FAQ

FAQ — Frequently Asked Questions

---

Q1: Does this skill work with GPT-4, Gemini, or other LLMs?

Yes — mostly. The core techniques are universal and work with any large language model. System prompt templates are written for Claude but require only minor wording adjustments for other models. Token counts may differ slightly between models.

---

Q2: Do I need to upload all files or just SKILL.md?

Just SKILL.md is enough to get all core techniques. The files in references/ and examples/ are optional depth. For Claude Projects, uploading all reference files gives the best results.

---

Q3: What does "star-required license" mean exactly?

You must star this repository on GitHub before using the skill in any project. Starring is free and takes one click. It is not a payment — it is attribution. See LICENSE for full legal terms.

---

Q4: How do I measure token counts?

Use one of these free tools:

• Anthropic token counter — built into the Claude API (count_tokens endpoint)
• OpenAI Tokenizer — platform.openai.com/tokenizer — good for rough estimates
• tiktoken (Python) — pip install tiktoken — accurate for most models
• Claude.ai — shows token usage in the interface on Pro plans

All benchmarks in this repo use the cl100k_base tokenizer.

---

Q5: Will these techniques reduce output quality?

No — in most cases quality improves. Compressed prompts are more precise, which reduces Claude's need to interpret vague instructions. In 20 benchmarked examples, quality improved in 14 cases and stayed the same in 6. It never dropped.

---

Q6: How do I use SKILL.md in the Claude API?

import anthropic

with open("SKILL.md", "r") as f:
    system_prompt = f.read()

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-4-20250514",
    max_tokens=1024,
    system=system_prompt,
    messages=[{"role": "user", "content": "optimize: [your prompt here]"}]
)
print(response.content[0].text)

---

Q7: Can I combine this with other skills?

Yes. In Claude Projects, upload multiple skill files. Token Optimizer will influence how Claude processes all other skill instructions — making your entire project more efficient.

---

Q8: What is the difference between a skill and a system prompt?

In practice they are the same thing. A skill is a structured Markdown file designed to be used as a system prompt. The YAML front matter makes skills self-describing and discoverable. Claude treats it as a regular system prompt when loaded.

---

Q9: Do XML tags actually help Claude parse prompts?

Claude does not parse XML like a parser. The tags are semantic markers that reduce ambiguity. Tags like <task>, <context>, and <constraints> are a consistent formatting convention that Claude interprets reliably. They work because they reduce ambiguity, not because of XML parsing logic.

---

Q10: How often is this skill updated?

Updates happen when new compression patterns are validated, Claude model updates change how techniques perform, or community contributions add high-quality content.

Check CHANGELOG.md for the latest version.

Subscribe to the WhatsApp Channel for update notifications.