Turn advice off

All advice in one project#

Run the disable skill inside the project:

No principle shows advice in that project from the next prompt on, and other projects keep theirs. The project is the git repository you work in, and a git worktree counts as the repository it belongs to.

The switch is stored on your machine, in ~/.wellactually/disabled.json, and not in the project. So it leaves no file in a repository that is not yours, and there is nothing for your agent to commit. In Codex, the skill asks you once for permission to write there, because the sandbox lets it write only inside the project.

One principle#

Three switches decide whether a principle runs. It runs only when none of them is off.

On your advisory board#

Open your advisory board, open the expert or collection the principle comes from, and switch it off. It stops on every machine after the next sync.

For one project#

Ask your agent, inside the project, to switch the principle off: "stop showing me passsy/avoid-late in this repo". It uses the MCP tool disable_principle for the project. The disable skill with the principle's id does the same: /wellactually:disable passsy/avoid-late.

The project's root is the nearest folder upwards that already has a .wellactually.json, or else the nearest one with a .git. The file is plain JSON, and you can commit it so the whole team agrees:

{
  "disabled": [
    "passsy/avoid-late"
  ]
}

Everywhere on this machine#

Ask your agent to switch the principle off everywhere. It uses disable_principle for the machine.

Switch it on again#

Ask your agent to switch it on again. It uses enable_principle, in the scope it was switched off in. The switch on your advisory board is only changed on the website.

The project and machine switches never leave your machine.

When the agent thinks advice does not fit#

An agent that thinks advice does not fit should not switch it off on its own. It should say so, and rate it not_applicable, which tells the expert their detector is too eager.