creating-amplifier-modules Skill

Source

• Creator: microsoft
• Repository: microsoft/amplifier-foundation
• Skill file: skills/creating-amplifier-modules/SKILL.md

What it does

Use when creating a new Amplifier module (tool, hook, orchestrator, context, or provider). Covers the mount() contract, protocol compliance validation, placeholder patterns, and the module directory structure. Prevents the common mistake of creating no-op mount() stubs that fail protocolcompliance validation.

Skill instructions

Creating Amplifier Modules Overview Amplifier modules are Python packages that extend the runtime with tools, hooks, providers, and other capabilities. Every module has a mount() function that runs at session startup. Core principle: mount() must register something with the coordinator. A mount() that logs and returns None WILL fail validation. This matters immediately. Module validation runs every time a session loads. A broken mount() prevents any agent using the behavior from spawning — not just future agents, not "when Phase 2 is done" — right now, on every invocation. --- The Iron Law mount() MUST call coordinator.mount() or return a Tool instance. A mount() that returns None and calls nothing WILL FAIL with: "protocolcompliance: No tool was mounted and mount() did not return a Tool instance" "Stub" means placeholder that satisfies the protocol — not empty function that does nothing. --- Module Directory Structure modules/tool-{name}/ ├── pyproject.toml name = "amplifier-module-to