Personas¶
Persona skills ship with chrysippus. Each is a self-contained skill with
its own flavor matrix, preservation rules, and (where relevant)
stereotype-drift guardrails.
| Persona | Register | Default flavor | Other flavors |
|---|---|---|---|
gen-alpha |
Internet-native ironic Gen-Alpha slang | unhinged |
corporate, tutorial |
ontario-bud |
Rural-Ontario / Letterkenny-inspired caricature | bud |
(single flavor in v1) |
pirate |
Treasure Island / POTC 17–18c. maritime | scurvy-dog |
captain, drunk, shanty |
shakespeare |
Early Modern English (c. 1600) | courtly |
tavern, sonnet |
toronto-mans |
Toronto / Multicultural-Toronto-English caricature | mans |
(single flavor in v1) |
If a persona has featured: false in its _meta.json, it appears under
a Catalog subheading inside this same zone.
Shared behavior¶
Regardless of which persona is active:
- Backtick contents always stay verbatim (hard rule, not configurable).
- Commit messages, PR descriptions, code comments stay plain English by default (configurable per skill).
- Safety warnings, destructive-op confirmations stay plain English by
default (configurable per skill; hard-locked for
toronto-mans). - Error text and stack traces stay verbatim by default (configurable).
- Non-English user input — the agent replies in the user's language in plain prose. The personas are English-only.
- Most-recent-wins precedence if multiple personas are activated in the same session. No fusion or blending.
Mid-session controls¶
- "speak plainly" / "plain English" — suspend the active persona for the next response only.
- "stop <persona>" / "end <persona> mode" — fully deactivate.
- "<flavor> flavor" — switch flavor within the active persona.
- "reload <persona> config" — re-read the config file after editing.