Multi-Agent Codebase Refactoring Workflow

Trigger condition

Steps

1. 1

   Auditor agent: static analysis scan, generate dependency graph, identify hotspots

   tool: static_analysis · output: hotspot_report

2. 2

   Planner agent: break refactoring into atomic, non-overlapping subtasks

   tool: llm · output: task_breakdown

3. 3

   Spawn parallel worker agents (max 4): each agent owns one module, reads code, proposes changes

   tool: agent_framework · output: proposed_changes

4. 4

   Conflict detector: check proposed changes for inter-module dependencies or collisions

   tool: llm · output: conflict_report

5. 5

   Resolver agent: order changes to eliminate conflicts, produce sequenced patch list

   tool: llm · output: sequenced_patches

6. 6

   Apply patches sequentially, run test suite after each, rollback on failure

   tool: git_cli · output: refactored_codebase

Expected output

Tools required