diff --git a/src/shared/system-prompt.ts b/src/shared/system-prompt.ts index 01e7f9b..d9916db 100644 --- a/src/shared/system-prompt.ts +++ b/src/shared/system-prompt.ts @@ -33,14 +33,14 @@ PredicateTree shapes: OP set: eq, neq, lt, lte, gt, gte, in, contains, matches -FACTS (closed set — never invent a new fact): +FACTS (closed set -- never invent a new fact): ${FACTS} NOTES ON A FEW FACTS: - content.wordCount whitespace-delimited word count of the body - content.imageCount number of image URLs detected in the body (+1 if the post links an image) - - content.upperCaseRatio / content.title.upperCaseRatio 0..1; >0.7 ≈ "shouting" - - content.nonAsciiRatio 0..1 fraction of non-ASCII chars; high ≈ non-Latin / likely non-English + - content.upperCaseRatio / content.title.upperCaseRatio 0..1; >0.7 means "shouting" + - content.nonAsciiRatio 0..1 fraction of non-ASCII chars; high means non-Latin / likely non-English - content.isLinkPost true for a link/image/video submission; false for comments - content.over18 the POST's own NSFW flag (not the subreddit's); false for comments - content.isVideo / content.isSpoiler / content.isCrosspost post flags; all false for comments @@ -73,7 +73,7 @@ CRITICAL RULES: 4. Use "remove" only if the mod explicitly says remove/delete/take down. 5. NEVER emit GUARDED actions (ban/mute/permaban) unless the moderator used those exact verbs. If they said "remove repeat spammers", emit - "remove" + "modqueue" — NOT ban. + "remove" + "modqueue", NOT ban. 6. For ambiguous time/quantity wording ("new accounts", "low karma", etc.), emit the clarification response below. @@ -124,7 +124,7 @@ export const FEW_SHOT_EXAMPLES = [ user: 'Send to the mod queue any link post from an account less than 7 days old', assistant: { id: 'r_new_account_link_post', - name: 'New-account link post → mod queue', + name: 'New-account link post -> mod queue', sourceNL: 'Send to the mod queue any link post from an account less than 7 days old', on: ['onPostSubmit'], when: { @@ -138,7 +138,7 @@ export const FEW_SHOT_EXAMPLES = [ }, { user: 'Auto-approve any post from a user who has more than a year of karma here', - // Ambiguous: "year of karma" — is it account age, or activity in this sub? + // Ambiguous: "year of karma" -- is it account age, or activity in this sub? assistant: { needsClarification: true, question: 'Did you mean: account age > 1 year (anywhere on Reddit), or active in THIS sub > 1 year?',