ADR-0002: protoWorkstacean ↔ protoMaker integration boundary

Status: Superseded (2026-07-03). protoMaker was decommissioned from the fleet; the org→execution board integration described here no longer exists. Retained as historical record.

• Status: Accepted — 2026-05-27
• Deciders: Josh (operator)
• Related: ADR-0001, protoMaker#3975

Context

protoMaker advertises an A2A agent card, so protoWorkstacean's SkillBroker auto-discovered its skills and registered A2A executors for them. But protoMaker serves no /a2a JSON-RPC endpoint — /a2a returns 404, and the card's advertised URL (ava:3008/a2a) resolves to the protoWorkstacean container itself. So every workstacean→protoMaker A2A dispatch failed silently:

• the board.health ceremony failed every 30 min (since retired),
• the Linear→protoMaker manage_feature bridge never delivered,
• a dead bug_triage A2A executor shadowed Quinn's working in-process one via health-weighted resolution.

We need a durable contract for how the two systems talk, before building the ADR-0001 intakes on top of it.

Decision

protoWorkstacean ↔ protoMaker communicate over HTTP + GitHub webhooks + bus POST /publish. Never A2A. There is no skill-dispatch chain from protoWorkstacean into protoMaker's agents.

• protoWorkstacean → protoMaker:
  • Create-project intake — Ava calls a protoMaker HTTP endpoint to start the create-project → decompose flow (ADR-0001, initiative-level).
  • GitHub issues — protoMaker ingests issues via GitHub's native webhooks (ADR-0001, issue-level; protoMaker#3975).
• protoMaker → protoWorkstacean:
  • Lifecycle events via POST /publish (feature.completed / feature.failed), consumed by feature-notifier + the Linear bridge.
• Reads: protoWorkstacean polls protoMaker GET /api/settings/global for the project registry (the source of truth for projects).
• protoMaker owns its own execution — Roxy and proto run inside protoMaker. protoWorkstacean does not dispatch into them.

Retirements that follow from this:

• Remove the dead protomaker A2A entry from workspace/agents.yaml (stops the dead-executor pollution + the bug_triage shadow).
• Retire the linear-protomaker-bridge manage_feature A2A dispatch — superseded by issue ingestion + the create-project intake. Keep the reverse feature.completed / feature.failed consumer.

Consequences

• No more dead A2A pollution; contracts are explicit and each side is self-contained (protoMaker resolves its own repo→project routing from its own .automaker/settings.json, not from a protoWorkstacean file).
• This decides the open question on the bridge: retire it, do not rebuild manage_feature over HTTP — the two ADR-0001 intakes replace it.
• Aligns with the broader "no dispatch chain to protoMaker" direction. If multi-node messaging is ever needed, that's a bus-bridge concern, not a reason to revive A2A here.