[&] Protocol — AmpersandBoxDesign

Open, language-agnostic specification for capability composition in AI agents. Defines how agents declare memory, reasoning, time, space, body (sensorimotor), and governance capabilities that compile into MCP and A2A configurations.

Implementation prompts (source-of-truth for code generation)

• prompts/PROTOCOL_PROMPT.md — the full implementation prompt used to generate [&] Protocol reference code. Read this before modifying the Elixir reference implementation — it contains the project identity, architecture summary, formal grammar, coding standards, and generation rules.

• prompts/GRAPHONOMOUS_PROMPT.md — autonomous codebase traversal prompt for Graphonomous MCP integration.

• prompts/KAPPA_BUILD_PROMPT.md — κ cyclicity detection implementation guide (v2). Canonical schema, field name rules, directed edge rules. Proved on 1,926,351 finite systems.

• prompts/KAPPA_DELIBERATOR_PROMPT.md — deliberation loop architecture triggered by κ > 0 routing. Decompose → focus → reconcile pipeline.

• prompts/ATTENTION_ENGINE_PROMPT.md — proactive attention engine: survey → triage → dispatch → execute → reflect. The "ignition" for autonomous behavior.

• prompts/MODEL_TIER_PROMPT.md — hardware-adaptive tier budgets (local_small/8B, local_large/70B+, cloud_frontier). Same topology, different depth.

Prompt dependency order:

PROTOCOL_PROMPT → KAPPA_BUILD → KAPPA_DELIBERATOR → ATTENTION_ENGINE → MODEL_TIER

These prompts serve the same role as docs/spec/ in other portfolio projects. When the protocol or reference impl needs changes, align with the prompt first.

Build and verify (Elixir reference implementation)

cd reference/elixir/ampersand_core
mix deps.get
mix escript.build
mix test

CLI usage

cd reference/elixir/ampersand_core
./ampersand validate ../../../examples/infra-operator.ampersand.json
./ampersand compose ../../../examples/infra-operator.ampersand.json
./ampersand generate mcp ../../../examples/infra-operator.ampersand.json
./ampersand generate a2a ../../../examples/infra-operator.ampersand.json

Repository structure

• SPEC.md — protocol specification (draft v0.1.0)

• protocol/schema/v0.1.0/ — JSON Schema artifacts (ampersand.schema.json, capability-contract.schema.json, registry.schema.json)

• examples/ — reference agent declarations (*.ampersand.json)

• contracts/v0.1.0/ — capability contract definitions

• reference/elixir/ampersand_core/ — Elixir reference implementation + CLI

• sdk/npm/validate/ — @ampersand-protocol/validate npm package

• sdk/python/ampersand_protocol/ — ampersand-protocol Python SDK

• capabilities/ — individual capability documentation

• docs/ — quickstart, runtime walkthrough, FAQ, comparison table

• site/ — website source files

• playground/ — browser-based validation playground

Protocol change rules

When changing the protocol, update ALL affected layers together:

1. SPEC.md

2. protocol/schema/v0.1.0/*.schema.json

3. examples/*.ampersand.json

4. Reference implementation logic

5. CLI output

6. docs/

A protocol change is not complete if only the prose changes. See CONTRIBUTING.md.

Core model

Six capability primitives: &memory, &reason, &time, &space, &body, &govern Two composition operators: & (combine) and |> (pipeline) Capabilities are interfaces, providers are implementations. [&] does not replace MCP or A2A — it compiles into them.

&body is the sensorimotor primitive (added in draft v0.1.0) describing the agent's instantiation in an environment — perception, typed action, affordance enumeration, state encoding for replay. See OS-011 (Embodiment Protocol) for the behavioral loop that consumes &body.*.

Schema targets

JSON Schema draft 2020-12. Prefer explicit validation rules over descriptions.