Developer releases Claude Code plugin to shorten verbose replies

By Splitfeed EditorialLast updated Oct 8, 2026, 10:30 UTC

On October 7, 2026, a developer posted a tool called Terse to Hacker News, describing it as a Claude Code plugin built to cut the length of Claude's replies by approximately half. The post focused on specific stylistic complaints: recurring metaphors, recap passages, slogans, unusual noun choices, and constructions like "it's not X, it's Y". The developer reported that Claude's built-in concise mode did reduce overall length but left behind what they called "slop", meaning the unwanted stylistic patterns persisted. System prompts written to target verbosity fared worse, being forgotten by the model after only a few conversation turns. Terse is presented as a more consistent enforcement mechanism than anything previously available in the Claude Code plugin ecosystem.

The story matters because it points to a practical friction between how large language models are tuned to communicate and what at least some professional users actually want during focused technical work. Claude's default output style is shaped by training choices that favor completeness, warmth, and explanation, qualities that can feel useful in some contexts and exhausting in others. The fact that the developer found neither the native concise setting nor manually written system prompts to be reliable fixes suggests the issue sits deeper than a simple toggle. Whether Terse actually solves this consistently, or whether it introduces its own tradeoffs by stripping context that turns out to be useful, is the real open question.

None of the three camps have published reactions to this story yet. The Pro-AI camp would typically treat a story like this as evidence that the ecosystem around AI tools is maturing, with developers filling gaps that model makers have not prioritized. The Anti-AI camp would likely use it to argue that the verbosity and formulaic phrasing are symptoms of models optimized for appearing helpful rather than being useful, a form of performance baked into the training process. The Middle Ground camp would probably acknowledge the legitimate frustration while noting that a community plugin addressing the problem is exactly how open tooling is supposed to work, and that the burden falling on individual developers rather than Anthropic itself is worth scrutinizing.

The most concrete thing to watch is whether Terse maintains its claimed fifty-percent reduction across varied tasks and longer sessions, and whether Anthropic responds by adjusting Claude Code's native behavior or expanding its style controls. User feedback accumulating on the Hacker News thread over the days following the October 7 post will give the earliest signal of whether the plugin holds up in practice.