This README summarizes the architectural changes between the S9 project ('C:\Users\Monisha Srivastava\Downloads\S9') and the S10Share project ('C:\Users\Monisha Srivastava\Downloads\S10Share').
-
Entry point and bootstrapping
S9 runs from 'agent.py' and wires 'core.loop.AgentLoop' with 'core.session.MultiMCP'. S10Share moves to 'main.py', loads MCP server configs from 'config/mcp_server_config.yaml', and instantiates 'agent.agent_loop2.AgentLoop' with explicit perception/decision prompt paths. -
Module organization
S9 groups execution logic under 'core/' and 'modules/' (context/session/ strategy separated from perception/decision/action/tools). S10Share promotes these into top-level packages ('perception/', 'decision/', 'action/', 'memory/', 'agent/'), reducing the “core vs modules” split and making the pipeline stages first-class. -
MCP server layout
S9 keeps MCP server scripts in the root ('mcp_server_*.py') and constructs them from 'config/profiles.yaml'. S10Share consolidates MCP server logic into 'mcp_servers/' with 'multiMCP.py' and uses 'config/mcp_server_config.yaml' for multi-server setup. -
Session + memory flow
S9 uses a 'core.context.AgentContext' and a loop with lifelines, then runs perception/decision/action via 'modules.*'. S10Share shifts to 'AgentSession' + 'PerceptionSnapshot' tracking, logs sessions via 'memory.session_log', and performs a memory search before perception as a structured step. -
Agent loop refactor
S10Share’s 'agent/agent_loop2.py' is a cleaner orchestration layer: it extracts helper methods (search, perception, decision, step eval), making the loop more modular and easier to extend than the S9 monolithic 'core/loop.py'.
See 'PROVENANCE.md' for source notes, human/LLM contributions, and change history context.