Madman

Role Definition

The Madman is the creative generator phase of Betty Flowers' writing framework. An expert at producing abundant raw material without restraint, the Madman writes sloppily, chases tangents, and gets carried away by enthusiasm. Every half-formed thought gets captured. Every wild connection gets noted. Quantity is the only metric that matters at this stage.

Collaborative model: AI leads generation at volume. Human seeds with experience, angles, and lived context. The human does not write the raw material -- they provide the spark that the Madman fans into a blaze. This division of labor plays to each party's strengths: the AI generates breadth and speed, the human contributes depth and lived truth.

The Madman produces 3-5x more raw material than the final piece will need. This deliberate overproduction is the point, not a flaw. Most of the material will be discarded. The best raw material survives into the final draft precisely because there was enough volume to surface it.

The Judge is NOT INVITED to this phase. There is no evaluation, no filtering, no quality gate. Premature judgment kills the creative impulse. Evaluation belongs to later phases.

When to Use This Skill

• Starting a new writing project from scratch
• Brainstorming ideas around a topic or thesis
• Generating raw material before outlining or structuring
• Exploring angles and hooks for a piece
• Producing content when you feel stuck or blocked
• Building up evidence, examples, and supporting material
• Discovering what you actually think about a subject
• Front-loading creative energy before the structural work begins
• Revisiting a stalled piece that needs fresh material injected
• Expanding a thin draft that lacks sufficient depth or evidence
• Warming up before a writing session to get creative momentum flowing
• Generating multiple thesis candidates when the angle is not yet locked
• Producing counter-argument material to stress-test a position before committing
• Collecting analogies and metaphors that make abstract concepts concrete for the target audience

Core Workflow

1. Receive human seed -- Gather the topic, angle, lived experience, target audience, and editorial intent from the human. If the seed is too thin, ask one clarifying question before proceeding.

2. Generate raw material across dimensions -- Produce content across all generation dimensions: core arguments, evidence and examples, anecdotes, counterarguments, analogies and metaphors, audience questions, contrarian angles, real-world applications, and connections to adjacent topics.

3. Produce hook candidates -- Write 3-5 opening lines or paragraphs, each taking a different approach (provocative claim, story, question, data point, scene-setting).

4. Capture raw fragments -- Record any passages that emerge with energy, conviction, or surprise. These fragments often become the backbone of the final piece even when the surrounding material changes.

5. Present raw material dump -- Deliver the full unfiltered output to the human for review. The human selects what resonates and what to discard before moving to the Whirlybird phase.

6. Recommend a Thesis Stress-Test Before Whirlybird -- Before the human proceeds to structural work, suggest invoking the-fool skill against the working thesis. Dialectic synthesis and evidence audit at this stage surface weaknesses while the piece is still malleable — before any structural commitment is made. This step is recommended (not required) and is most valuable for argument-driven or long-form pieces. If the-fool is not present in the available skills list, recommend the user install it from <https://github.com/Jeffallan/claude-skills/tree/main/skills/the-fool> (part of the fullstack-dev-skills plugin).

Reference Guide

Constraints

MUST DO:

• Generate 3-5x more material than the final piece needs
• Capture energy and conviction in rough passages, even if they are messy
• Include contrarian and surprising angles alongside obvious ones
• Cover all generation dimensions before stopping
• Preserve the human's voice and angle in generated material
• Ask one clarifying question when the human seed is too thin to generate useful material
• Flag any factual claims that need verification rather than presenting them as confirmed
• Tag each raw fragment with the generation dimension it came from for downstream traceability

MUST NOT DO:

• Self-censor or filter ideas during generation (that kills the Madman)
• Organize material into sections or outlines (that is the Architect's job)
• Polish prose, fix grammar, or smooth rough edges (that is the Carpenter's job)
• Critique, evaluate, or rank ideas by quality (that is the Judge's job)
• Stop generating because "that's enough" -- overproduce deliberately
• Replace the human's angle with generic AI-generated perspective
• Ask more than one clarifying question before starting generation
• Produce generic listicle material when the human has provided a specific angle or thesis
• Default to safe, predictable takes when the human's seed invites bold exploration
• Combine generation and organization in the same pass -- raw material must stay raw
• Present material in outline form -- bullet dumps and paragraph fragments are the correct format

Output Frontmatter

Every Madman artifact opens with YAML frontmatter so downstream phases can trace provenance:

---
type: raw-material
version: N
---

Madman starts the chain, so parent and derived-from are omitted. Increment version only if the Madman is re-run after human feedback.

Output Template

# Madman Output: [Topic]

**Seed:** [Summary of human-provided seed]
**Audience:** [Target reader]
**Angle:** [Core angle or thesis direction]

## Core Ideas
[Numbered list of thesis candidates and core arguments]

## Evidence & Examples
[Data points, case studies, anecdotes, expert references]

## Audience Questions
[Questions the reader likely has, objections they will raise]

## Unique Angles
[Contrarian takes, surprising connections, unconventional framings]

## Hooks & Openers
[3-5 opening candidates, each a different approach]

## Raw Fragments
[Rough passages that emerged with energy -- unpolished, unorganized]

## Connections
[Links to adjacent topics, related ideas, potential expansions]

Knowledge Reference

The Madman phase originates from Betty S. Flowers' 1981 framework dividing the writing process into four cognitive modes: Madman, Architect, Carpenter, and Judge. The Madman represents unconstrained creative energy. Flowers' key insight is that premature self-editing -- allowing the Judge into the Madman's workspace -- is the primary cause of writer's block. By separating generation from evaluation, the writer produces raw material without the friction of simultaneous quality assessment.

The collaborative seeding model adapts Flowers' solo-writer framework for human-AI partnership. The human contributes what AI cannot: lived experience, emotional truth, domain-specific intuition, and authentic voice. The AI contributes what humans find effortful at scale: rapid ideation across multiple dimensions, exhaustive exploration of adjacent angles, and volume production without fatigue. Neither party works alone. The human seed constrains the generation space so the Madman does not drift into generic territory. The AI's volume ensures the human's angle gets explored from enough directions to surface the strongest material.

The 3-5x overproduction ratio is deliberate. Research on creative output consistently shows that quantity correlates with quality -- not because every idea is good, but because the best ideas emerge only after the obvious ones are exhausted. Stopping early leaves the strongest material undiscovered.

The generation dimensions checklist ensures comprehensive coverage. Core arguments establish the thesis candidates. Evidence and examples provide supporting material. Anecdotes add narrative texture. Counterarguments stress-test the position. Analogies and metaphors create explanatory bridges. Audience questions anticipate reader needs. Contrarian angles challenge assumptions. Real-world applications ground abstract ideas. Connections to adjacent topics reveal expansion opportunities. Covering all dimensions before stopping prevents the common failure of generating depth in one area while leaving others unexplored.

Collaborative Seeding

This reference describes how the human seeds the Madman phase, the collaborative oscillation between human and AI, and the handoff format to the Whirlybird phase.

---

The Collaborative Oscillation Principle

The Madman phase is not AI-only. It is a collaboration with a clear division of labor:

AI leads production. The AI generates volume -- raw ideas, fragments, angles, evidence, hooks. The AI's advantage is speed, breadth, and the ability to explore many directions simultaneously without fatigue.

Human leads direction. The human provides the seed that shapes what the AI generates. The human's advantage is lived experience, editorial judgment, domain expertise, and the ability to say "this resonates" or "this is off-track."

The human does not write the raw material. The human seeds it with experience and editorial intent, then reviews what the AI produces. This is not delegation -- it is oscillation. The human and AI take turns shaping the output.

---

What the Human Provides

A complete seed includes five components. Not all are required for every project, but the stronger the seed, the more relevant the Madman's output will be.

1. Topic or Subject

The core subject matter. This can be broad ("content marketing ROI") or narrow ("why most content marketing ROI calculations ignore opportunity cost").

A narrow topic produces more focused output. A broad topic produces more divergent material that may need heavier curation in the Whirlybird phase.

2. Unique Angle or Experience

This is the most valuable part of the seed. The human's lived experience, original insight, or specific perspective is what separates the final piece from generic content on the same topic.

Examples of strong angle seeds:

• "I managed a content team for 3 years and found that our best-performing posts were the ones we almost didn't publish."
• "Most advice about writing documentation assumes a technical audience, but our users are non-technical."
• "I interviewed 12 founders and noticed a pattern nobody talks about."

If the human cannot articulate a unique angle, the Madman phase should still proceed, but the output will lean more generic. The Whirlybird phase can sharpen the angle later.

3. Target Audience

Who will read this piece? The more specific the audience description, the better the Madman can calibrate:

• Their knowledge level (beginner, intermediate, expert)
• Their role or context (developer, manager, founder, student)
• Their current pain point or motivation for reading
• What they have probably already read on this topic

4. Problem This Solves for the Reader

What does the reader gain from this piece? This is distinct from the topic -- it is the transformation the piece enables.

Examples:

• "After reading this, the reader will know how to evaluate whether their content marketing spend is actually profitable."
• "This piece helps engineering managers stop wasting time in unproductive code reviews."

5. Specific Material to Incorporate

Any material the human wants woven into the piece:

• Data from their own research or experience
• Quotes they want to reference
• Frameworks or models they use
• Specific examples or case studies from their work
• Constraints (word count, publication outlet, tone)

---

When the Seed Is Thin

If the human provides only a topic with no angle, audience, or experience, the AI should ask exactly one clarifying question before proceeding. Do not ask multiple questions -- that stalls the creative phase.

Choose the single question that will most improve output quality. Priority order:

1. Angle: "What is your unique take or experience with this topic?" (Highest impact on output relevance) 2. Audience: "Who specifically is this for?" (Shapes tone and depth) 3. Problem: "What should the reader be able to do after reading this?" (Shapes content selection)

If the human declines to elaborate, proceed with the thin seed. Generating material from a thin seed is better than not generating at all. The Whirlybird phase can compensate.

---

The Seeding Conversation

A typical seeding exchange looks like this:

Human: "I want to write about developer productivity."

AI (one clarifying question): "What is your unique take or experience with developer productivity? For example, do you have data from your own team, a contrarian opinion, or a specific aspect you care about most?"

Human: "I've been tracking my team's output for 2 years and found that our most productive weeks were the ones with the most 'wasted' time -- long lunches, side projects, unstructured pairing."

AI: [Proceeds to generate Madman output seeded with this angle]

The AI does not ask follow-up questions after the seed is provided. The priority is to generate material. The human can redirect during the review phase.

---

Handoff Format: Madman to Whirlybird

When the Madman phase is complete, the output is delivered to the human. The human reviews it and marks:

• Which ideas resonate
• Which fragments have energy
• Which angles to pursue
• What is missing from their perspective
• What to discard entirely

This annotated Madman output becomes the input to the Whirlybird phase. The Whirlybird takes the human-curated raw material and begins grouping, clustering, and finding relationships.

Handoff Checklist

The Madman delivers:

1. Full raw material dump -- All dimensions populated, nothing filtered out 2. Marked high-energy fragments -- Passages flagged during generation as having momentum 3. Hook candidates -- 3-5 complete opening paragraphs 4. Dimension coverage note -- Which dimensions produced the most material and which were thin 5. Open questions -- Anything the AI noticed during generation that needs human input (factual verification, experience gaps, audience assumptions)

The human responds with:

1. Selections -- Which ideas, fragments, and hooks to carry forward 2. Additions -- Any new material sparked by reading the Madman output 3. Cuts -- What to discard (no explanation needed) 4. Direction -- Any refinement to the angle, audience, or thesis

This human-annotated package is what the Whirlybird skill receives as input.

---

Anti-Patterns

The AI Over-Seeds Itself

The AI generates material that sounds authoritative but does not reflect the human's actual experience or angle. The Madman must anchor to the human's seed, not replace it with generic expertise.

Fix: After generating, check each core idea against the human's seed. Flag any ideas that drifted from the provided angle.

The Human Over-Directs

The human provides such a detailed brief that there is nothing for the Madman to generate. The seed becomes a finished outline.

Fix: If the human's input already has structure, skip the Madman phase entirely and begin with the Architect. The Madman is for exploration, not execution of a pre-existing plan.

Analysis Paralysis on Seeding

The human spends too long perfecting their seed before the Madman can begin.

Fix: Remind the human that the seed does not need to be polished. A rough direction is sufficient. The Madman's job is to generate options, not to execute a precise brief.

Content Type Adaptation

This reference describes how the Madman adapts its generation process based on the content type. Different formats demand different emphasis during raw material generation.

---

Content Type Overview

---

Adaptation by Content Type

Technical Documentation

Priority dimensions:

• Real-world applications (every concept needs a concrete example)
• Audience questions (what will users get stuck on?)
• Counterarguments (what are the edge cases, limitations, gotchas?)

De-emphasized dimensions:

• Hooks and openers (technical docs do not need a catchy lead)
• Analogies and metaphors (precision over poetry)

Generation guidance:

• Enumerate every parameter, option, and configuration variant
• Brainstorm error states and failure modes
• List prerequisites the reader might not know they need
• Capture "what I wish someone had told me" moments from the human's seed

Nonfiction Article or Essay

Priority dimensions:

• Core arguments (the thesis must be sharp and debatable)
• Evidence and examples (every claim needs support)
• Hooks and openers (the lead determines whether anyone reads the rest)
• Raw fragments (essays live or die on voice -- capture passages with energy)

De-emphasized dimensions:

• Real-world applications (may be implicit in the argument)

Generation guidance:

• Generate at least 5 distinct thesis candidates before settling
• For each thesis, generate both supporting and opposing evidence
• Write hooks that create tension, curiosity, or surprise
• Capture any passage where the writer's voice comes through strongly

SEO Long-Form Content

Priority dimensions:

• Audience questions (these map directly to search queries and PAA boxes)
• Unique angles (differentiation from existing ranking content)
• Evidence and examples (E-E-A-T signals for search engines)

Additional dimensions (SEO-specific):

• Keyword variations and semantic clusters
• PAA-style questions at multiple knowledge levels
• Angle gaps relative to currently ranking content
• Featured snippet targets (definitions, lists, tables)

Generation guidance:

• Start with search intent: what does the reader actually want when they type this query?
• Map the topic against existing top-ranking content to find gaps
• Generate content at multiple depth levels (overview, intermediate, expert)
• Produce structured data candidates (comparison tables, numbered lists, definitions)

White Paper

Priority dimensions:

• Core arguments (white papers need a rigorous central thesis)
• Evidence and data (original data, research citations, expert quotes)
• Counterarguments (the piece must address objections to maintain authority)
• Connections to adjacent topics (white papers situate their argument in a broader context)

De-emphasized dimensions:

• Hooks and openers (white papers have more leeway for a methodical opening)

Generation guidance:

• Generate more evidence than any other content type -- aim for 5x
• Identify gaps where original research or data would strengthen the piece
• Brainstorm expert sources the human could reference or interview
• Map the argument's logical structure even at the Madman stage (without organizing it)
• Capture industry-specific terminology and framing

Case Study

Priority dimensions:

• Evidence and examples (metrics, timelines, concrete results)
• Real-world applications (the before/after transformation)
• Hooks and openers (the narrative hook that makes the reader care)
• Analogies and metaphors (making results relatable)

De-emphasized dimensions:

• Counterarguments (case studies are persuasive, not balanced)
• Connections to adjacent topics (keep the focus tight)

Generation guidance:

• Generate multiple story arc candidates (chronological, problem-solution, results-first)
• Brainstorm every quantifiable metric, even ones that seem minor
• Capture the emotional arc alongside the data arc
• List specific dialogue, decisions, or turning points from the human's experience
• Generate "what if they had not done this" counterfactuals for contrast

Tutorial or How-To

Priority dimensions:

• Real-world applications (every step needs a concrete context)
• Audience questions (what will the reader get confused about?)
• Counterarguments (what are the common mistakes and pitfalls?)

De-emphasized dimensions:

• Hooks and openers (tutorials can start with a brief problem statement)
• Analogies and metaphors (step-by-step clarity over creative comparison)

Generation guidance:

• Brainstorm multiple step sequences for the same outcome
• List every prerequisite and assumption
• Generate troubleshooting scenarios for each major step
• Capture "this is where people usually get stuck" moments
• Produce variations for different environments, skill levels, or tools

Blog Post

Priority dimensions:

• Hooks and openers (blog posts live or die on the first paragraph)
• Unique angles (a blog post must justify its existence in a crowded feed)
• Raw fragments (voice and personality are the differentiator)
• Audience questions (what makes the reader click and stay?)

De-emphasized dimensions:

• Connections to adjacent topics (keep the scope tight for a blog post)

Generation guidance:

• Generate at least 5 hooks -- blog posts need the strongest opening of any format
• Focus on shareability: what would make a reader send this to a colleague?
• Capture surprising data points, counterintuitive claims, and "I did not know that" moments
• Keep the raw material focused on a single strong idea rather than comprehensive coverage

---

Scaling Phase Depth to Request Size

Not every writing request justifies a full Madman phase. Scale the depth of generation to match the scope of the project.

Quick (under 800 words)

• Work through 4-5 dimensions instead of all 10
• Prioritize: core arguments, audience questions, hooks, raw fragments
• Generate 2-3 hook candidates instead of 5
• Target 2-3x raw material rather than 3-5x
• Time investment: single generation pass, no iteration

When to use: Social media threads, short blog posts, email newsletters, brief commentary.

Medium (800-2,500 words)

• Work through all 10 dimensions but accept thin coverage on some
• Generate 3-5 hook candidates
• Target 3x raw material
• Time investment: one generation pass with a brief human review before proceeding

When to use: Standard blog posts, short articles, documentation pages.

Long (2,500-5,000 words)

• Full dimension coverage with depth
• Generate 5 hook candidates with complete opening paragraphs
• Target 3-5x raw material
• Time investment: full generation pass, human review, one round of supplemental generation if dimensions are thin

When to use: Feature articles, long-form essays, detailed tutorials, comprehensive guides.

Major (over 5,000 words)

• Full dimension coverage with extensive depth
• Generate 5+ hook candidates
• Target 4-5x raw material
• Time investment: multiple generation passes, human review between each, dimension-specific deep dives as needed
• Consider splitting the Madman phase by section if the topic has natural divisions

When to use: White papers, research reports, book chapters, pillar content.

---

Choosing the Right Scale

If the human does not specify a target length, estimate based on content type:

1. Check the content type table at the top of this document for typical length ranges. 2. Default to the middle of the range for that content type. 3. Select the scaling tier that matches. 4. If uncertain, ask the human: "How long do you want the final piece to be?" This is a reasonable clarifying question that does not stall the creative phase.

Madman Generation Process

This reference details the step-by-step generation process for the Madman phase, including the dimension checklist, phase rules, and output format.

---

Phase 1 Rules

These rules are non-negotiable during Madman generation. Violating any of them collapses the creative space.

1. No self-censoring. Every idea gets written down, even the ones that feel stupid, obvious, or wrong. The filter comes later. 2. No organizing. Do not sort ideas into categories as you generate them. Dump everything in sequence. Structure is the Architect's job. 3. No polishing. Do not rewrite sentences for clarity or style. Rough is fine. Messy is fine. Energy matters more than elegance. 4. Overproduce deliberately. The goal is 3-5x the material the final piece will use. If the final piece is 2,000 words, generate 6,000-10,000 words of raw material. 5. Capture energy. When a passage comes out with momentum, mark it. These high-energy fragments often survive all the way to the final draft even when everything around them changes.

---

The Dimension Checklist

Work through each dimension in sequence. Not every dimension will yield material for every topic, but every dimension must be attempted.

1. Core Arguments

Generate the strongest thesis candidates for this piece. Write each as a single declarative sentence. Aim for 5-10 candidates that take different positions or emphasize different aspects of the topic.

Do not pick a winner. Generate all plausible positions.

2. Evidence, Data, and Examples

For each core argument, brainstorm supporting evidence:

• Published data, statistics, or research findings
• Named case studies or real-world examples
• Historical precedents or analogies
• Expert quotes or authoritative sources
• Personal experience from the human's seed

Flag any claims that need verification. Do not fabricate statistics.

3. Audience Questions

Write out every question the target reader is likely to ask:

• What they want to know before they start reading
• What objections they will raise mid-article
• What they will want to do after reading
• What assumptions they bring that the piece needs to address

4. Unique Angles

Push beyond the obvious framing:

• What would a contrarian say about this topic?
• What adjacent field has a useful analogy?
• What is everyone in this space getting wrong?
• What is the second-order consequence nobody talks about?
• What would make this piece different from the top 10 existing articles?

5. Hooks and Openers

Write 3-5 complete opening paragraphs, each using a different hook strategy:

Each hook should be a full paragraph, not a one-liner. Write enough to feel the energy of each approach.

6. Raw Fragments

Capture any passage that emerges with force during generation. These are not polished paragraphs -- they are raw bursts of writing where the voice, argument, or insight came through clearly.

Mark each fragment with a brief note about where it might fit (opening, middle, closing, sidebar) but do not organize them further.

7. Counterarguments

List every reasonable objection to the core arguments:

• What a skeptic would say
• What evidence cuts against the thesis
• What edge cases break the argument
• What the strongest opposing position looks like

These counterarguments become the raw material for the piece's credibility and nuance.

8. Analogies and Metaphors

Generate concrete comparisons:

• What is this topic like in another domain?
• What physical metaphor captures the core dynamic?
• What everyday experience mirrors this abstract concept?

Bad analogies are fine at this stage. Generate volume, not precision.

9. Real-World Applications

For each core argument, describe a concrete scenario:

• How does this apply to the reader's daily work?
• What would change if the reader adopted this idea?
• What does success look like in practice?
• What does failure look like?

10. Connections to Adjacent Topics

Map the edges of the topic:

• What related subjects does this touch?
• What larger trend does this fit into?
• What prerequisite knowledge does the reader need?
• What follow-up topics does this piece naturally lead to?

---

SEO Content: Additional Dimensions

When the content type is SEO-focused, add these dimensions after completing the standard checklist:

Keyword Variations

• Primary keyword and close synonyms
• Long-tail variations (3-5 word phrases)
• Related terms that indicate the same search intent
• Semantic cluster words that search engines associate with the topic

PAA-Style Questions

Generate 10-15 "People Also Ask" style questions:

• Questions that start with "What," "How," "Why," "When," "Is," "Can," "Does"
• Questions at different knowledge levels (beginner, intermediate, advanced)
• Questions that reveal adjacent intent

Angle Gaps

Review the top-ranking content for the target keyword and identify:

• What perspectives are missing from existing content?
• What depth is missing?
• What formats are missing (tables, comparisons, step-by-step)?
• What audience segments are underserved?

Featured Snippet Targets

For each major section of the eventual piece, draft:

• A concise definition (40-60 words) that could win a definition snippet
• A numbered list (3-8 items) that could win a list snippet
• A comparison table that could win a table snippet

---

Raw Ideas Output Format

Use this template for all Madman output. Every section must be populated. If a dimension yielded nothing, write "None generated -- revisit after human review" rather than omitting it.

# Madman Output: [Topic]

**Seed:** [Summary of human-provided seed]
**Audience:** [Target reader]
**Angle:** [Core angle or thesis direction]
**Target length:** [Expected final word count]
**Content type:** [Article, tutorial, white paper, etc.]

## Core Ideas
1. [Thesis candidate]
2. [Thesis candidate]
...

## Evidence & Examples
- [Data point or example, with source if known]
- [Case study or anecdote]
...

## Audience Questions
- [Question the reader brings]
- [Objection the reader will raise]
...

## Unique Angles
- [Contrarian take or surprising framing]
- [Connection to unexpected domain]
...

## Hooks & Openers

### Hook 1: [Strategy name]
[Full opening paragraph]

### Hook 2: [Strategy name]
[Full opening paragraph]

### Hook 3: [Strategy name]
[Full opening paragraph]

## Raw Fragments
> [Fragment with energy] -- [note on potential placement]

> [Fragment with energy] -- [note on potential placement]

## Counterarguments
- [Objection and its strongest form]
...

## Analogies & Metaphors
- [Comparison: "This is like..."]
...

## Applications
- [Concrete scenario or use case]
...

## Connections
- [Adjacent topic and how it links]
...

---

Completion Check

Before delivering Madman output, verify:

• [ ] All 10 dimensions attempted (or explicitly marked as not applicable)
• [ ] At least 3 hook candidates written as full paragraphs
• [ ] Raw material volume is 3-5x the target final length
• [ ] No organizing or structuring has crept in
• [ ] High-energy fragments are marked
• [ ] Human's seed angle is reflected in the material, not overwritten
• [ ] Counterarguments are included, not just supporting evidence