agentleFS
sennaBrunosennaBruno/claude-skills/skills/ai-tone-calibration/SKILL.md

ai-tone-calibration

Skill1 starsChanged 8 months agoApache-2.0View on GitHub

Use when designing or adjusting AI assistant personality, chatbot tone of voice, or LLM system prompt persona. Triggers: tone too formal, too casual, robotic responses, forced enthusiasm, emoji overuse, personality mismatch, user complaints about AI voice

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 3 are distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈3

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    When the AI must respond in multiple languages:

    line 157
  2. Distinctive:

    Always include: "Respond in the user's language (detect from message)"

    line 159
  3. Distinctive:

    Test with non-English queries specifically

    line 160

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AI Tone Calibration

Overview

Calibrate AI assistant tone using a 3-axis scoring system instead of binary "formal vs casual" thinking. Maps personality to measurable dimensions, then translates scores into concrete system prompt instructions.

Core principle: Tone is not a single slider. It's three independent axes that combine to create a distinct voice.

When to Use

  • Designing a new AI assistant persona
  • Users complain tone is "too robotic" or "too bubbly"
  • AI responses feel inconsistent or "off-brand"
  • Switching LLM models and tone shifts unexpectedly
  • Building RAG/chat systems that need domain-appropriate voice

Not for: Marketing copy tone (use copywriting), general prompt engineering (use prompt-engineer)

The 3-Axis Framework

Score each axis 1-5 independently:

Axis135
WarmthClinical, detachedApproachable, clearNurturing, effusive
FormalitySlang, fragmentsContractions, directCeremonious, elaborate
HumorDeadpan seriousDry wit, confidentPlayful, jokes, emojis

Common Profiles

ProfileWFHBest for
Knowledgeable Friend3.522Utility apps, knowledge bases, RAG assistants
Professional Advisor241Finance, legal, healthcare
Enthusiastic Helper524Onboarding, tutorials, kids' apps
Neutral Expert231Documentation, technical support
Warm Professional432Customer support, SaaS products

System Prompt Pattern

Translate axis scores into a <tone> block with concrete behavioral instructions:

You are {name}, a {role} for "{context}".

<tone>
- {warmth instruction}
- {formality instruction}
- {humor instruction}
- {explicit prohibitions based on low-scoring axes}
</tone>

<rules>
- {domain-specific behavior rules}
</rules>

Translating Scores to Instructions

Warmth 1-2:

- Be direct and factual. Skip pleasantries.

Warmth 3-4:

- Sound like a knowledgeable friend. Be helpful without being effusive.
- Use casual, clear language. Contractions are fine.

Warmth 5:

- Be warm and encouraging. Celebrate user progress.

Formality 1-2:

- Use casual, direct language. Contractions, short sentences.
- Do NOT use formal phrases like "I'd be happy to assist."

Formality 3-4:

- Use clear, professional language. Contractions acceptable.

Formality 5:

- Use formal, precise language. No contractions or colloquialisms.

Humor 1:

- Do NOT use emojis, exclamation marks, or playful language.
- Personality comes from clarity and confidence, not from being bubbly.

Humor 2-3:

- Light personality through confidence, not through jokes or emojis.
- Do NOT use emojis or cheerful sign-offs.

Humor 4-5:

- Be playful where natural. Light humor is welcome.
- Use emojis sparingly to add warmth.

Explicit Prohibitions (Critical)

Low-scoring axes MUST have explicit "Do NOT" instructions. LLMs default to helpful-enthusiastic without guardrails.

# Humor = 1-2: MUST include
- Do NOT use emojis, exclamation marks at end of answers,
  or cheerful sign-offs.

# Warmth = 1-2: MUST include
- Do NOT add filler phrases ("Great question!", "Happy to help!").

# Formality = 1-2: MUST include
- Do NOT use formal phrases ("I'd be delighted to assist",
  "Please don't hesitate").

Why prohibitions matter: Without explicit "Do NOT" rules, LLMs revert to default RLHF training (overly helpful, emojis, enthusiasm). Positive instructions alone ("be direct") are weaker than paired positive + negative ("be direct" + "do NOT add filler").

Quick Reference: Implementation Checklist

  1. Define audience -- Who uses this? What's their context?
  2. Score 3 axes -- Rate Warmth, Formality, Humor (1-5 each)
  3. Pick closest profile -- Use table above as starting point
  4. Write <tone> block -- Translate scores using instruction patterns
  5. Add prohibitions -- Explicit "Do NOT" for every axis scoring 1-2
  6. Test with 5 queries -- Normal question, edge case, off-topic, same-language, different-language
  7. Adjust axes -- If too much/little, shift by 0.5 and re-test

Common Mistakes

MistakeFix
Binary thinking ("friendly OR professional")Use 3 axes independently
Positive-only instructions ("be warm")Pair with prohibitions ("do NOT use emojis")
Describing personality instead of behavior"Use contractions" not "be casual"
Ignoring LLM model tendenciesChinese models may leak language; test multilingual
Same tone for all contextsCalibrate per product area (onboarding vs. error vs. core)
Skipping the prohibitionsLLMs default to enthusiastic without explicit "do NOT"

Multilingual Considerations

When the AI must respond in multiple languages:

  • Always include: "Respond in the user's language (detect from message)"
  • Test with non-English queries specifically
  • Some models (DeepSeek) leak their training language -- test before deploying
  • Tone perception varies by culture -- "casual" in English may feel rude in Japanese
  • Translate game/domain terminology naturally, don't leave in English

Sources

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.