Skip to content

cyberswat/claude-plugin-analysis

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

6 Commits
 
 
 
 
 
 
 
 

Repository files navigation

Analysis Plugin for Claude Code

A structured thinking framework for analysis and decision making, grounded in intelligence analysis methodologies and strategic consulting practices.

Commands

Command Purpose
analysis:analyze Plan your analysis: suggests personas, identifies the key question, assesses complexity, recommends command sequence
analysis:research Gather external context, industry trends, evidence base. Runs as a subagent to keep main context clean.
analysis:evaluate Five perspectives with assumption tracking: Informed, Steelman, Skeptic, Practical, Contrarian
analysis:critique Adversarial stress test: severity rated flaws, pre mortem, cross domain concerns
analysis:synthesize Structured output: summary, critique integration, recommendation with confidence, categorized open questions
analysis:decide Decision landscape with viable paths, type detection, user checkpoint, structured decision record
analysis:quick Lightweight mode: compressed analysis in a single pass under 500 words
analysis:test Validate the framework: runs three scenarios and scores output quality

Usage

Full Framework

For complex, high stakes, or ambiguous decisions:

/analysis:analyze [your topic]

This will identify the key question, suggest personas, and output the commands to run:

/analysis:research [topic]. Approach as a [persona].
/analysis:evaluate [topic]. Approach as a [persona].
/analysis:critique [topic]. Approach as a [persona].
/analysis:synthesize
/analysis:decide

The analyze step also detects when adjacent domains are relevant and adds cross domain flags to the critique command automatically.

Quick Mode

For straightforward, reversible, or time constrained decisions:

/analysis:quick [your topic]. Approach as a [persona].

Skipping Steps

You can skip steps or run them out of order. The framework is designed for flexibility:

  • Run just research and evaluate if you want evidence and perspectives without the full pipeline
  • Jump straight to decide if you have already done your own analysis
  • Use critique standalone to stress test any idea

Testing the Framework

After making changes to any skill, validate with:

/analysis:test

This runs three predefined scenarios covering strategic, people, and technical decisions, then scores each step against a quality rubric.

Personas

Each command supports a persona that layers domain expertise onto the analysis:

  • "Approach as a financial analyst"
  • "Approach as a systems architect"
  • "Approach as a product manager"
  • "Approach as a cognitive systems designer"

The persona affects what evidence matters, what concerns arise, and how recommendations are framed. The analyze step generates rich personas with specific experience, context, and formative events rather than generic role titles.

Architecture

Research Subagent

The research step runs in a forked context via the analysis-researcher agent. This keeps high volume web search output out of the main conversation window. Research output is structured with key findings, evidence quality assessment, source citations, and identified gaps.

Decision Type Detection

The decide step detects three decision types and adapts its output format:

  • Clear Choice: For binary or technical decisions with strong evidence
  • Phased Approach: For decisions with significant uncertainty or multiple stages
  • Navigational: For people, coaching, and relationship decisions where principles matter more than rigid actions

Framework Validation

The test skill runs the full pipeline against three embedded scenarios and scores output quality across multiple dimensions per step. This provides a repeatable way to validate changes.

Installation

claude plugins add github:cyberswat/claude-plugin-analysis

License

MIT

About

Structured thinking framework for Claude Code: research, evaluate, critique, synthesize

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors