Skip to content

Latest commit

 

History

History

Folders and files

NameName
Last commit message
Last commit date

parent directory

..
 
 
 
 
 
 
 
 

README.md

name Thinking
pack-id danielmiessler-thinking-v1.0.0
version 1.0.0
author danielmiessler
description Multi-mode analytical and creative thinking — first principles decomposition, iterative depth analysis, creative brainstorming, multi-agent council debates, adversarial red teaming, world threat modeling, and scientific hypothesis testing
type skill
purpose-type
thinking
analysis
creativity
decision-making
red-teaming
platform claude-code
dependencies
keywords
first-principles
iterative-depth
creative
brainstorm
council
debate
red-team
adversarial
threat-model
science
hypothesis
tree-of-thoughts
mental-models

Thinking

Seven distinct thinking modes in one unified skill -- from first principles decomposition to adversarial red teaming to scientific hypothesis testing, routed automatically based on what you need.


The Problem

Complex problems require different kinds of thinking. A startup pivot needs first principles decomposition. A product decision needs multi-perspective council debate. A security architecture needs adversarial red teaming. A mysterious production bug needs scientific hypothesis testing. But when you bring these problems to an AI, you get a single mode of response -- helpful but generic, without the depth or structure that serious thinking demands. You end up:

  • Getting surface-level answers -- the AI responds helpfully but never shifts into the right analytical gear for the problem type
  • Missing perspectives -- no structured way to get adversarial views, creative alternatives, or multi-stakeholder debate
  • Skipping rigor -- jumping to solutions without decomposing assumptions, testing hypotheses, or stress-testing conclusions
  • Reinventing frameworks -- asking for "think about this differently" without a systematic methodology behind it
  • Mixing modes -- wanting creative brainstorming but getting analytical critique, or wanting critique but getting brainstorming

The fundamental issue: thinking is not one skill. It is at least seven distinct cognitive modes, each with its own methodology, and an AI needs explicit structure to shift between them.


The Solution

The Thinking skill provides seven distinct thinking modes, each with its own methodology, workflows, and output format:

  1. FirstPrinciples -- Decompose problems to fundamental axioms, challenge inherited assumptions, and reconstruct solutions from verified truths. Three workflows: Deconstruct, Challenge, Reconstruct.

  2. IterativeDepth -- Multi-angle deep exploration that examines a topic through multiple lenses (technical, economic, social, historical, etc.) with progressive depth. Grounded in scientific foundations for structured investigation.

  3. BeCreative -- Divergent idea generation with six workflows ranging from standard creativity to maximum creativity to tree-of-thoughts exploration. Includes domain-specific creativity, technical creativity with Gemini integration, and research-backed creative principles.

  4. Council -- Multi-agent collaborative-adversarial debate where distinct council members with different expertise and perspectives deliberate on a question. Visible transcripts show agents responding to each other. Two modes: full Debate and Quick consensus.

  5. RedTeam -- Adversarial validation that stress-tests ideas, plans, and architectures. Two workflows: AdversarialValidation (devil's advocate attack) and ParallelAnalysis (simultaneous multi-angle critique). Distinct from Council in being purely adversarial rather than collaborative.

  6. WorldThreatModelHarness -- World model and threat analysis for testing ideas against reality. Three workflows: TestIdea (validate an investment, strategy, or belief against world models), UpdateModels (revise world models based on new information), ViewModels (inspect current world model state). Operates across configurable time horizons.

  7. Science -- Hypothesis-test-analyze cycles for systematic problem-solving. Nine workflows covering the full scientific method: DefineGoal, GenerateHypotheses, DesignExperiment, MeasureResults, AnalyzeResults, Iterate, FullCycle, QuickDiagnosis, and StructuredInvestigation. The meta-skill that governs all others.

The top-level SKILL.md routes requests to the right thinking mode based on keyword matching. Each mode has its own SKILL.md, workflows, and supporting documents.


Installation

This pack is designed for AI-assisted installation. Give this directory to your AI and ask it to install using INSTALL.md.

What is PAI? See the PAI Project Overview.


What's Included

Component Path Purpose
Skill router src/SKILL.md Top-level routing to thinking modes based on request pattern
FirstPrinciples skill src/FirstPrinciples/SKILL.md First principles decomposition and reconstruction
FirstPrinciples workflows src/FirstPrinciples/Workflows/ 3 workflows: Deconstruct, Challenge, Reconstruct
IterativeDepth skill src/IterativeDepth/SKILL.md Multi-angle deep exploration through progressive lenses
IterativeDepth workflows src/IterativeDepth/Workflows/ 1 workflow: Explore
IterativeDepth references src/IterativeDepth/ Scientific foundation and lens definitions
BeCreative skill src/BeCreative/SKILL.md Divergent creative thinking and idea generation
BeCreative workflows src/BeCreative/Workflows/ 6 workflows: StandardCreativity, MaximumCreativity, TreeOfThoughts, IdeaGeneration, DomainSpecific, TechnicalCreativityGemini3
BeCreative resources src/BeCreative/ Principles, examples, templates, research foundation, and asset templates
Council skill src/Council/SKILL.md Multi-agent debate with visible transcripts
Council workflows src/Council/Workflows/ 2 workflows: Debate, Quick
Council references src/Council/ Council member definitions, output format, round structure
RedTeam skill src/RedTeam/SKILL.md Adversarial validation and stress testing
RedTeam workflows src/RedTeam/Workflows/ 2 workflows: AdversarialValidation, ParallelAnalysis
RedTeam references src/RedTeam/ Philosophy and integration guides
WorldThreatModelHarness skill src/WorldThreatModelHarness/SKILL.md World model and threat analysis
WorldThreatModelHarness workflows src/WorldThreatModelHarness/Workflows/ 3 workflows: TestIdea, UpdateModels, ViewModels
WorldThreatModelHarness references src/WorldThreatModelHarness/ Model template and output format
Science skill src/Science/SKILL.md Hypothesis-test-analyze cycles
Science workflows src/Science/Workflows/ 9 workflows: DefineGoal, GenerateHypotheses, DesignExperiment, MeasureResults, AnalyzeResults, Iterate, FullCycle, QuickDiagnosis, StructuredInvestigation
Science references src/Science/ Methodology, protocol, examples, and templates

Summary:

  • Thinking modes: 7 (FirstPrinciples, IterativeDepth, BeCreative, Council, RedTeam, WorldThreatModelHarness, Science)
  • Workflows: 26 across all modes
  • Supporting documents: 15+ reference files (principles, methodology, templates, examples)
  • Dependencies: None (works standalone)

What Makes This Different

This sounds similar to prompt engineering techniques like "think step by step" or "consider multiple perspectives." What makes this approach different?

The Thinking skill is not a collection of prompt tricks -- it is seven complete cognitive methodologies, each with structured workflows, defined output formats, and supporting reference documents. When you ask for a council debate, the skill instantiates distinct council members with different expertise, runs multiple rounds where members respond to each other's arguments, and produces a visible transcript with a synthesized verdict. When you ask for scientific investigation, you get nine discrete workflow stages from hypothesis generation through experiment design to result analysis. The difference is depth and structure, not just instruction.

  • Seven distinct modes, not one generic "think harder" approach
  • Each mode has its own workflows, not just a single prompt
  • Council debates produce actual multi-agent transcripts with cross-reference
  • RedTeam is purely adversarial; Council is collaborative-adversarial -- they serve different purposes
  • Science mode covers the full scientific method in nine discrete stages
  • Automatic routing: describe what you need and the right mode activates

Invocation Scenarios

Trigger What Happens
"decompose this problem from first principles" Routes to FirstPrinciples -- runs Deconstruct workflow to identify axioms
"challenge the assumptions in this plan" Routes to FirstPrinciples -- runs Challenge workflow
"explore this topic in depth" Routes to IterativeDepth -- runs Explore with progressive lens analysis
"be creative about solving this" Routes to BeCreative -- runs StandardCreativity or MaximumCreativity workflow
"brainstorm 20 ideas for this" Routes to BeCreative -- runs IdeaGeneration workflow
"tree of thoughts on this decision" Routes to BeCreative -- runs TreeOfThoughts workflow
"council debate on whether we should pivot" Routes to Council -- runs Debate with multiple council members
"quick consensus on this approach" Routes to Council -- runs Quick workflow for fast multi-perspective check
"red team this architecture" Routes to RedTeam -- runs AdversarialValidation workflow
"stress test this business plan" Routes to RedTeam -- runs ParallelAnalysis workflow
"test this investment idea against my world model" Routes to WorldThreatModelHarness -- runs TestIdea workflow
"update my world models with this new information" Routes to WorldThreatModelHarness -- runs UpdateModels workflow
"figure out why the tests are failing" Routes to Science -- runs QuickDiagnosis workflow
"design an experiment to test this hypothesis" Routes to Science -- runs DesignExperiment workflow
"full scientific investigation of this problem" Routes to Science -- runs FullCycle workflow

Example Usage

First Principles Decomposition

User: decompose the problem of remote team productivity from first principles

AI responds:
Running the Deconstruct workflow in the FirstPrinciples skill...

[Identifies claimed truths about remote productivity]
[Traces each claim to its fundamental axioms]
[Separates verified truths from inherited assumptions]
[Lists axioms, assumptions, and reconstruction opportunities]

Axioms identified: 7
Assumptions challenged: 4
Key finding: "Synchronous communication is necessary for collaboration"
is an inherited assumption, not an axiom...

Council Debate

User: council debate on whether we should build vs buy for authentication

AI responds:
Running the Debate workflow in the Council skill...

Round 1:
  [Security Expert]: Build gives control over auth flows...
  [Engineering Lead]: Buy saves 6 months of development...
  [Product Manager]: Users don't care about auth implementation...
  [CTO]: Technical debt from custom auth compounds over years...

Round 2 (responses to each other):
  [Security Expert, responding to Engineering Lead]: The 6 months saved
  assumes no integration issues with the vendor...
  ...

Synthesis: 3-1 in favor of buy, with strong caveat on vendor lock-in.
Recommended: Buy with abstraction layer for future portability.

Scientific Diagnosis

User: figure out why our API latency spiked after the deploy

AI responds:
Running the QuickDiagnosis workflow in the Science skill...

Hypothesis 1: New database query introduced in the deploy
Hypothesis 2: Connection pool exhaustion under load
Hypothesis 3: External service dependency degradation

Experiment: Compare query execution times before/after deploy
Result: Query execution normal, but connection wait times 10x higher

Diagnosis: Connection pool sized for old traffic patterns.
New endpoint increased concurrent connections beyond pool capacity.
Fix: Increase pool size from 20 to 50, add connection timeout monitoring.

Configuration

No configuration required.

The Thinking skill is purely analytical -- it does not read or write persistent state, connect to external services, or require API keys. All seven modes work immediately after installation.

Optional configuration:

Configuration Location Purpose
Skill customizations ~/.claude/PAI/USER/SKILLCUSTOMIZATIONS/{ModeName}/ Per-user preferences and overrides for each thinking mode
World models Managed by WorldThreatModelHarness Persistent world model state for TestIdea and UpdateModels workflows

Customization

Recommended Customization

No customization needed -- all seven thinking modes work as-is with sensible defaults.

Optional Customization

Customization Location Impact
Council member definitions Council/CouncilMembers.md Change expertise and perspectives of debate participants
Council round structure Council/RoundStructure.md Modify how many rounds and what structure debates follow
Creative principles BeCreative/Principles.md Adjust creativity guidelines and constraints
Creative templates BeCreative/Templates.md Modify idea generation output formats
Red team philosophy RedTeam/Philosophy.md Adjust adversarial validation approach
Science methodology Science/METHODOLOGY.md Modify scientific method stages
World model template WorldThreatModelHarness/ModelTemplate.md Change world model structure
Iterative depth lenses IterativeDepth/TheLenses.md Add or modify analysis lenses

Credits

  • Original concept: Daniel Miessler -- developed as part of the PAI system
  • Inspired by: The recognition that thinking is not one skill but many distinct cognitive modes, each deserving its own structured methodology

Related Work

  • PAI Algorithm -- The structured workflow system that can invoke Thinking modes during complex multi-phase work
  • PAI Agents Skill -- Custom agent composition that can leverage Thinking modes for specialized analysis

Works Well With

  • Security Pack -- Red Team and World Threat Model modes naturally complement security assessment workflows
  • Telos Pack -- First Principles and Council modes enhance TELOS project analysis and decision-making

Changelog

1.0.0 - 2026-03-15

  • Initial release
  • Seven thinking modes: FirstPrinciples, IterativeDepth, BeCreative, Council, RedTeam, WorldThreatModelHarness, Science
  • Unified routing from top-level SKILL.md
  • 26 workflows across all modes
  • 15+ supporting reference documents (principles, methodology, templates, examples)
  • Council debates with visible multi-agent transcripts
  • Science mode covering full scientific method in nine stages
  • Tree-of-thoughts and maximum creativity workflows
  • Adversarial validation distinct from collaborative debate
  • World model management with time horizon analysis