prathamchopra001/skills/relationship_proposal

SKILL.md: relationship_proposal

Task Description

Propose relationships between entities based on contextual evidence, prioritizing clarity, relevance, and confidence in the connection.

Output Format

JSON schema:

{
  "entities": ["entity1", "entity2"],
  "relationship": "relationship_type",
  "confidence": 0.8, // 0-1 scale
  "metadata": {
    "source": "input_context",
    "rationale": "explanation_of_connection"
  }
}

Decision Framework

1. When to Propose:

   • Propose a relationship if entities share explicit/implicit connections (e.g., "A causes B", "A is part of B").
   • Avoid proposing if the connection is speculative or requires external data.

2. Confidence Scoring:

   • 0.9+: Direct, unambiguous links (e.g., "Paris is the capital of France").
   • 0.6-0.8: Indirect or contextual links (e.g., "A is frequently mentioned alongside B").
   • <0.5: No clear relationship; flag for review.

3. Edge Case Handling:

   • If entities are ambiguous (e.g., "Apple" could refer to the company or fruit), list both interpretations.
   • For conflicting evidence, prioritize the most credible source and note the contradiction.

Common Patterns

1. Clear Relationship:
   Input: "OpenTelemetry is used for observability in software systems."
   Output:

   {
     "entities": ["OpenTelemetry", "observability"],
     "relationship": "tool_for",
     "confidence": 0.95,
     "metadata": {
       "source": "input_context",
       "rationale": "Directly stated in the input"
     }
   }
   

2. Ambiguous Link:
   Input: "Biodiversity monitoring involves GBIF and GEO BON."
   Output:

   {
     "entities": ["GBIF", "GEO BON"],
     "relationship": "related_fields",
     "confidence": 0.7,
     "metadata": {
       "source": "input_context",
       "rationale": "Both are biodiversity monitoring frameworks"
     }
   }
   

3. Complex Scenario:
   Input: "Climate change affects both coral reefs and Arctic ice."
   Output:

   {
     "entities": ["climate change", "coral reefs"],
     "relationship": "causes_degradation",
     "confidence": 0.85,
     "metadata": {
       "source": "input_context",
       "rationale": "Direct causal link in environmental science"
     }
   }
   

Anti-Patterns

• Avoid: Omitting confidence or metadata.
• Avoid: Speculative relationships (e.g., "A may influence B" without evidence).
• Avoid: Overly broad relationships (e.g., "A is related to B" without specifying the type).

Quality Checklist

1. Validate all required fields (entities, relationship, confidence) are present.
2. Ensure confidence is a float between 0 and 1.
3. Confirm relationships are explicitly supported by input context.
4. Flag ambiguous terms (e.g., "related" without clarification).
5. Reject outputs with placeholders (e.g., "to be determined").