers can wire into their harness	MCP as a primitive

On-disk layout

.apm/
  skills/
    my-skill/
      SKILL.md
      scripts/
      references/
      assets/
  prompts/
    review.prompt.md
  instructions/
    style.instructions.md
  agents/
    cli-logging-expert.agent.md
  hooks/
    pre-commit.hook.md
  commands/
    deploy.command.md

Every primitive type follows the same pattern: a markdown file (or directory containing a primary markdown file) with frontmatter declaring its name and its trigger conditions. apm compile reads .apm/, applies any policy, and writes per-target output to the right directories on the target’s filesystem.

Recommended reading order

1. Skills — the densest primitive type and the one most newcomers hit first.
2. Prompts — next-most-common; how reusable templates work.
3. Instructions and agents — when behavior should persist across sessions.
4. Hooks and commands — runtime extension points.
5. MCP as a primitive — when your package needs to bring along a tool server.

When you’re ready to ship, continue to Compile and Pack a bundle.