CtrlK
BlogDocsLog inGet started
Tessl Logo

juliusbrussee/caveman

Compressed caveman-style prose for AI coding agents — cuts ~65% output tokens while keeping full technical accuracy

96

1.00x
Quality

100%

Does it follow best practices?

Impact

96%

1.00x

Average score across 38 eval scenarios

SecuritybySnyk

Passed

No known issues

Overview
Quality
Evals
Security
Files

criteria.jsonevals/scenario-36/

{
  "context": "Tests whether the agent asks clarifying questions or provides a structured diagnostic approach when given a vague problem description, rather than guessing at a solution. Compressed output mode should not skip the diagnostic step.",
  "type": "weighted_checklist",
  "checklist": [
    {
      "name": "Asks for more information or provides diagnostic steps",
      "description": "Asks what kind of app, what 'slow' means (latency, throughput, startup), or provides a systematic diagnostic checklist rather than jumping to a specific fix",
      "max_score": 15
    },
    {
      "name": "Does not prescribe a single fix without context",
      "description": "Does not jump to a specific solution (e.g., 'add an index' or 'use caching') without understanding the problem first",
      "max_score": 15
    },
    {
      "name": "Covers multiple possible causes",
      "description": "Mentions at least 2-3 common performance bottleneck categories (database, network, CPU, memory, I/O)",
      "max_score": 10
    }
  ]
}

evals

README.md

SKILL.md

tile.json