Change control workflow¶

Primary MCP edit cycle (sole sequence diagram for change control in the guide):

sequenceDiagram
    participant Agent
    participant MCP as CodeClone MCP
    Agent ->> MCP: analyze_repository(root=<abs>)
    MCP -->> Agent: run_id
    Agent ->> MCP: start_controlled_change(root=<abs>, scope, intent, dirty_scope_policy?)
    MCP -->> Agent: intent_id, blast_radius, budget, edit_allowed
    Agent ->> MCP: get_relevant_memory(root, scope|intent_id)
    MCP -->> Agent: ranked memory context
    Agent ->> MCP: get_implementation_context(root, paths, intent_id?)
    MCP -->> Agent: bounded structural context
    Note over Agent: edit files
    opt Python structural / governance config
        Agent ->> MCP: analyze_repository
        MCP -->> Agent: after_run_id
    end
    Agent ->> MCP: finish_controlled_change(intent_id, changed_files|diff_ref, after_run_id?, claims_text?)
    MCP -->> Agent: status, summary, workspace_hygiene_after, intent_cleared

Tool tiers¶

Normative tool params: MCP workflow tools. Finish pipeline and hygiene: finish_controlled_change, Finish hygiene.

Related recipes¶

• Agent edit cycle
• Queue & recovery
• Atomic debug path
• Engineering Memory recipes