Skip to main content
Recently, well-designed Agent Computer Interfaces (ACIs) like Cursor and Claude Code have emerged to help agents understand the state of the software development process. These interfaces manage history to avoid unnecessary context from prior observations and provide actions that LLMs can use efficiently and reliably. The ACI specifies both the commands available to LLMs and how the environment state is communicated back to the language model. It also tracks the history of all previous commands and observations. At each step, it manages how these should be formatted and combined with high-level instructions into a single input for the LM. We propose ByteRover CLI as a Hybrid Computer Interface (HCI) where humans and agents collaborate on the same interface to accomplish tasks. This approach shapes context and environment feedback to complement the agent’s limitations and abilities. Drawing inspiration from ACIs and Context Engineering, where studies reveal how well different interfaces align with human intuition and performance, we leverage a command line interface that supports both human users and agents to curate and build optimal context for software engineering tasks.