Skip to content

feat: add amplifier-agent-adapter-designer bundle - #93

Draft
manojp99 wants to merge 1 commit into
mainfrom
feat/amplifier-agent-adapter-designer-bundle
Draft

feat: add amplifier-agent-adapter-designer bundle#93
manojp99 wants to merge 1 commit into
mainfrom
feat/amplifier-agent-adapter-designer-bundle

Conversation

@manojp99

Copy link
Copy Markdown
Collaborator

What this bundle does

Provides /mode amplifier-agent-adapter-designer — a runtime mode that turns any Amplifier session into a focused workspace for designing adapters between host applications and amplifier-agent.

Why it lives in this repo

The bundle is tightly coupled to the engine: it knows the three integration surfaces (Python SDK, TypeScript SDK, chat-completions HTTP server), the three existing host adapter case studies (paperclip, nanoclaw, opencode), and the cross-cutting concerns every integrator hits (credentials, MCP config, bundle cache priming, protocol version pinning, env allowlist, workspace isolation). Shipping it inside amplifier-agent keeps the knowledge co-versioned with the engine that generates it.

Mechanism mix

  • Mode (user-facing entry point): /mode amplifier-agent-adapter-designer
  • Expert agent (adapter-design-expert) — context sink carrying the ~2,500-token integration reference
  • Two context files: thin awareness pointer (always-on) + full reference (agent-only)
  • Behavior YAML wires it all together

No recipe, no skill. Adapter design is a conversational, branching exercise — a recipe would impose rigid step-gating on something that needs to breathe.

Context discipline

Layer When loaded Approximate weight
adapter-design-awareness.md always-on (thin pointer) ~200 tokens
Mode body only while mode active ~900 tokens
integration-reference.md only inside expert agent's child session ~2,500 tokens

The full reference never touches the root session.

Verification

Behavioral model in docs/BEHAVIORAL_MODEL.md covers 10 scenarios — fresh activation, surface selection, case-study deep dive, cross-cutting concerns walkthrough, error handling, mode-not-active isolation, protocol skew. All pass.

Open items to resolve before ready-for-review

  1. Verify the @foundation:context/shared/common-system-base.md path matches the deployed foundation fork.
  2. Confirm lazy mode discovery works for composed bundles; if not, add hooks-mode search_paths config to behaviors/agent-adapter-designer.yaml.
  3. amplifier-agent-py install path currently says "not yet on PyPI, install from git source" — will need update when PyPI publishes.
  4. Verify protocol version 0.3.0 in integration-reference.md matches current SDK source.
  5. Verify displayMode (camelCase) matches actual amplifier-agent-ts API surface vs Python's display_mode (snake_case).
  6. Confirm bundle namespace agent-adapter-designer isn't already taken in the registry.

Draft status

These 6 open items should be resolved and the bundle exercised via test-drive before ready-for-review.

Bundle providing /mode amplifier-agent-adapter-designer — a runtime mode
that turns any Amplifier session into a focused workspace for designing
adapters between host applications and amplifier-agent.

🤖 Generated with [Amplifier](https://github.com/microsoft/amplifier)

Co-Authored-By: Amplifier <240397093+microsoft-amplifier@users.noreply.github.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant