Caveman Ultra Mode: Make Your Coding Agent Say Less, Not Think Less
AI coding agents have a communication problem. They can be technically capable and still make a development loop feel slow because every small action arrives with a paragraph of ceremony: a restatement of the ticket, a promise to investigate, an explanation of an obvious command, and a summary that repeats the first three things. That style is sometimes useful. It is not useful all the time. When…
AI coding assistants often provide excessive detail in their communication, which can slow down the development process. When developers need to quickly diagnose issues, inspect diffs, or make specific fixes, a more concise approach is often preferred. Caveman is a plugin designed to address this communication gap by making coding agents use short, direct language while keeping the code, commands, and errors unchanged.
Caveman's Ultra Mode is an interface option that allows users to opt for a more compact response style. This mode focuses on the execution phase of work, ensuring that the agent thoroughly inspects the repository, follows the test suite, and detects any ambiguities. However, it refrains from adding unnecessary social filler to simple findings.
The project claims that using Caveman can lead to a 65% reduction in output tokens, but this figure should be treated as the project's own estimate rather than an independent benchmark. The actual savings depend on factors such as the agent, the task, the amount of tool output, and the prompt. Despite these variations, the basic principle remains effective: fewer tokens mean less attention required from the user.
Caveman is particularly useful in known-scope debugging, where the agent can quickly locate relevant middleware, compare conditions with test expectations, make the necessary changes, and report the results. In small implementation loops, concise updates keep the human in the loop, allowing for clear progress reports and easy file changes and test results visibility.
While concise output can help convey ideas more succinctly, it may also lead to a lack of necessary uncertainty. This mode is not recommended for parts of engineering that require nuance, such as discovery, requirements, architecture, security, privacy, incident response, and onboarding. In these instances, explicit language and thorough explanations are essential.
To maintain accountability, a compact reporting contract should be established. Before making edits, the agent should state likely causes in 1-3 bullet points. After edits, it should list changed files, validation results, outcomes, and remaining risk. Before undertaking irreversible or scope-expanding decisions, the agent should be asked for confirmation. This approach ensures that the agent remains brief while still providing the necessary evidence and context.
A practical operating model involves dividing agent work into decision mode and execution mode. In decision mode, the agent can explore options, explain comparisons, and surface uncertainties. Once an approach is approved, the agent should provide short plans and acceptance checks. Users can then approve or revise the approach and switch to Ultra Mode for implementation.
The final step is returning to normal mode for postmortems or architectural follow-ups. This method allows for a careful process combined with a compact execution interface, avoiding the false choice between verbose and reckless agents. Installation instructions for Caveman are provided in the project's documentation, supporting various operating systems and environments.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.
