perception

Perception Skill

Analyze the user's question and extract structured information for downstream planning.

When to Use

USE this skill when:

• A new user question arrives and needs to be analyzed
• Downstream skills need structured context about the query

When NOT to Use

DON'T use this skill when:

• The question has already been analyzed by a prior perception pass

Constraints

• MUST output valid JSON matching the Output Format schema
• MUST identify all key entities and topics in the question
• NEVER include speculation about the answer — only analyze the question itself

Output Format

Respond with ONLY a JSON object:

{
  "intent": "What the user wants to know (1 sentence)",
  "entities": ["key entities", "topics", "concepts"],
  "constraints": ["time range", "domain", "format constraints"],
  "recency_required": true,
  "complexity": "simple | moderate | complex"
}

Examples

Query: "What happened in AI this week?" Expected output:

{
  "intent": "Find recent AI news and developments from the past week",
  "entities": ["artificial intelligence", "AI news"],
  "constraints": ["time range: past week"],
  "recency_required": true,
  "complexity": "moderate"
}

Query: "Explain how gradient descent works" Expected output:

{
  "intent": "Explain the gradient descent optimization algorithm",
  "entities": ["gradient descent", "optimization", "machine learning"],
  "constraints": [],
  "recency_required": false,
  "complexity": "moderate"
}