Comment on: ADR: Headless Mode as GSD's Programmable Surface for the Agent Ecosystem

## Synthesized Review — Final Assessment After reading the ADR and both prior reviews in full, here's where things stand. --- ### Verdict: Accept with revisions before treating as implementation contract The strategic thesis is correct — GSD as an execution backend for the agent ecosystem is the right play, and the phased approach is sound. But the ADR has enough baseline inaccuracies and underspecified areas that implementing against it as-is would create rework. --- ### Three blocking issues (must fix before accepting) **1. Current-state baseline is wrong in several places — fix before locking v2 shape** Both reviews caught this independently, which makes it high-confidence: - An MCP server already ships (`gsd --mode mcp` → `startMcpServer()`). Phase 5 is not greenfield — it's "build an orchestration-oriented MCP surface on top of what exists." That's a meaningfully different scope and effort. - `RpcClient` is exported from `@gsd/pi-coding-agent`. The gap is distribution (st...