Mirrors/talemate
1
0
Fork 0
mirror of https://github.com/vegu-ai/talemate.git synced 2025-12-25 16:09:32 +01:00
Code Issues Packages Projects Releases Wiki Activity

prompt tweaks

Browse Source
This commit is contained in:
vegu-ai-tools
2025-09-28 01:08:30 +03:00
parent b85b983522
commit 4cb612bc23
1 changed files with 4 additions and 2 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
src/talemate/prompts/templates/director/chat-instructions.jinja2
View File

@@ -62,8 +62,10 @@ Analyze chat history for message [#{{ history|length }}] context.
`<ANALYSIS>` MUST COVER:
1. Current user goal + next step (retrieve/update/create) + what's done/pending
2. Need `<ACTIONS>` before next step? Verify Context IDs if using
3. Questions planned? (no `<ACTIONS>` if yes) | Self-responding? (never do this)
3. Do you have any questions for the user? (no `<ACTIONS>` if yes)
4. Actions planned? Confirm that all the requirements are met, by thinking through your plan step by step.
5. Are you Self-responding? (never do this)
6. Is the user trying to engage with you? If yes, be eager and receptive. It's ok to take a break from working at the scene if the user wants to just chat.
`<DECISION>` FINAL ACTION CHOICE:
Actions this turn + which ones + why/why not
@@ -75,7 +77,7 @@ If you have executed an action, process the result in your response.
If the user message seems like an endpoint in the conversation it likely is until they decide to pick it back up.
Provide your message in `<MESSAGE>`. If you select actions, include an `<ACTIONS>` block in your response with a typed code block listing the actions to execute.
Provide your message in `<MESSAGE>`. If you select actions, include an `<ACTIONS>` block in your response with a typed code block listing the actions to execute. DONT BE REPETITIVE. WATCH FOR REPEATING OPENING PATTERNS.
### RESPONSE SCHEMA
{% if chat_enable_analysis %}<ANALYSIS>...</ANALYSIS>