Built CONCISE cuts prose filler from AI coding agent responses.
Trailing summaries, restated context, hedging, over-long preambles: gone. Code, commit messages, analysis depth: untouched.
Before:
"Sure! I'd be happy to help. Let me take a look... Great, I found the issue! [explains at length]... To summarize what I did..."
After:
"Line 42: assignment, not comparison. Fixed."
Two ways to run it:
→ Claude Code plugin full ruleset + a live per-response word-budget bar right in the statusline
claude plugin marketplace add 0xyph3r/CONCISE && claude plugin install concise@concise
→ Agent Skill, any compatible tool (Cursor, Copilot, Codex, Gemini CLI, more) same compression rules, portable
npx skills add 0xyph3r/CONCISE
Open source, MIT.
https://t.co/guHaQEPkMR
CONCISE v0.2.0: now covers subagents too, not just the main session.
Dispatch a Task-tool subagent and it gets the compression ruleset injected into its own prompt before it runs verified by reading its raw transcript, not just trusting its word.
Their responses get measured on the live statusline bar too.
npm install -g @0xyph3r/concise
Gemini 3.8 Flash Coming Tomorrow
- gemini-3.8-flash has been deployed and coming to release tomorrow.
- Coming with fixes many of the issues Gemini 3.7 Flash struggled with, with a lot of the slop cut down.