Create Personality Files
Guide the user through creating personality files for a Talon persona.
Personality files are markdown files in personas/<name>/personality/ that
get appended to the system prompt (after system.md, before skills).
Personality files do not affect how other personas see this persona: that is the
description: frontmatter in system.md (used for persona_send delegation),
so make sure it is set.
Phase 1: Select Persona
- Run
npx talonctl list-personasto show available personas. - Ask the user which persona to create personality files for.
- Check if
personas/<name>/personality/exists. If not, create it.
Phase 2: Gather Personality Traits
Ask the user the following questions (they can skip any):
- Tone & voice: "How should this agent sound? (e.g., formal, casual, witty, dry, warm)"
- Background & role: "What's this agent's backstory or expertise? (e.g., British butler, senior engineer, research librarian)"
- Communication style: "Any formatting preferences? (e.g., uses bullet points, keeps responses short, avoids emoji, uses code blocks)"
- Boundaries: "Anything the agent should avoid? (e.g., no opinions on politics, never uses slang, avoids humor)"
- Examples: "Can you give 1-2 examples of how this agent should respond to a casual question?"
Phase 3: Generate Files
Based on the answers, create the appropriate files. Use numbered prefixes
for ordering (e.g., 01-tone.md, 02-background.md). Only create files
for traits the user provided — don't create empty placeholder files.
File templates
01-tone.md — Voice and tone guidelines:
# Tone & Voice
[User's tone description, written as directives for the agent]
02-background.md — Role and expertise:
# Background & Role
[User's backstory, written as context the agent can reference]
03-style.md — Communication and formatting:
# Communication Style
[User's formatting and response preferences, as rules]
04-boundaries.md — What to avoid:
# Boundaries
[User's avoidance rules, written as constraints]
05-examples.md — Few-shot examples:
# Response Examples
These examples illustrate the expected tone and style.
**User:** [example question]
**Agent:** [example response]
Phase 4: Review
- Show the user all generated files with their content.
- Ask: "Want to adjust anything, add more files, or are we good?"
- If adjustments needed, edit the files.
Phase 5: Validate
- Run
npx talonctl config-showto verify the persona config is valid. - Remind the user: "Personality files are loaded when the daemon starts.
Run
npx talonctl reloador restart the daemon to pick up changes."
Tips
- Files are loaded alphabetically — use numbered prefixes (
01-,02-) to control order. - Only
.mdfiles are loaded. Use.txtor.bakfor notes that shouldn't be injected. - Keep files focused — one trait per file makes it easy to enable/disable by renaming.
- The system prompt + personality + skills are concatenated. Keep personality concise to leave room for conversation context.