Claude Agent: approving permission prompt always saves permissions to .claude/settings.local.json

[tredondo]

Reproduction steps

1. Start a Claude Agent
2. Run a prompt that will execute shell commands, e.g. "Investigate what processes are pinning CPU cores"

Current vs. Expected behavior

Currently, the permission approved by the prompt is persisted to .claude/settings.local.json, which is by default gitignored.

Claude.agent.permissions.always.persisted.to.settings.local.webm

Expected: ability to decide, per prompt, where the permission should be persisted, similar to the VS Code extension:

Zed version and system specs

Zed: v1.10.0+preview.315.8d56867088a4d4269bed518cdb1f913a037179de (Zed Preview)
OS: Linux Wayland ubuntu 26.04
Memory: 62.5 GiB
Architecture: x86_64
GPU: Intel(R) Iris(R) Xe Graphics (ADL GT2) || Intel open-source Mesa driver || Mesa 26.0.3-1ubuntu1

Attach Zed log file

No response

Relevant Zed settings

settings.json

   "agent_servers": {
      "claude-acp": {
         "default_config_options": {
            "effort": "low",
            "model": "sonnet",
            "mode": "plan"
         },
         "type": "registry"
      }
   },

Relevant Keymap

No response

(for AI issues) Model provider details

No response

If you are using WSL on Windows, what flavor of Linux are you using?

No response