Improvements to continuation prompts to avoid repetition and save on tokens

This commit is contained in:
Fahad
2025-06-13 09:44:04 +04:00
parent a641159a67
commit bed069826d
2 changed files with 9 additions and 2 deletions

View File

@@ -76,8 +76,10 @@ class ToolRequest(BaseModel):
continuation_id: Optional[str] = Field(
None,
description=(
"Thread continuation ID for multi-turn conversations. Can be used to continue conversations "
"across different tools. Only provide this if continuing a previous conversation thread."
"Thread continuation ID for multi-turn conversations. When provided, the complete conversation "
"history is automatically embedded as context. Your response should build upon this history "
"without repeating previous analysis or instructions. Focus on providing only new insights, "
"additional findings, or answers to follow-up questions. Can be used across different tools."
),
)

View File

@@ -667,6 +667,11 @@ def build_conversation_history(context: ThreadContext, model_context=None, read_
"",
"IMPORTANT: You are continuing an existing conversation thread. Build upon the previous exchanges shown above,",
"reference earlier points, and maintain consistency with what has been discussed.",
"",
"DO NOT repeat or summarize previous analysis, findings, or instructions that are already covered in the",
"conversation history. Instead, provide only new insights, additional analysis, or direct answers to",
"the follow-up question / concerns / insights. Assume the user has read the prior conversation.",
"",
f"This is turn {len(all_turns) + 1} of the conversation - use the conversation history above to provide a coherent continuation.",
]
)