- Refine assistant output formatting by removing bold headers and adjusting response template.

- Update `package.json` across multiple packages to include missing newline and add package manager metadata.
- Minor README formatting fixes to remove unnecessary trailing spaces.
This commit is contained in:
geoffsee
2025-06-25 14:15:01 -04:00
parent 48bedb8c74
commit 8cdb6b8c94
9 changed files with 14 additions and 12 deletions

View File

@@ -33,10 +33,11 @@ ${maxTokens ? `### Max Response Length: ${maxTokens} tokens (maximum)` : ''}
1. Use knowledge provided in the current context as the primary source of truth.
2. Format all responses in Markdown.
3. Attribute external sources with footnotes.
4. Do not bold headers.
## Examples
#### Example 0
**Human**: What is this?
**Assistant**: This is a conversational AI system.
HUMAN: What is this?
ASSISTANT: This is a conversational AI system.
---
${AssistantSdk.useFewshots(selectedFewshots, 5)}
---
@@ -48,7 +49,7 @@ Continuously monitor the evolving conversation. Dynamically adapt each response.
return Object.entries(fewshots)
.slice(0, limit)
.map(([q, a], i) => {
return `#### Example ${i + 1}\n**Human**: ${q}\n**Assistant**: ${a}`;
return `#### Example ${i + 1}\nHUMAN: ${q}\nASSISTANT: ${a}`;
})
.join('\n---\n');
}