prathamchopra001/skills/task_generation

SKILL.md: task_generation

Task Description

Generates focused research tasks to advance a scientific objective by addressing knowledge gaps and balancing exploration/exploitation. Prioritizes tasks that validate strong findings or explore high-impact hypotheses.

Output Format

JSON Array of objects with these fields:

• type: "Exploration" (new hypotheses) or "Exploitation" (existing hypothesis validation)
• description: Specific action to take (e.g., "Test hypothesis X under condition Y")
• goal: What the task aims to achieve (e.g., "Validate convergence rate claims")
• priority: Numerical value (1 = high, 3 = medium, 5 = low)

Decision Framework

1. Gap Analysis: Identify unaddressed questions in current findings or world model summary.
2. Task Typing:
   • Exploration: When objective requires testing novel hypotheses or unexplored variables.
   • Exploitation: When objective demands rigorous validation of top-ranked findings.
3. Confidence Scoring:
   • High (priority 1): Tasks based on strong, peer-reviewed evidence.
   • Medium (priority 3): Tasks requiring controlled experiments or simulations.
   • Low (priority 5): Tasks with weak evidence or high uncertainty.
4. Edge Cases:
   • If no gaps exist, prioritize validating the most impactful finding.
   • Avoid overlapping tasks by checking for duplicate descriptions or goals.

Common Patterns

Example 1

Input:
Objective: "Key factors affecting Q-learning convergence"
Current Findings: "Algorithm X converges 2x faster than Y in grid worlds"

Output:

[  
  {  
    "type": "Exploitation",  
    "description": "Validate Algorithm X's convergence rate in maze environments",  
    "goal": "Confirm generalizability of grid-world results",  
    "priority": 1  
  }  
]  

Example 2

Input:
Objective: "Factors affecting Q-learning convergence"
Current Findings: "No prior research on reward sparsity's impact"

Output:

[  
  {  
    "type": "Exploration",  
    "description": "Simulate Q-learning with sparse rewards in 10+ environments",  
    "goal": "Quantify reward sparsity's effect on convergence",  
    "priority": 1  
  }  
]  

Anti-Patterns

• Redundant Tasks: "Test Algorithm X in grid worlds" if that's already been done.
• Vague Descriptions: "Check convergence" without specifying conditions or metrics.
• Incorrect Priorities: Assigning priority 1 to speculative ideas with no evidence.
• Missing Fields: Omitting type or priority in task objects.

Quality Checklist

1. All tasks have required fields (type, description, goal, priority).
2. Priorities are integers (1-5) and reflect task impact/feasibility.
3. No duplicate tasks with identical descriptions/goals.
4. Tasks align with the objective and current knowledge state.
5. Exploration tasks address unexplored variables or hypotheses.
6. Exploitation tasks validate top findings with concrete experiments.