I'd like to propose adding support for defining custom tools with

.openhands/tools/

e.g. I could have something like

name: run_linter
description: Execute the project's linter using npm run lint
inputSchema:
  type: object
  properties:
    fix:
      type: boolean
      description: Whether to attempt automatic fixes
      default: true
  required: []
run: |
  if [ "$fix" = "true" ]; then
    npm run lint:fix
  else
    npm run lint
  fi

Is this part of the big refactor or are you also talking about how openhands works now?

I think it can be independent of the refactor

The definition is a MCP-like definition. Then we execute something when called. What would be the advantage of supporting this, when we could say, use MCP?

Mainly that it's a single file that can be checked into a repository, rather than the user needing to manage running a server

if this is gonna get implemented, I'd really like to be able to be able to scope them ... when it could be activated ... where ... perhaps with overloadable attributes and deep merging .... as well as being able to dynamically alter them... durably or ephemerally ... they should be able to, and not required to, live in the repo in question .... or the directory in question ... or be associatable with a workdir but not live in the repo ... (org-scoped with project subdirs? or perhaps a list of project/dir enable/disable lists) not that I have opinions or anything (( doubly so: not that all/most of my opinions are necessarily meritous )) also I want a pony. 😛