It should be possible to configure a set of commands the agent can run without permission

### Describe the feature or problem you'd like to solve

_No response_

### Proposed solution

Currently you can only set allow-all to allow the agent to execute commands without permission. Otherwise with every new session i have to approve every command over and over again.

The --allow-all flag bypasses all permission checks, which create real risks:
- Destructive commands like" rm-rf" could be e executed without confirmation
- Permission changes (chmod, chown) could silently alteer system security

A granular allowlist ( eg allow read, find, cat,ls but block rm, chmod, would provide a much better balance between autonomy and safety.

This should be configurable via instruction or agent files

### Example prompts or workflows

_No response_

### Additional context

_No response_

Start by reviewing the existing --allow-all permission path and how instruction or agent files are configured. Define how a command allowlist coexists with confirmation and blocked commands, then verify that configured safe commands run without repeated approval while destructive commands still require permission.

Written by the indexing model from the issue text.