perplexity-search

name	perplexity-search
description	Web-grounded SOTA discovery using Perplexity Sonar models via OpenRouter. Use when academic APIs return foundational papers but you need to know what's CURRENT — what has superseded older methods, what practitioners actually use today, and what was released in the last weeks/months.
user-invocable	false

When to Use This vs Academic Skills

Need	Use
Structured paper metadata (citations, abstracts, code repos)	arxiv-search, semantic-scholar-search, hf-papers-search
What's current SOTA — what replaced older methods	perplexity-search --sota
What happened in the last N days	perplexity-search --recent
Practitioner adoption, blog posts, framework releases	perplexity-search
Whether a 2022 paper is still relevant or superseded	perplexity-search --sota

Key insight: Citation count measures influence, not currency. A paper with 5000 citations from 2022 may be foundational but no longer SOTA. Perplexity provides the web-grounded context to distinguish "seminal" from "current best".

Setup

Requires OPENROUTER_API_KEY environment variable. Get one at https://openrouter.ai/keys (pay-as-you-go, ~$0.002-0.005 per query with sonar-pro).

export OPENROUTER_API_KEY='sk-or-v1-...'

Usage

# Locate the script
PPX=$(find ~/.claude/plugins -path "*/ai-frontier/skills/perplexity-search/scripts/search.mjs" 2>/dev/null | head -1)

# SOTA discovery — what's current best for a topic
node "$PPX" --sota "LLM agent reasoning frameworks beyond ReAct"

# Recent developments — what happened in last N days
node "$PPX" --recent "retrieval augmented generation" --days=30

# General web-grounded search
node "$PPX" "comparison of Tree-of-Thoughts vs ReAct vs LATS for agent planning"

# Use reasoning model for complex analysis
node "$PPX" --sota "efficient attention mechanisms" --model=sonar-reasoning-pro

Models

Model	Cost/query	Best for
sonar	~$0.001	Quick fact checks
sonar-pro (default)	~$0.003	SOTA discovery, general research
sonar-reasoning	~$0.007	Complex comparisons, tradeoff analysis
sonar-reasoning-pro	~$0.010	Deep technical analysis

Output

Returns unified JSON envelope compatible with other ai-frontier skills:

{
  "success": true,
  "query": "...",
  "source": "perplexity",
  "result_count": 1,
  "results": [{
    "answer": "Synthesized web-grounded answer with dates and sources...",
    "citations": ["https://arxiv.org/...", "https://github.com/..."],
    "model": "sonar-pro"
  }],
  "meta": {
    "timestamp": "...",
    "api_version": "openrouter-perplexity-v1",
    "tokens_used": { "prompt_tokens": N, "completion_tokens": N }
  }
}

Integration with Agents

Agents should use Perplexity after academic search, not instead of it:

1. Search arXiv + S2 + HF Papers → get papers with structured metadata
2. Call Perplexity --sota → get current SOTA context and supersession info
3. Merge: use Perplexity to annotate which papers are "foundational but superseded" vs "current SOTA"

The --sota mode automatically asks Perplexity to:

• Focus on the last 6 months
• Name what has superseded older approaches
• Include arxiv IDs and publication dates
• Distinguish "widely adopted" from "just published"

Graceful Degradation

If OPENROUTER_API_KEY is not set, the script returns:

{ "success": false, "error": "OPENROUTER_API_KEY not set...", "source": "perplexity" }

Agents should check success and continue with academic-only data if Perplexity is unavailable. This keeps the plugin functional without an API key.