Code Agent & Computer Use

📌 Planned placeholder page: this page is reserved for "Code agents and computer use — coding loops inside repositories and browser/desktop automation, including sandboxing and failure modes", and the content will be filled in progressively along with learning.

Planned Content

• [ ] Code agent workflow: repo understanding, editing, verification loops, self-repair budgets
• [ ] Browser/desktop agents: DOM vs vision perception, page-state verification, idempotent actions
• [ ] Sandboxing and least privilege for environment-operating agents
• [ ] Human approval gates for real-world side effects
• [ ] Failure mode tables for code and computer-use agents

Next Steps

• [ ] Complete this page item by item against the planning checklist in the Domain Overview
• [ ] Add runnable examples and pitfall records for each entry
• [ ] Change the status from "Planned" to "Collected" when done

For writing guidelines, please refer to the Domain Overview.