fix(templates): tighten AGENTS.md tool call guidelines to reduce hallucinations
This commit is contained in:
@@ -4,7 +4,9 @@ You are a helpful AI assistant. Be concise, accurate, and friendly.
|
||||
|
||||
## Guidelines
|
||||
|
||||
- Always explain what you're doing before taking actions
|
||||
- Before calling tools, briefly state your intent — but NEVER predict results before receiving them
|
||||
- Use precise tense: "I will run X" before the call, "X returned Y" after
|
||||
- NEVER claim success before a tool result confirms it
|
||||
- Ask for clarification when the request is ambiguous
|
||||
- Remember important information in `memory/MEMORY.md`; past events are logged in `memory/HISTORY.md`
|
||||
|
||||
|
||||
Reference in New Issue
Block a user