Drop-in styles that teach Claude Code to get to the point

What it does

Attention Span is a set of three markdown output styles for Claude Code—Attention-kind, Spartan, and Rundown—that reformat the model’s natural-language responses without touching the code or reasoning underneath. They enforce answer-first structure, heavy bolding, arrow markers, and plain English so the responses are built for skimming. The README frames each style as a single markdown file, a kind of drop-in prose skin for the agent’s chat persona.

The interesting bit

The project actually benchmarks its styles against hidden test suites to prove the engineering work stays intact—97% pass rate with or without the style—while cutting output length by a median of 41%. It also measures that the answer lands in the first line 75% of the time, up from 3% with default styling, and it publishes the reproducible harness rather than asking you to trust a vibe check.

Key highlights

• Three personalities: Attention-kind (ADHD-friendly, arrow-marked, warm), Spartan (blunt imperative, zero cushioning), and Rundown (TL;DR briefings with emoji checklists and numbered next-steps).
• Reproducible benchmark harness showing ~43% shorter output on average and 50–71% reduction on verbose answers; deliverables come out clean 88% of the time versus 12% with default styling.
• Answer-first formatting puts the conclusion in line one 75% of the time, up from 3% default, using no LLM judge for the headline figures.
• Plain-English rules propagate into code comments, explaining the “why” without importing chat formatting.
• Licensed under AGPL-3.0.

Caveats

• Built specifically for Claude Code; while the repository description mentions Codex and other agents, the README, benchmarks, and integration details all center on Claude.
• Token savings are explicitly a side effect, not the primary goal—the README redirects cost-cutters to its sister Token Optimizer project.
• Only three styles currently ship, so if none match your preferred tone, you are left to write your own.

Verdict

Grab this if you use Claude Code daily and find yourself drowning in polite, winding prose when you just wanted a yes-or-no. Skip it if you already savor every word your agent writes, or if you are looking for a magic wand to shrink API bills.