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

Independent audit against current `main` plus the cited external surfaces. I think the direction is good, but I would not merge this ADR as written yet. There are a few baseline mismatches in the “current state” section, and one protocol design gap that I think needs to be resolved before the v2 shape is locked in. 1. High: “No MCP server” is not accurate in the current tree. The repo already ships `gsd --mode mcp` from the main CLI, and it already exposes a stdio MCP server over the current tool registry. Relevant code paths: `src/cli.ts` wires `mode === 'mcp'` into `startMcpServer()`. `src/mcp-server.ts` implements `tools/list` and `tools/call` over stdio. `src/tests/mcp-server.test.ts` covers the module shape. The missing thing is not “MCP exists vs does not exist”; it is “there is no orchestration-oriented MCP surface for long-running GSD workflows”. That distinction matters. As written, Phase 5 reads like greenfield work, when the real decision is whether to evolve the existing...