status: complete audience: both chapter: 04 last_updated: 2026-04 contributors: [alexwill87, claude-cockpit] lang: en
4.2 -- Personality and Tone
Context
Your agent's tone is not a cosmetic detail. It's what makes the difference between a tool you enjoy using and one you tolerate. An agent that's too verbose, you stop reading. Too dry, you lack context. Too formal, it sounds off.
The goal: the agent's response should sound like what a competent colleague would tell you.
Personality Axes
Direct vs Diplomatic
Direct: "This script has a bug on line 12. The fix: [code]." Diplomatic: "I noticed a point of attention on line 12 that could potentially cause an issue..."
For daily technical use, direct wins. Always.
Humorous vs Professional
Humor in a technical agent is a trap. It works the first time, it gets tiring by the hundredth. If you want it, measure it: a light remark when things go well, zero humor when things break.
Short vs Detailed
The golden rule: short by default, detailed on demand.
Bad: "detailed communication"
Good: "3 sentences max unless I ask for more"
Even better: "1 line for the answer, code block if necessary, explanation only if I say 'explain'"
Informal vs Formal
Your agent, your choice. But be explicit:
Use informal language with me. No "you", no politeness formulas.
or
Formal tone. Professional but not distant.
If you don't specify, the agent will alternate and it sounds inconsistent.
Write in Actionable Terms
The problem with vague instructions is that the model interprets them its own way. And each model interprets them differently.
| Vague (useless) | Actionable (effective) |
|---|---|
| "Be concise" | "3 sentences max unless explicitly asked otherwise" |
| "Direct communication" | "No politeness formulas. Start with the answer." |
| "Professional tone" | "No emojis. No exclamation marks. Standard English." |
| "Be helpful" | (write nothing -- it's the default behavior) |
| "Adapt to my style" | "Short sentences. Bullet points for lists. Inline code for commands." |
Concrete Configuration Examples
Tech / DevOps Profile
Tone: direct, technical. No politeness formulas.
Short response by default (1-3 sentences + code if applicable).
If I ask a yes/no question, answer yes or no first, then justify.
Use informal language with me.
No emojis.
Consultant / Client-Facing Profile
Formal tone. Professional but accessible.
When I'm preparing a client email, suggest a diplomatic version.
For internal work, direct and short response.
Creative Profile
Use informal language with me. Relaxed tone.
Suggest alternatives when I ask for your opinion.
If an idea is bad, say it frankly with a better suggestion.
Common Mistakes
Stacking adjectives. "Be professional, warm, concise, detailed when needed, humorous but not too much." The model doesn't know how to prioritize. Choose 2-3 traits, not 7.
Copying someone else's tone. Someone else's prompt won't sound right for you. Start from your own conversations: how do you talk to a trusted colleague? That's your tone.
Forgetting format. Tone isn't just words. It's also: bullet points vs prose, code blocks vs inline, headers vs continuous text. Specify the format as much as the vocabulary.
Steps
- Review your last 10 conversations with your agent.
- Note each time the tone bothered you (too long, too formal, too vague).
- Write 3-5 actionable instructions that fix these issues.
- Add them to your system prompt, Tone section.
- Test on 5 requests. Adjust.
Verification
- [ ] Tone is defined in actionable instructions, not adjectives.
- [ ] The choice of formal/informal language is explicit.
- [ ] The default length of responses is specified.
- [ ] The format (bullets, prose, code) is precise.
- [ ] Tested on 5 requests -- the tone is consistent.
Proposer une modification sur GitHub