Runtime Auto-Approve

Runtime Auto-Approve is a built-in plugin that can automatically approve or deny some ACP agent permissions and supervised terminal permissions based on rules you configure.

It is enabled by default in current builds, but it still changes the safety boundary around agent actions and should be reviewed intentionally.

When to use it

Use this plugin when you want to inspect or tune how RealTimeX auto-approves or denies well-understood agent workflows.

Good examples:

• repeated safe read-only inspection work
• tightly controlled terminal workflows
• supervised environments where you want policy consistency

Not a good example:

• broad experimentation where you are still learning what the agent might ask to do

Where to configure it

1. Open Settings > Plugins.
2. Find Runtime Auto-Approve.
3. Click Configure.

If you previously disabled it, re-enable it first.

Main configuration areas

ACP safe tool kinds

This list controls which ACP tool kinds can be auto-approved.

Examples include:

• read
• edit
• delete
• move
• search
• fetch
• switch_mode
• other

execute is intentionally more sensitive than most other tool kinds.

Terminal approval preset

The preset controls the overall approval posture.

• Ask Every Time: no automatic trust
• Recommended: the safer guided default
• Custom: you explicitly choose allowed actions and command groups

Auto-allowed terminal actions

In Custom mode, you can allow specific supervised terminal actions, such as:

• reading terminal output
• web search
• web fetch
• sending signals
• writing raw terminal input
• approving structured commands

Raw writes and command approvals are the most sensitive options in this group.

Auto-allowed command groups

Also in Custom mode, you can approve recognized command groups such as:

• read-only shell commands
• git inspection commands
• test commands

Extra allowed command prefixes

Use this when you want to allow additional structured command prefixes that are not covered by the built-in command groups.

Blocked command prefixes

These are denied before allow rules are considered.

Treat this list as an explicit guardrail layer, not just an optional convenience setting.

Log decisions

Keeps an audit trail of approval and denial behavior in system logging.

This is recommended when you are tuning a policy.

LLM command analysis

This optional mode allows an LLM to review commands that pass static pattern checks.

Use this only if:

• you understand the extra complexity it adds
• you have configured the related LLM slot
• you still keep strict static rules in place

Safe rollout advice

1. Start with Ask Every Time or Recommended.
2. Observe what the agent actually requests.
3. Move to Custom only when you understand the request patterns.
4. Keep blocked prefixes conservative.
5. Leave decision logging on while tuning.

Related guides

• For general plugin lifecycle actions, see Install & Manage Plugins
• For configuration behavior, see Configure Plugin Settings