hermes-mcp/upgrade_0.6.0.md

Hermes-mcp upgrade to 0.6.0 :

hermes-mcp - trader-mcp communication:

desired behaviour:

• hermes-mcp only acivates and deactivates strategies on trader-mcp. no observe mode. it only switches to another stratgey. therefode the mode parameter can be dropped for now from decision logic and dashboard and deision history, if hermes wants to suspend a strategy, it deactivates it.
  
• investigate the back reporting from trader and minimize load or unnecessary communication.

• hermes does not need to care about passing parameters to strategies on trader or mess with the configuration, except for maybe reading it.

• Hermes supervises the trader by activating the new and deactivating old strategies if appropriate and receives minimal reports like wallet composition and balances from the strategy on trader.

hermes-mcp decision logic

background: we recently introduced playbooks. the current decision logic is mostly based on the grid-trend-rebalancer playbook. the trend-only playbook is here to have another playbbok while bootstrapping the app.

• the repo contains a directoy named decision_families, this is to be renamed to playbooks.
• and the code needs to be refactored so that the actual decision logic used by a playbook does really reside in the corresponding python code in the playbooks folder.

playbooks

• i am ok with having json files in the repo for definition of each playbook eg grid-trend-rebalancer.json and trend-only.json

• a playbook defines

  • a set of strategies and
    
  • parameters as well as
    
  • the logic
    that allow hermes to make decisions to activate or deactivate strategies if decided appropriate on trader.

• a playbook defines roles and which strategy family or class, can be used to fill each role. eg for our grid-trend-rebalancer it would require those four roles : grid, buyer, seller, rebalancer we already have a playbooks endpoint in the dashboard that lets us assign strategies to roles.

• a playbook defines a set of parameters, that are used to fine tune the decision logic. these parameters are tunable via the dashboard

• each playbook has a corresponding file in the playbooks folder, where most if not all of the logic used by that playbook is defined.

• the main decision_engine.py should only contain logic that can be used by all playbooks, eg general input data, utility functions ...

• playbook specific functions should be moved to the playbooks specific python file.

• i dont know if we need a dispatcher.py !!?? imho, the main decision_engine.py should play the dispatcher for the playbooks.

• the dashboard and decision logic can use the playbook definitions to choose the right parameter set for processing, persisting and display.

expectations and constraints

• define more clearly how playbooks are used throughout hermes-mcp,
• clearly pack the decision logic and parameters of a playbook into its defilition and code file
• make the use of playbooks and their different parameters an decision logic code tock solid as a abse for further development
• clean out leftovers from old architectural mistakes or obsolete code, simplify where possible

• we need to use the term playbook in the code whenever appropriate and drop the other misleading terms like decision_family and also maybe decision_configuration for the more appropriate idea of a playbook and its configuration parameters

• we do not need to validate if strategies fit the expected family to be used to fill a role. we do not need to be able to edit playbooks in the dashboard. the current implementation is sufficient.

• we want to bring hermes-mcp to a point where fine tuning the deciding weights for transitions between strategies as defined by a playbook is solid and reliable.