Morty AgentSkill

Ultra Deep Research

Multi-agent, multi-provider research that outperforms single-pass AI research tools. Research quality is an architectural outcome, not a prompting problem.

Pipeline Steps

Search Providers

5–10

Parallel Agents

~$0.40

Avg Cost

⚖️ Three Laws

Never synthesize from memory when evidence is required.

Every important claim requires 2+ independent sources.

Always know what you don't know — track gaps explicitly.

🎯 Research Modes

⚡

Quick

1–2 agents
1–2 search rounds
30–90 seconds

~$0.05

📊

Standard

3–5 agents
2–3 rounds per stream
3–8 minutes

~$0.40

🔍

Exhaustive

5–10 agents
3–5 rounds + recursive
8–20 minutes

~$1.50

📡

Monitoring

2–3 agents
Delta-focused
2–5 minutes

delta

🔗 The 10-Step Pipeline

Classify Research Mode

Analyze query complexity to select Quick, Standard, Exhaustive, or Monitoring. Determines agent count, search depth, verification rigor, and budget.

Decompose the Query

Create a Research Plan with 3–12 independent workstreams, mandatory contrarian angle, and dependency mapping.

Route to Providers

Score providers (Tavily, Brave, Exa, Perplexity) per workstream. Assign primary + backup for cross-verification.

Spawn Parallel Subagents

Launch all independent workstreams as parallel Opus 4.6 subagents. Types: Scout, Extractor, Verifier, Contrarian, Entity.

Wait & Collect

Subagents auto-announce completion. No polling. All must return before proceeding.

Build Evidence Store

Deduplicate sources & claims. Score confidence (0–1). Map contradictions. Tag source tiers (1–4).

Gap Analysis & Recursive Deepening

Coverage → contradiction → diversity → freshness → adversarial checks. Spawn gap-fillers if needed. Up to 3 rounds.

Adversarial Challenge

Exhaustive mode: dedicated agent tries to break the thesis. Tests assumptions, searches for counter-evidence.

Synthesize Final Report

Opus 4.6 director reorganizes by theme. Inline citations, confidence assessment, mandatory contradictions section.

Output & Archive

Deliver report. Optional structured JSON for pipelines. Optional Obsidian save with YAML frontmatter.

🧭 Smart Provider Routing

Tavily

General research workhorse. AI-optimized results with citations. Site crawling & content extraction.

Brave

Fastest provider (<600ms). Independent non-Google/Bing index. Best for freshness and verification.

Exa

Semantic precision. 70M+ companies, 1B+ profiles. Academic papers. Grounded structured outputs.

Perplexity

Answer engine + raw search. Multi-query. Academic & SEC modes. Chain-of-thought reasoning.

📐 Architecture Flowchart

Complete pipeline flow from query classification through recursive deepening to final synthesis

🏆 What Makes It Better

✦Multi-provider triangulation — Different search indexes surface different results. Agreement across independent sources = high confidence.
✦Recursive deepening — Doesn't stop after one pass. Identifies gaps and fills them with targeted follow-up agents (up to 3 rounds).
✦Adversarial testing — Actively tries to disprove its own conclusions. If the thesis survives, report credibility increases.
✦Structured evidence scoring — Every claim has a confidence score (0–1) and source tier (1–4). No vibes-based conclusions.
✦Mandatory uncertainty disclosure — Reports always include Contradictions & Open Questions. Its absence = low-quality research.
✦Opus-powered throughout — Opus 4.6 workers and director. No quality compromises — research is only as good as the agents gathering evidence.

🚫 Anti-Patterns

✕Spawning agents with vague directives like "look into this more"

✕Letting synthesis become summarization instead of reasoning over evidence

✕Trusting a single search provider for important claims

✕Skipping gap analysis — the highest-ROI step in the entire pipeline

✕Over-decomposing: 50 agents for a 3-agent question = noise and waste

✕Producing reports without the Contradictions & Open Questions section

✕Presenting vendor claims as facts without flagging source bias