A2A Campaign Harness Integration Contract — v0.7.0

GH issue: #718 (cross-repo). Harness repo: alphaonedev/ai-memory-a2a-v0.7.0. This repo (substrate): alphaonedev/ai-memory-mcp.

The A2A harness is maintained in a separate repository so that node provisioning, droplet wiring, multi-agent orchestration, and result aggregation can iterate independently of the substrate release cadence. This document pins the v0.7.0 contract the harness must honor when exercising this repo’s binary.

Cross-repo dependency

Direction	From	To	Contract
binary	`ai-memory-mcp` v0.7.0	`ai-memory-a2a-v0.7.0` setup script	scp-built `target/release/ai-memory` MUST be used; harness scripts MUST NOT overwrite via `AI_MEMORY_VERSION=0.6.0` tarball download. See #718 root cause §1.
config	`ai-memory-mcp` `config.toml` schema v3	harness rendered config	harness MUST set `schema_version = "3"` and respect every `[agents.defaults]` / `[permissions]` / `[federation]` block. See `src/config.rs`.
federation wire	`ai-memory-mcp` substrate	harness 2-of-N or 4-of-N nodes	`x-peer-id` header REQUIRED on every `/api/v1/sync/push` (#716 substrate cure). v0.7.0 also adds `X-Memory-Sig: ed25519=<base64>` on every outbound POST (#791); receivers MAY require it via `AI_MEMORY_FED_REQUIRE_SIG=1` (default in v0.7.0).
TLS	optional	harness	`TLS_MODE=mtls` requires `/etc/ai-memory-a2a/tls/server.pem`; `TLS_MODE=off` is acceptable for harness-only smoke runs (no real-world peer exposure).
identity	`AI_MEMORY_AGENT_ID` env	harness boot	every harness-spawned `ai-memory` process MUST set a unique `AI_MEMORY_AGENT_ID` (see CLAUDE.md §Agent Identity for the resolution ladder).

v0.8.0 substrate guarantees the harness can rely on

101 MCP tools at --profile full; 7 at --profile core + always-on memory_capabilities.
92 production HTTP routes (78 unique URL paths) registered.
Schema v78 (single logical version across sqlite + postgres; see src/storage/migrations.rs::CURRENT_SCHEMA_VERSION and src/store/postgres.rs::CURRENT_SCHEMA_VERSION).
Per-message Ed25519 federation signing (X-Memory-Sig header).
Per-peer attestation via x-peer-id header on every push.
28-field Memory model with reflection_depth, memory_kind, entity_id, persona_version, citations, source_uri, source_span, confidence_source, confidence_signals, confidence_decayed_at, version (Gap-1 optimistic concurrency, schema v45), lifecycle_state (schema v64), and cid (schema v74).

v0.7.0 substrate features the harness MUST NOT depend on

E2E encrypted_envelope field — deferred to v0.7.1-blocker (#228 in this repo).
memory_mark_outcome outcome-feedback tool — deferred to v0.7.1-blocker.
memory_skill_apply — deferred to v0.7.1-blocker.

What “closes” #718

Per the issue body: when the harness repo lands a workflow that exercises the locally-built v0.7.0 binary end-to-end (NOT a tarball download of v0.6.0), AND the boot-script env contract is documented, AND TLS provisioning is either fixed or explicitly opted out, then #718 closes. None of those changes live in this repo; the present file is the substrate-side handshake.

This file is the canonical cross-repo reference. The substrate-side ticket (#718) can be closed with a cross-ref to:

alphaonedev/ai-memory-a2a-v0.7.0 (harness repo)
this file
the v0.7.0 contract pin above

Provenance

Operator directive: 28860423-d12c-4959-bc8b-8fa9a94a33d9
Triage: .local-runs/issue-triage-2026-05-18.md
Substrate cure for #238 (parent of #717/#791 attestation gap): PR #716