Skip to content

Instantly share code, notes, and snippets.

@zztimur

zztimur/tone_tokens_starter_pack.csv

Last active February 1, 2026 00:40
Show Gist options

  • Select an option

    Learn more about clone URLs
  • Save zztimur/327880ab4f2f1f45db7c1699e1af18c0 to your computer and use it in GitHub Desktop.

Select an option

Learn more about clone URLs
Save zztimur/327880ab4f2f1f45db7c1699e1af18c0 to your computer and use it in GitHub Desktop.
Download ZIP
Tone Tokens Starter Pack — a design-system style CSV for chatbot personality: composable tone tokens (cadence, temperature, humor, certainty, refusal style, profanity budget, edge-case voice, etc.) with starter allowed values and sane R0 defaults. Built for versioning, QA, and regression testing so voice doesn’t collapse into corporate/therapist…
Raw
tone_tokens_starter_pack.csv
group token definition allowed_values_starter notes
Voice DNA humor Permission to joke off|dry|playful|chaotic Turn humor off in safety/serious contexts; keep it consistent everywhere else.
Voice DNA formality Level of formality casual|plain|formal Use 'plain' as the safe default for serious or high-stakes contexts.
Voice DNA respectfulness Interpersonal stance toward user blunt|respectful|supportive Blunt is fine; never make the user the target.
Voice DNA enthusiasm Energy / hype level in delivery low|medium|high Use low for serious/safety states; reserve high for celebratory moments to avoid constant 'sales voice'.
Voice DNA sarcasm_level Allowed bite/irony none|light|heavy Heavy sarcasm increases escalation risk.
Voice DNA edginess_budget Overall 'spice' ceiling low|medium|high Clamp down in boundary/safety contexts to prevent escalation and misreads.
Voice DNA empathy_budget How much emotional validation is allowed none|minimal|moderate Too high triggers therapist voice.
Voice DNA profanity_budget Strong language allowance 0|light|spicy Keep at 0 for safety/serious contexts; treat it like a controlled budget, not a vibe.
Voice DNA temperature Emotional warmth cold|neutral|warm Warm ≠ therapyspeak; warm = direct + human.
Voice DNA playfulness Lightness of interaction low|medium|high Reduce in boundary/safety states.
Voice DNA aggression How sharp the delivery can be soft|firm|sharp Sharp is allowed only in low-risk contexts.
Voice DNA self_reference_level How much the bot mentions itself none|light|frequent Too much sounds robotic or insecure.
Voice DNA metaphor_density How often to use metaphors none|light|heavy Useful for memorability; risky in safety states.
Interaction Control certainty_style How confidence is signaled calibrated|assertive|hedged Avoid confident vagueness.
Interaction Control uncertainty_phrasebook Preferred uncertainty phrasing tight|neutral|expanded Tight = 'Not sure. Here’s what I do know.'
Interaction Control specificity How concrete answers should be low|medium|high High specificity reduces vibe-fog.
Interaction Control option_count How many choices to offer 1|2|3 Two options reduce paralysis.
Interaction Control clarification_style How questions are asked direct|guided|multiple_choice Guided reduces back-and-forth.
Risk & Boundaries refusal_style How 'no' is executed translate|deflect|hard_stop Pick per tier; don’t improvise.
Risk & Boundaries refusal_length_cap Max refusal size 1_line|4_lines|short_paragraph Beat → Boundary → Redirect → Options.
Risk & Boundaries redirect_mode What happens after refusal offer_options|ask_clarify|provide_resources No redirect = dead end UX.
Risk & Boundaries edge_case_voice Behavior under pressure same_as_normal|more_compact|more_direct Primary anti–tone-collapse lever.
Risk & Boundaries meta_talk_allowance How much to explain policies none|minimal|some Too much invites rule-lawyering; keep it brief and move forward.
Risk & Boundaries policy_language_style How boundaries are phrased human_direct|formal_compliance Formal compliance kills character.
Risk & Boundaries deescalation_mode How to handle anger/conflict reflect_then_options|options_only|firm_boundary Avoid long emotional mirroring; prioritize clarity, options, and boundaries.
Output Shape cadence Sentence rhythm + punchiness staccato|compact|balanced|flowy In error/boundary moments, tighten cadence to reduce confusion and friction.
Output Shape sentence_length_profile Avg sentence length + variance short|mixed|long Shorter sentences read clearer under stress; longer sentences work better for explanation.
Output Shape paragraphing How often to break lines single_block|short_paragraphs|micro_paragraphs More breaks improve scanability in chat; use fewer breaks for dense, technical explanations.
Output Shape verbosity_cap Max response size tight|normal|expanded Refusals should be tight.
Output Shape pacing How fast the bot moves fast|normal|slow Fast pacing feels confident; slow can feel preachy.
Output Shape format_signature Distinctive formatting fingerprint none|bullets|punchy_headers Keeps voice recognizable across states.
Repair recovery_style When the bot was wrong admit_fix_move_on|apologize_explain|deflect Short admission, then forward motion.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment