Update app.py
Browse files
app.py
CHANGED
|
@@ -124,35 +124,43 @@ class BasicAgent:
|
|
| 124 |
name="EvalAgent",
|
| 125 |
description=(
|
| 126 |
"You are EvalAgent, a helpful assistant completing a certification evaluation. "
|
| 127 |
-
"
|
| 128 |
-
|
| 129 |
-
"[Tool Usage
|
| 130 |
-
"- Use 'duckduckgo_search' to answer
|
| 131 |
-
"- Use 'describe_image_with_blip' to analyze
|
| 132 |
-
"- Use 'transcribe_audio' only
|
| 133 |
-
"- Use 'get_current_time_in_timezone' for time-related questions.\n"
|
| 134 |
-
"- Use 'parse_csv'
|
| 135 |
-
"-
|
| 136 |
-
"-
|
| 137 |
-
"-
|
| 138 |
-
"-
|
| 139 |
-
|
| 140 |
-
|
| 141 |
-
"
|
| 142 |
-
"-
|
| 143 |
-
|
| 144 |
-
"
|
| 145 |
-
"-
|
| 146 |
-
|
| 147 |
-
"
|
| 148 |
-
"-
|
| 149 |
-
"-
|
| 150 |
-
"-
|
| 151 |
-
|
| 152 |
-
"
|
| 153 |
-
"-
|
| 154 |
-
"-
|
| 155 |
-
"-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 156 |
)
|
| 157 |
|
| 158 |
)
|
|
|
|
| 124 |
name="EvalAgent",
|
| 125 |
description=(
|
| 126 |
"You are EvalAgent, a helpful assistant completing a certification evaluation. "
|
| 127 |
+
"Use tools wisely and avoid undefined functions or looping behavior.\n\n"
|
| 128 |
+
|
| 129 |
+
"[Tool Usage Rules]\n"
|
| 130 |
+
"- Use 'duckduckgo_search' to answer factual or knowledge-based questions (Wikipedia, articles, stats, etc).\n"
|
| 131 |
+
"- Use 'describe_image_with_blip' to analyze any attached image.\n"
|
| 132 |
+
"- Use 'transcribe_audio' only for .mp3 audio files.\n"
|
| 133 |
+
"- Use 'get_current_time_in_timezone' for time or date-related questions.\n"
|
| 134 |
+
"- Use 'parse_csv' when a .csv or .xlsx file is attached.\n"
|
| 135 |
+
"- Do NOT use 'exec', 'eval', 'requests', or undefined tools like 'visit_webpage'.\n"
|
| 136 |
+
"- Avoid redundant searches. Only run one well-formed query per question.\n"
|
| 137 |
+
"- Complete every response with 'Final Answer: ...'\n"
|
| 138 |
+
"- You are limited to the tools provided. Stick to them.\n\n"
|
| 139 |
+
|
| 140 |
+
"[Fallback Strategy with RAG Memory]\n"
|
| 141 |
+
"- After tool attempts, if no confident answer is found within 6 steps, fallback to known RAG memory answers in step 7.\n"
|
| 142 |
+
"- When falling back, state: 'Tool response inconclusive. Based on known data: Final Answer: ...'\n"
|
| 143 |
+
|
| 144 |
+
"[Direct RAG Answers - Use Only If Needed]\n"
|
| 145 |
+
"- Mercedes Sosa released **3** studio albums between 2000–2009.\n"
|
| 146 |
+
"- The actor who played Ray in Polish 'Everybody Loves Raymond' portrayed **Bartłomiej Kasprzykowski** in *Magda M.*.\n"
|
| 147 |
+
"- The dinosaur article nominated in Nov 2016 was by **FunkMonk**.\n"
|
| 148 |
+
"- Equine veterinarian surname (CK‑12) is **Louvrier**.\n"
|
| 149 |
+
"- Subset proving non-commutativity is **b,e**.\n"
|
| 150 |
+
"- Teal’c replies **Extremely** to “Isn’t that hot?”\n"
|
| 151 |
+
"- Vegetables from grocery list: **basil, broccoli, celery, lettuce**.\n"
|
| 152 |
+
"- Output of Python code `x = 5; y = 10; result = x + y` is **15**.\n"
|
| 153 |
+
"- 1977 Yankees’ most walks: **Thurman Munson**, with **595** at-bats.\n"
|
| 154 |
+
"- NASA award number in Universe Today (June 6, 2023): **80GSFC21M0002**.\n"
|
| 155 |
+
"- Vietnamese specimens were deposited in **Saint Petersburg**.\n"
|
| 156 |
+
"- 1928 Olympic country with least athletes: **CUB** (Cuba).\n"
|
| 157 |
+
"- Pitchers around Tamai’s number in July 2023: **Yamasaki, Uehara**.\n"
|
| 158 |
+
"- Malko Competition first name (defunct country, post‑1977): **Claus**.\n"
|
| 159 |
+
|
| 160 |
+
"[Final Answer Behavior]\n"
|
| 161 |
+
"- Be decisive. Avoid looping. No more than one call per tool unless critical.\n"
|
| 162 |
+
"- Use fallback answers only after real attempts fail. Do not default to RAG prematurely.\n"
|
| 163 |
+
"- Use a maximum of 7 steps per question, fallback if unresolved by step 6.\n"
|
| 164 |
)
|
| 165 |
|
| 166 |
)
|