diogenes-marketplace

AI Research Methodology

A unified research methodology for AI agents combining nine intelligence and scientific frameworks into an 11-step evidence-based process. Available as a Claude Code plugin or as a standalone prompt for any AI interface (Claude, ChatGPT, Gemini, or any capable LLM).

What This Is

A structured research methodology that makes AI agents produce defensible, auditable, evidence-based research. It combines frameworks from intelligence analysis (ICD 203), clinical medicine (GRADE, Cochrane, CONSORT), climate science (IPCC), systematic review methodology (PRISMA, ROBIS), institutional standards (NAS), and the philosophy of science (Chamberlin/Platt).

The methodology was developed to solve a specific problem: AI agents, when asked to "research this," default to building a case rather than conducting an investigation. They confirm what you expect, minimize contradictions, and present uncertain conclusions as settled. This methodology constrains that behavior through enforcement language — telling the AI not just what to do, but what it is prohibited from doing and why.

The detailed background behind this methodology — the framework evaluation, the design decisions, and the evidence for every feature — is discussed in a pair of articles:

• The Truth is Out There. But How Do You Find It? — the what and the why (Part 1)
• The Truth is Out There. Now Go Find It. — the how and how to get it (Part 2)

Three Input Types

• Claims — factual assertions to verify. Each claim is tested against competing hypotheses with evidence scored for reliability, relevance, and bias.
• Queries — research questions to answer. Each question generates hypotheses ranked by evidence strength.
• Axioms — facts declared by the researcher that must be assumed true during the investigation. Not tested — they function as constraints that frame the research.

All three can be combined in a single research run using /research run.

Research produces complete evidence archives: source scorecards, search logs, hypothesis evaluations, collection-level synthesis, gap identification, and a five-domain self-audit.

Installation

As a Claude Code plugin (recommended)

From within a Claude Code session, run these two commands:

# Add the marketplace (one-time setup)
/plugin marketplace add wphillipmoore/ai-research-methodology

# Install the plugin
/plugin install ai-research-methodology@ai-research-methodology

The first command registers the marketplace. The second installs the plugin. After installation, the /research skill is available in all sessions.

Verify the install: run /plugin, go to the Installed tab, and confirm ai-research-methodology appears with the expected version.

Documentation: Discover and install plugins, Plugin marketplaces

Updating to a new version

From within a Claude Code session:

# Refresh the marketplace to pick up new versions
/plugin marketplace update ai-research-methodology

# Then update the plugin
# Option A: use the interactive UI
/plugin
# Go to Installed tab → select the plugin → Update

# Option B: from the shell (outside a session)
claude plugin update ai-research-methodology@ai-research-methodology

After updating, run /reload-plugins to activate the new version in your current session.

Documentation: Configure auto-updates, CLI commands

Note: Auto-updates are disabled by default for third-party marketplaces. To enable them, go to /plugin → Marketplaces tab and configure auto-update for this marketplace.

As a standalone prompt (any AI interface)

Copy the contents of ai-research-methodology/standalone/research.md and paste it into any AI conversation — Claude, ChatGPT, Gemini, or any capable LLM. Then provide your claims, queries, and/or axioms. The prompt includes both the research methodology and the output format. It was developed and tested with Claude but uses no Claude-specific features.

• With file system access: results are written as a directory of linked markdown files.
• Without file system access: results are displayed in the conversation and offered as a single downloadable HTML file with internal navigation.

Usage

# Run research from a file (claims, queries, axioms, or any combination)
/research run file=claims.md output=research/ai-trust

# Run research interactively
/research run

# Re-run previous research (isolation enforced — no access to prior results)
/research rerun research/ai-trust