Agent Bridges

agent-bridges

Config and templates for a fixed-role delegation pipeline that pairs **Claude Code** with three MCP bridges:

• Antigravity (agy-bridge MCP server, pinned to the Antigravity pin) plays three roles across the pipeline: Document Analyzer, Coder/Executor, Release/Changelog Writer.
• Codex QA (codex-qa, QA pin) — QA Engineer.
• Codex Security (codex-security, Security pin) — Security Engineer.
• Claude Code is Planner and Code Reviewer, on whatever model you pick per plan. It's the only agent with tool access, repo/session state, and write permission — it plans, reviews first, and has final say on every bridge output.

Why

Routing everything through one model is wasteful when another provider fits a role better. Instead of routing task-by-task, each bridge gets a fixed job: Antigravity ingests specs, writes the code, then drafts release notes; Codex QA always tests the diff; Codex Security always reviews it for exploits; Claude Code always plans, reviews before either Codex role sees a diff, and has final say — including over the docs Antigravity drafts.

Claude Code owns verification: a delegated finding is never the sole basis for an edit until checked (see "Parallelism, failures, and verification" in `templates/CLAUDE.md`). Reading a large file, diffing a big repo, or running QA/security burns context fast — often only the *answer* is needed, not the investigation — so the bridges return results without Claude Code carrying that context itself.

Architecture

• One direction only. Claude Code calls all three bridges; none of them call back into Claude Code or each other. Execution authority stays at one point.
• Claude Code = MCP host. Antigravity (antigravity server, running agy-bridge) and two separate Codex instances — codex-qa and codex-security — are three independent MCP servers over JSON-RPC 2.0. Both Codex instances run the same binary but under different