@willbooster/agent-skills 1.21.12 → 1.22.1
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/package.json
CHANGED
|
@@ -0,0 +1,15 @@
|
|
|
1
|
+
---
|
|
2
|
+
name: improve-prompt
|
|
3
|
+
description: Systematically improve system and user prompts by comparing current and refined model outputs.
|
|
4
|
+
---
|
|
5
|
+
|
|
6
|
+
# Prompt Improvement Workflow
|
|
7
|
+
|
|
8
|
+
1. Run the production code to capture the actual prompts and the model's corresponding responses.
|
|
9
|
+
2. Analyze the captured prompts and responses to identify specific areas for improvement.
|
|
10
|
+
3. Create debug tests without assertions in the `tests/debug` directory. These tests should output both the initially captured prompts/responses and the results of the modified prompts for direct comparison.
|
|
11
|
+
4. Iteratively refine the prompts using the debug tests. Run the tests to compare the model's new responses against the captured baseline until the desired improvement is achieved.
|
|
12
|
+
- Prompts should be concise and clear.
|
|
13
|
+
- Prompts in different threads should be independent, i.e., one should not rely on the context of another.
|
|
14
|
+
- When adding content, rewrite the entire prompt to keep it well-structured and highly cohesive, avoiding any signs of later additions.
|
|
15
|
+
5. Update the production code with the refined prompts.
|