I benchmarked Claude Code's caveman plugin against "be brief."

Caveman is a popular Claude Code compression plugin. The pitch is in the name: ultra-compressed responses, ~75% fewer tokens, all the technical accuracy. Six modes, slash commands, intensity dials, classical Chinese variants.

I benchmarked it against two words: "be brief."

Same quality. Same range of tokens. The plugin didn't beat the boring default on either axis.

What I tested

I ran 24 prompts across six categories: bug diagnosis, concept explanations, architecture tradeoffs, multi-step setup, security and destructive ops, and error interpretation. Each prompt had a per-prompt rubric including key points, required terms, and claims to avoid.

Five arms were tested:

1. Baseline: Claude default.
2. Brief: "Be brief." prepended to every prompt.
3. Lite, Full, Ultra: Caveman plugin at three intensity levels.

Each arm ran through claude-opus-4-7, and a separate claude-sonnet-4-6 scored every response against the rubric.

Quality didn't move

Compression didn't hurt correctness. Every arm scored within 1.5% of each other. Baseline and Brief both hit 0.985, while Caveman modes ranged from 0.970 to 0.976. Every arm hit 100% of its key points.

The headline result

| Arm | mean tokens | |---|---| | baseline | 636 | | brief | 419 | | lite | 401 | | full | 404 | | ultra | 449 |

"Be brief." cut tokens by 34% versus baseline. Caveman lite and full landed close to brief. Ultra, the strictest mode, produced the longest answers due to its "Auto-Clarity" rule, which drops compression for safety warnings and multi-step sequences.

So what's caveman actually for?

If a two-word prompt matches it on tokens and quality, the value isn't just compression—it's structure:

• Consistent output shape: Predictable patterns that "be brief" lacks.
• Intensity dial: Slash commands to switch modes mid-session.
• Persistence: Re-injects rules via hooks to prevent drift in long sessions.
• Safety escape: Explicitly knows when to stop compressing for destructive operations.

Conclusion

If all you want is shorter outputs, start with "be brief." in your prompt or CLAUDE.md. Reach for Caveman when you need consistent output structure across sessions. The bigger lesson: most prompt-engineering advice hasn't been measured against the boring default. Measure it.