@llm-optimize/cli

Command-line tools for analyzing, linting, and benchmarking LLM prompts. Find token waste, duplicate instructions, and quality issues before they reach your model.

Install

npm install -g @llm-optimize/cli
# or use without installing:
npx @llm-optimize/cli <command>

Commands

analyze — Inspect a single prompt file

npx llm-opt analyze prompt.md

Output:

{
  "file": "prompt.md",
  "promptQualityScore": 87,
  "tokenCount": 342,
  "suggestions": [
    "Compress excess whitespace",
    "Remove 2 duplicate line(s)"
  ]
}

• promptQualityScore — 0–100, higher is better. Penalizes duplicate lines and excess whitespace.
• tokenCount — estimated token count using the same estimator as @llm-optimize/core
• suggestions — actionable improvements

lint — Scan all prompt files in a folder

npx llm-opt lint prompts/

Recursively scans .md, .txt, and .prompt files.

Output:

[
  { "file": "prompts/system.md", "score": 95, "suggestions": [] },
  { "file": "prompts/user.md",   "score": 72, "suggestions": ["Compress excess whitespace"] }
]

benchmark — Measure token savings across a folder

npx llm-opt benchmark prompts/

Output:

{
  "files": 8,
  "beforeOptimization": { "tokenCount": 4820 },
  "afterOptimization":  { "tokenCount": 4210 },
  "savedTokens": 610,
  "compressionRatio": 0.873,
  "estimatedMonthlySavingsUsd": 0.0183
}

Shows real token savings from whitespace normalization across your actual prompt files.

Supported File Types

| Extension | Description | |---|---| | .md | Markdown prompt files | | .txt | Plain text prompts | | .prompt | Custom prompt files |

Use in CI

Add to your CI pipeline to catch prompt quality regressions:

- name: Lint prompts
  run: npx llm-opt lint prompts/ | jq '.[] | select(.score < 80)'