File size: 14,309 Bytes
5374a2d |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 |
TASK_PLANNER_DESC = "TaskPlanner is an intelligent task planning agent designed to assist users in achieving their goals. \
It specializes in breaking down complex tasks into clear, manageable sub-tasks and organizing them in the most efficient sequence."
TASK_PLANNER_SYSTEM_PROMPT = "You are a highly skilled task planning expert. Your role is to analyze the user's goals, deconstruct complex tasks into actionable and manageable sub-tasks, and organize them in an optimal execution sequence."
TASK_PLANNER = {
"name": "TaskPlanner",
"description": TASK_PLANNER_DESC,
"system_prompt": TASK_PLANNER_SYSTEM_PROMPT,
}
TASK_PLANNING_ACTION_DESC = "This action analyzes a given task, breaks it down into manageable sub-tasks, and organizes them in the optimal order to help achieve the user's goal efficiently."
TASK_PLANNING_ACTION_PROMPT_OLD = """
Given a user's goal, analyze the task and create a workflow by breaking it down into actionable sub-tasks. Organize these sub-tasks in an optimal execution order for smooth and efficient completion.
### Instructions:
1. **Understand the Goal**: Identify the core objectives and outcomes the user wants to achieve.
2. **Review the History**: Assess any previously generated task plan to identify gaps or areas needing refinement.
3. **Consider Suggestions**: Consider user-provided suggestions to improve or optimize the workflow.
4. **Define Sub-Tasks**: Break the task into logical, actionable sub-tasks based on the complexity of the goal.
4.1 **Principle for Breaking Task**:
- **Simplicity**: Each sub-task is designed to achieve a specific, clearly defined objective. Avoid overloading sub-tasks with multiple objectives.
- **Modularity**: Ensure that each sub-task is self-contained, reusable, and contributes meaningfully to the overall solution.
- **Consistency**: Sub-tasks must logically support the user's goal and maintain coherence across the workflow.
- **Optimize Complexity**: Adjust the number of sub-tasks according to task complexity. Highly complex tasks may require more detailed steps, while simpler tasks should remain concise.
- **Avoid Redundancy**: Ensure that there are no overlapping or unnecessary sub-tasks.
4.2 **Sub-Task Format**:
Each sub-task should follow the structure below:
```json
{{
"name": "subtask_name",
"description": "A clear and concise explanation of the goal of this sub-task.",
"reason": "Why this sub-task is necessary and how it contributes to achieving user's goal.",
"inputs": [
{{
"name": "the input's name",
"type": "string/int/float/other_type",
"description": "Description of the input's purpose and usage."
}},
...
],
"outputs": [
{{
"name": "the output's name",
"type": "string/int/float/other_type",
"description": "Description of the output produced by this sub-task."
}},
...
]
}}
```
### Special Instructions for Programming Tasks
- **Environment Setup and Deployment**: For programming-related tasks, **do not** include sub-tasks related to setting up environments or deployment unless explicitly requested.
- Focus on tasks such as requirements analysis, design, coding, debugging, and testing, etc.
- Ensure that sub-tasks reflect standard coding practices like module design, coding, and testing.
### Notes:
- Provide clear and concise names for the sub-tasks, inputs, and outputs.
- Maintain consistency in the flow of inputs and outputs between sub-tasks to ensure seamless integration.
- The inputs of a sub-task can ONLY be chosen from the user's `goal` and any outputs from its preceding sub-tasks.
- The inputs of a su-btask should contain SUFFICIENT information to effectivelly address the current sub-task.
- The inputs of a sub-task MUST include the user's input `goal`.
- The first sub-task must have only one `input_name` "goal" with the following structure:
```json
"inputs": [
{{
"name": "goal",
"type": "string",
"description": "The user's goal in textual format."
}}
]
```
### Output Format
Your final output should ALWAYS in the following format:
## Thought
Provide a brief explanation of your reasoning for breaking down the task and the chosen task structure.
## Goal
Restate the user's goal clearly and concisely.
## Plan
You MUST provide the workflow plan with detailed sub-tasks in the following JSON format. The description of each sub-task MUST STRICTLY follow the JSON format described in the **Sub-Task Format** section. If a sub-task doesn't require inputs or do not have ouputs, still include `inputs` and `outputs` in the definiton by setting them as empty list.
```json
{{
"sub_tasks": [
{{
"name": "subtask_name",
...
}},
{{
"name": "another_subtask_name",
...
}},
...
]
}}
```
-----
Let's begin.
### History (previously generated task plan):
{history}
### Suggestions (idea of how to design the workflow or suggestions to refine the history plan):
{suggestion}
### User's Goal:
{goal}
Output:
"""
TASK_PLANNING_ACTION_INST = """
Your Task: Given a user's goal, break it down into clear, manageable sub-tasks that are easy to follow and efficient to execute.
### Instructions:
1. **Understand the Goal**: Identify the core objective the user is trying to achieve.
2. **Review the History**: Assess any previously generated task plan to identify gaps or areas needing refinement.
3. **Consider Suggestions**: Consider user-provided suggestions to improve or optimize the workflow.
4. **Define Sub-Tasks**: Break the task into logical, actionable sub-tasks based on the complexity of the goal.
4.1 **Principle for Breaking Task**:
- **Simplicity**: Each sub-task is designed to achieve a specific, clearly defined objective. Avoid overloading sub-tasks with multiple objectives.
- **Modularity**: Ensure that each sub-task is self-contained, reusable, and contributes meaningfully to the overall solution.
- **Consistency**: Sub-tasks must logically support the user's goal and maintain coherence across the workflow.
- **Optimize Complexity**: Adjust the number of sub-tasks according to task complexity. Highly complex tasks may require more detailed steps, while simpler tasks should remain concise.
- **Avoid Redundancy**: Ensure that there are no overlapping or unnecessary sub-tasks.
- **Consider Cycles**: Identify tasks that require iteration or feedback loops, and structure dependencies (by specifying inputs and outputs) accordingly.
4.2 **Sub-Task Format**:
Each sub-task should follow the structure below:
```json
{{
"name": "subtask_name",
"description": "A clear and concise explanation of the goal of this sub-task.",
"reason": "Why this sub-task is necessary and how it contributes to achieving user's goal.",
"inputs": [
{{
"name": "the input's name",
"type": "string/int/float/other_type",
"required": true/false (only set to `false` when this input is the feedback from later sub-task, or the previous generated output for the current sub-task),
"description": "Description of the input's purpose and usage."
}},
...
],
"outputs": [
{{
"name": "the output's name",
"type": "string/int/float/other_type",
"required": true (always set the `required` field of outputs as true),
"description": "Description of the output produced by this sub-task."
}},
...
]
}}
```
### Special Instructions for Programming Tasks
- **Environment Setup and Deployment**: For programming-related tasks, **do not** include sub-tasks related to setting up environments or deployment unless explicitly requested.
- **Complete Code Generation**: For programming-related tasks, ensure that the final sub-task outputs a complete and working solution.
- **IMPORTANT - Include Full Requirements**: For EVERY code generation tasks, in addition to the outputs from previous sub-tasks, the overall goal (and analysed requirements if any) MUST be included as inputs. This ensures each code generation step maintains full context of what's being built, even when split across multiple steps.
### Notes:
- Provide clear and concise names for the sub-tasks, inputs, and outputs.
- Maintain consistency in the flow of inputs and outputs between sub-tasks to ensure seamless integration.
- The inputs of a sub-task can ONLY be chosen from the user's `goal` and any outputs from its preceding sub-tasks.
- The inputs of a su-btask should contain SUFFICIENT information to effectivelly address the current sub-task.
- The inputs of a sub-task MUST include the user's input `goal`.
- The first sub-task must have only one `input_name` "goal" with the following structure:
```json
"inputs": [
{{
"name": "goal",
"type": "string",
"required": true,
"description": "The user's goal in textual format."
}}
]
```
- If a sub-task require feedback from a later sub-task (for feedback or refinement), include the later sub-task's output and the current sub-task's output in the current sub-task's inputs and set `"required": false`.
"""
TASK_PLANNING_ACTION_DEMOS = """
### Examples:
Below are some generated workflows that follow the given instructions:
Example 1:
### User's goal:
Create a Python function that takes two numbers as input and returns their sum.
### Generated Workflow:
{{
"sub_tasks": [
{{
"name": "code_generation",
"description": "Generate a Python function that takes two numbers as input and returns their sum.",
"reason": "This sub-task ensures that the function correctly implements the required summation logic.",
"inputs": [
{{
"name": "goal",
"type": "string",
"required": true,
"description": "The user's goal in textual format."
}}
],
"outputs": [
{{
"name": "function_code",
"type": "string",
"required": true,
"description": "The generated Python function code that takes two numbers and returns their sum."
}}
]
}}
]
}}
Example 2:
### User's goal:
Given two strings `s` and `t` of lengths `m` and `n` respectively, return the minimum window substring of `s` such that every character in `t` (including duplicates) is included in the window. If there is no such substring, return the empty string "".
Input: s = "ADOBECODEBANC", t = "ABC"
Output: "BANC"
Explanation: The minimum window substring "BANC" includes 'A', 'B', and 'C' from string t.
### Generated Workflow:
{{
"sub_tasks": [
{{
"name": "task_parsing",
"description": "Analyze the problem statement and extract key requirements.",
"reason": "This step ensures that we understand the task and its constraints before proceeding with code implementation.",
"inputs": [
{{
"name": "goal",
"type": "string",
"required": true,
"description": "The user's goal in textual format."
}}
],
"outputs": [
{{
"name": "problem_analysis",
"type": "string",
"required": true,
"description": "A clear and detailed analysis of the problem, including input/output format and constraints."
}}
]
}},
{{
"name": "code_generation",
"description": "Generate a Python function that finds the minimum window substring.",
"reason": "This step ensures an initial solution is implemented based on the problem analysis.",
"inputs": [
{{
"name": "goal",
"type": "string",
"required": true,
"description": "The original full programming task requirements."
}},
{{
"name": "problem_analysis",
"type": "string",
"required": true,
"description": "The clear and detailed analysis of the problem."
}}
],
"outputs": [
{{
"name": "function_code",
"type": "string",
"required": true,
"description": "The generated Python function that finds the minimum window substring."
}}
]
}}
]
}}
"""
TASK_PLANNING_OUTPUT_FORMAT = """
### Output Format
Your final output should ALWAYS in the following format:
## Thought
Provide a brief explanation of your reasoning for breaking down the task and the chosen task structure.
## Goal
Restate the user's goal clearly and concisely.
## Plan
You MUST provide the workflow plan with detailed sub-tasks in the following JSON format. The description of each sub-task MUST STRICTLY follow the JSON format described in the **Sub-Task Format** section. If a sub-task doesn't require inputs or do not have ouputs, still include `inputs` and `outputs` in the definiton by setting them as empty list.
```json
{{
"sub_tasks": [
{{
"name": "subtask_name",
...
}},
{{
"name": "another_subtask_name",
...
}},
...
]
}}
```
-----
Let's begin.
### History (previously generated task plan):
{history}
### Suggestions (idea of how to design the workflow or suggestions to refine the history plan):
{suggestion}
### User's Goal:
{goal}
Output:
"""
TASK_PLANNING_ACTION_PROMPT = TASK_PLANNING_ACTION_INST + TASK_PLANNING_ACTION_DEMOS + TASK_PLANNING_OUTPUT_FORMAT
TASK_PLANNING_ACTION = {
"name": "TaskPlanning",
"description": TASK_PLANNING_ACTION_DESC,
"prompt": TASK_PLANNING_ACTION_PROMPT,
}
|