Skip to content

Harness/Subagent for terminal commands auto-review #1486

Description

@Mastermind-U

I moved from Codex Agent to your platform, I really like Harness/Orchestration mode, but it lacks of terminal commands review unlike codex.
My motivation:
In current state I need to check commands security myself before execute. I can set auto-approve, but it's pure machinery that cannot separate rm -rf ./temp (temporary dir for current pipeline) and rm -rf /, so I need to check it myself. This is simplified example but in real cases I meet really weird commands that are hard to read so at the end of the day I just press approve button without a glimpse what running agent is launching.
In Codex there is an option to switch to auto-review to approve/reject all executable commands.

To change this behavior, pipeline needs to create subagent with zero context about pipeline that can quickly check command contents and call me if command is potentially harmful or automatically accept it.
This will lead to reduced interaction with chat.

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or request

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions