structured-analysis

Structured Analysis Skill for Claude Code

AI-augmented Structured Analytic Techniques (SATs) from US Intelligence Community doctrine — implemented as an interactive Claude Code skill.

18 techniques across 6 analytical phases, with automated evidence gathering, three-layer self-correction, and mandatory citation enforcement.

What is Structured Analysis?

Structured Analytic Techniques are rigorous methods developed by the US Intelligence Community to combat cognitive bias in high-stakes analysis. They emerged from decades of intelligence failures — Pearl Harbor, Iraq WMD — where the problem was never lack of information, but lack of structured challenge to dominant assumptions.

This skill brings SATs to Claude Code as an interactive /analyze command. Ask a question, and the skill orchestrates evidence gathering, technique selection, structured execution, self-correction, and a cited final report — all grounded in the CIA Tradecraft Primer (2009) and modern empirical updates.

Quick Start

Prerequisites

• Claude Code installed and configured
• (Optional) Firecrawl MCP server for enhanced OSINT web scraping

Install via Plugin (Recommended)

# Add the marketplace
/plugin marketplace add blevene/structured-analysis-skill

# Install the plugin
/plugin install structured-analysis@blevene

That's it. Run /analyze to start your first analysis.

Install via Git Clone

If you prefer to work from source (or want to contribute):

git clone https://github.com/blevene/structured-analysis-skill.git
cd structured-analysis-skill

Then open Claude Code in the cloned directory — the skill is discovered automatically from the skills/ directory.

Manual Install (Non-Claude Code)

The skill files are plain Markdown — they work with any AI assistant that supports structured prompting. To use them manually:

1. Clone the repository (see above)
2. Open skills/structured-analysis/SKILL.md — this is the skill entry point with all instructions
3. Feed the contents of SKILL.md and protocols/orchestrator.md into your AI assistant as system context
4. For each technique, provide the relevant protocol file from protocols/techniques/ and template from templates/techniques/
5. The docs/library/ directory contains the full reference knowledge base if your assistant needs theoretical grounding

The key files to provide as context:

• skills/structured-analysis/SKILL.md — orchestration instructions
• skills/structured-analysis/protocols/orchestrator.md — mode routing and technique selection
• skills/structured-analysis/protocols/evidence-collector.md — evidence gathering process
• skills/structured-analysis/templates/report-template.md — output structure
• docs/library/00-prime.md — master reference for all techniques

Optional: OSINT Setup

The skill gathers web evidence automatically. Without any setup, it uses Claude Code's built-in WebSearch and WebFetch tools. For faster, more reliable OSINT with richer content extraction, set up Firecrawl:

claude mcp add firecrawl -e FIRECRAWL_API_KEY=your-api-key -- npx -y firecrawl-mcp

Then whitelist the tools in .claude/settings.local.json:

{
  "permissions": {
    "allow": [
      "mcp__firecrawl__firecrawl_search",
      "mcp__firecrawl__firecrawl_scrape",
      "WebSearch",
      "WebFetch"
    ]
  }
}

This whitelists both Firecrawl (primary) and WebSearch/WebFetch (fallback) so evidence collection runs without permission prompts. If Firecrawl isn't configured, the skill falls back to WebSearch/WebFetch automatically and suggests Firecrawl setup after the analysis.

Without any web tools, the skill still works — it just uses conversation context and local files only. Use --no-osint to skip web research explicitly.

First Analysis

/analyze What are the strategic implications of quantum computing for national cybersecurity?

The skill will:

1. Assess the problem and select appropriate techniques
2. Gather evidence from conversation context, local files, and web sources
3. Execute each technique with structured protocols
4. Self-correct through three validation layers
5. Auto-remediate any HIGH-severity analytical weaknesses (if found)
6. Present findings with a human review gate before finalization