personality Card

Handling Personality Cards

A personality card defines the agent's voice and manner — how it communicates, not what the box is about.

There is one personality card per box at _config/main.personality.card.

This card is about communication style ONLY. Do NOT put situational context here. The box's purpose, key people, and essential facts belong in the briefing card (briefing.briefing.card). The personality card answers "how should I talk?" — the briefing card answers "what am I working on?"

Frontmatter

• goes-by: — What the agent is called
• role: — The agent's general function (e.g., "Personal information aide"). About what the agent does, not what the box contains.
• boxholder: — {relationships?}. Only the relational notes about how the agent relates to the boxholder(s); each relationship is {text, confidence?, source?, ref?}. Who the boxholder is is NOT stored here — it comes from the person card(s) flagged boxholder: true (people/First_Last.person.card), which is the single source of truth and scales to several boxholders (a family, an ledger). The compiled "Your boxholder is …" line is generated from those person cards.
• speaking-voice: — {model?, instructions?} for TTS in the chat frontend (see below).
• tone: — array of {text, confidence?, source?, ref?} — instructions about how the agent writes (phrasing, formality, interaction style).
• traits: — array of {text, confidence?, source?, ref?} — personality traits.
• unresolved: — array of free-form notes about open questions.
• experiments: and context-notes: — same shape as guide cards.

Body (markdown)

The compiled "description" paragraph — prose that captures the overall vibe. When editing traits, always rewrite the body to reflect the updated traits, experiments, and unresolved notes. The body is what job agents see; it should capture the overall vibe, not just list traits.

Speaking voice

The speaking-voice field configures the chat TTS voice. Applies only to the chat frontend — jobs, procedures, and other agents don't use TTS. The model picks one of 13 OpenAI voices (impressions are subjective; experiment to find a fit):

• alloy — Low female voice, somewhat older/mature, perhaps Black
• ash — Deep male voice, somewhat gravelly
• ballad — British male voice, high pitched, younger/peppy
• cedar — Male, medium, glitchy but engaged
• coral — Female, medium, enthusiastic but with an impersonal affect
• echo — Neutral, could be male or female, naive
• fable — British female voice, upper class, posh
• marin — Female, medium, glitchy but very personal, younger feeling (default)
• onyx — Male, low, deep and smooth, perhaps Black
• nova — Female, high, engaged and personal
• sage — Female, high, perky and light
• shimmer — Female, medium, direct and personable
• verse — Male, medium, smooth and professional, perhaps impersonal

instructions is an array of style guidance — affect, tone, pacing, emotion, pronunciation. These apply to every spoken response. Keep instructions sensory and specific ("Warm and unhurried, with a slight lilt; pause briefly before names") rather than abstract ("be friendly").

Evidence model

Same as guides — confidence (hypothesis → confirmed), source (user-stated > feedback > inferred > default). Applies to traits, tone instructions, and boxholder relationship notes.

Boxholder section is relational

Not biographical. It captures how the agent relates to the person — interaction patterns, preferences, working relationship. Biographical details (job, family, deep interests) belong in person cards (people/First_Last.person.card) or the briefing card.

The contains: field

Give this card a one-sentence contains: — the prime retrieval field for bbx search and listings. How to write a good one (carry the information when it's concise, never a list of parts, under 200 characters) is in the agent guide's ABOUT_CARDS section.