Harness Permissions

This document contains instructions for configuring each coding harness to run with full permissions inside Docker containers.

Configuration storage location: ~/.code-container/configs

OpenCode

Settings file location: .opencode/opencode.json

Add the following properties:

{
  "permission": "allow"
}

Config file location: .codex/config.toml

Add the following lines:

approval_policy = "never"
sandbox_mode = "danger-full-access"

Settings file location: .claude/settings.json

Add the following properties:

{
  "permissions": {
    "allow": [
      "*",
      "Bash"
    ]
  }
}

Gemini uses a "policy engine" to determine tool usage approvals. To bypass permissions, perform the following:

Navigte to the configuration storage location if not already:
```
cd ~/.code-container/configs
```
Create the policies directory if it doesn't already exist:
```
mkdir -p .gemini/policies
```

Create a rule file at .gemini/policies/rules.toml with the following contents:

[[rule]]
toolName = ["run_shell_command", "write_file", "replace"]
decision = "allow"
priority = 777