Multi-Agent Coding Orchestration System

You are a developer productivity engineer. Build a multi-agent orchestration system where specialized coding agents collaborate on repository tasks. Tech Stack: [INSERT - Python + LangGraph or TypeScript with a custom runtime] Agents: [INSERT - Planner, Coder, Reviewer, Test-Runner, Researcher] LLM: [INSERT - GPT-4o / Claude 3.5 / local code model] Workspace: [INSERT - Git worktrees, Docker job runners, S3 artifact store] Deliver the following modules in order: 1. Architecture: supervisor DAG runtime dispatching role agents; shared blackboard holding task plan, patches and review comments; max-step and budget guards. 2. Data Model: run/task/artifact entities, agent message envelope with trace_id, immutable patch records, plan revisions with lineage. 3. Repository & Tool APIs: Git branch/commit/PR operations, scoped workspace file read/write, shell and build commands, structured tool results fed back into the loop. 4. Coordination: task decomposition with dependencies, replanning on failure, file locks/ownership to prevent edit conflicts, critic-gated merge. 5. Human Gates: plan approval, confirmation for risky changes, pause/resume, comment injection mid-run. 6. Testing: mocked-LLM integration tests for the full DAG, SWE-bench-style task benchmark, determinism/replay tests, conflict and timeout scenarios.