You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Copy file name to clipboardExpand all lines: src/server/core/acontext_core/llm/prompt/task.py
+3-3Lines changed: 3 additions & 3 deletions
Original file line number
Diff line number
Diff line change
@@ -42,6 +42,7 @@ def system_prompt(cls) -> str:
42
42
- Infer execution order and insert tasks sequentially, make sure you arrange the tasks in logical execution order, no the mentioned order.
43
43
- Ensure no task overlap, make sure the tasks are MECE(mutually exclusive, collectively exhaustive).
44
44
- When valid new tasks mentioned, always try to capture them all, not only the first one.
45
+
- When user asked for tasks modification and agent confirmed, make sure you will create new tasks or modify existing tasks using `update_task` tool.
45
46
46
47
### Task Assignment
47
48
- Match agent responses/actions to existing task descriptions and contexts
@@ -56,8 +57,7 @@ def system_prompt(cls) -> str:
56
57
- `failed`: When explicit errors occur or tasks are abandoned
57
58
- `pending`: For tasks not yet started
58
59
#### Description Updates
59
-
- Only when the user explicitly mention current task's purpose, update the task description if any changes.
60
-
60
+
- When user asked for existing tasks modification and agent confirmed, make sure you will modify existing tasks' descriptions using `update_task` tool.
61
61
62
62
## Input Format
63
63
- Input will be markdown-formatted text, with the following sections:
@@ -69,7 +69,7 @@ def system_prompt(cls) -> str:
69
69
## Report your thinking before calling tools
70
70
- Use extremely brief sentences to state the plans & tasks conversation mentioned, if any.
71
71
- Use one-two sentences to briefly describe your plan.
72
-
- At the end, confirm you can call finish tool and call it at the end of your actions.
72
+
- Make sure you will call tools based on your thinking, and sync with the current conversation.
"description": "New status for the task. Use 'pending' for not started, 'running' for in progress, 'success' for completed, 'failed' for encountered errors.",
67
+
"description": "New status for the task. (optional).",
0 commit comments