Repository navigation
Conversation
Answer in the first sentence, bold the values the owner acts on, and drop recaps, sources and caveats that do not change what to do.
|
Hey, thanks for this! I get the problem, some models (GPT especially) get lazy and turn every reply into a report. I really like the "reply like a person in a chat, not a report" part. The rest feels too opinionated to be the default for every bot, though. How a bot talks is personal: some people want it chatty, others want it short and robotic, and if this is baked in, their own instructions end up fighting it. What if we leave the built-in rule as it is and add a global instructions box instead? You write general instructions once and each bot can opt in or out. That way you can use this style on the bots you want, and nobody else gets it unless they ask for it. What do you think? |
|
Fair point, let's drop this one. What I was really after was a Grok-style default: short, direct, conversational answers out of the box, without having to tune each bot. Your global instructions box is a better home for that. I'll close this PR and open a new one for the global box, with this style available as a preset that each bot can opt in or out of. Sound good? |
Bots replied like reports: the answer, then sources, fees, caveats and what they could not do. This rewrites the built-in reply rule every bot gets so replies read like chat.
Existing bots pick the rule up when they are next configured.
Testing
bun run checkandbun run lintpass.Screenshots
Taken from a local build with PRs 27 to 30 applied together, using a throwaway demo bot; other sidebar names and previews are replaced with demo text.