thinking-personas

Thinking Personas

Use this skill to instantiate a landscape of minds.

This skill has two jobs only:

cast distinct personas with materially different worldviews
optionally let those personas speak independently about the same target

This is not a parliament, debate stage, or forced-consensus machine. Its value comes from worldview diversity, not from fake closure.

If the user wants synthesis, verdicts, or decision pressure after the cast is created, treat that as a separate step rather than the core job of this skill.

Core Stance

Optimize for worldview distance, not job-title variety.
Build personas that differ in values, causal models, risk tolerance, time horizon, or evidence standards.
Prefer stable stance cards over disposable one-line characters.
Let each persona speak for itself; do not collapse them into your own voice too early.
Keep the target shared and the lenses different.
Treat emotional register as load-bearing when it changes how a persona interprets or attacks the target.
Do not force agreement, voting, or convergence unless the user explicitly asks for that elsewhere.
Treat the personas as useful simulations, not as truly independent minds.
When the target is still messy, compress it lightly before casting instead of multiplying confusion.

When To Use

Use this skill when the user:

wants several distinct personas, voices, or worldviews generated for a topic
says to roleplay 3, 6, or 10 different people reacting to something
wants to see how different kinds of minds interpret an idea, plan, text, or dilemma
wants a ready-made pack such as political, business, psychology, scientific, or emotionally varied personas
wants a reusable cast for later work, comparison, prompting, or downstream application
wants persona-specific opinions without debate or verdict mechanics

Do not use this skill when:

the user mainly wants one best answer or recommendation
the user needs strong challenge, adversarial critique, or red-teaming more than diversity of viewpoint
the task is straightforward implementation, lookup, or execution work
the input is so unshaped that persona output would just magnify confusion instead of clarifying it

Operating Rule

First determine which of the two responsibilities is needed:

cast - generate personas only
speak - generate personas, then have each one give an independent take

If the user asks for both, do both in that order.

If the user does not specify the count, ask how many personas they want. Useful defaults are 3, 6, and 8. If they do not care, default to 6. Hard warning: more than 10 usually turns into repetition, blur, or fake diversity.

Near the start, infer or capture:

the target: idea, decision, text, branch set, identity, problem, or artifact
whether the user wants a preset pack or a fresh cast
whether the user wants broad worldview spread or a more domain-specific cast
how many personas are wanted
whether emotional register should stay restrained or become vivid, passionate, unstable, or extreme
whether the result is for direct reading or downstream reuse

Use fewer when the target is narrow and depth matters more than spread. Use more only if the user asks or if the space truly needs it. More personas often blur together; do not inflate the cast for show.

Persona Count Guide

Use this rule instead of pure vibes:

3 personas - narrow target, fast contrast, or one-axis comparison
4-5 personas - one domain with several meaningful schools or camps
6 personas - default for broad worldview spread without too much blur
7-8 personas - complex target with multiple real axes of disagreement
9-10 personas - only when the user explicitly wants a crowd or a preset expansion and the cast can still stay distinct

If you cannot make each additional persona meaningfully different, stop adding personas. More bodies do not mean more insight. Above 10, warn that quality will likely degrade into repetitive or fake personas.

Persona Differentiation Rule

Every persona must differ from the others in at least one load-bearing way:

values
causal model
risk posture
time horizon
evidence standard
social orientation
abstraction style

Do not generate near-duplicates with new costumes.

Bad differentiation:

three experts with similar elite-institution analytical voices
several personas who all optimize for efficiency with minor tone changes
career labels that imply diversity but produce the same conclusions

Good differentiation:

order and control vs dignity and rights
incentives and games vs meaning and narrative
systems ecology vs immediate pragmatism
institutional trust vs anti-fragile skepticism

Persona Card

When casting personas, each card should usually include:

Name: a memorable handle
Archetype: a short role or orientation label
Worldview: the core frame through which they interpret reality
Optimizes for: what they protect or pursue first
Distrusts: what they are quick to challenge
Notices first: the feature they reliably see before others do
Blind spot: what they underweight or miss
Emotional register: cold, warm, urgent, bitter, ecstatic, wounded, detached, volatile, or another load-bearing tone
Voice: how they tend to sound or reason
Default move: what they tend to recommend or do

Keep cards compact but reusable. The point is to create personas that downstream skills can activate without re-inventing them.

Emotional register is not decorative. Use it when tone changes interpretation, not just style. A cold analyst, fervent activist, exhausted cynic, or unhinged prophet may share some beliefs while still producing very different reactions.

Preset Packs

The skill may use preset packs when the user asks for them. Preset packs are shortcuts, not cages; adapt them to the target while preserving internal contrast.

Common packs:

political - ideological and governance-oriented viewpoints around power, order, liberty, justice, institutions, and change
business-stakeholders - founder, operator, investor, customer advocate, regulator-minded voice, labor or execution voice
psychology-schools - behavioral, psychodynamic, cognitive, humanistic, trauma-aware, systems or family-systems lenses
scientific-community - theorist, experimentalist, statistician, skeptic, field practitioner, interdisciplinary synthesizer
emotional-spectrum - same target viewed through different emotional temperatures such as cold, compassionate, anxious, furious, ecstatic, cynical, or feral

Preset rule:

if the user names a preset, start there instead of rebuilding the casting logic from scratch
if the user names both a preset and a count, honor both unless the count destroys distinctness
if the user names a preset and wants stronger weirdness, intensity, or instability, vary emotional register inside that pack deliberately

If no preset is specified, cast from first principles.

Mode Guide

Choose the dominant mode based on what the user needs.

Cast Mode

Use when the user wants the personas themselves.

generate a distinct cast
expose each worldview clearly
optimize for reusability and contrast
use a preset pack immediately when the user names one
avoid unnecessary opinions unless lightly helpful

Good for:

downstream prompting
perspective libraries
worldview exploration
later branch evaluation

Speak Mode

Use when the user wants the cast plus each persona's own take.

cast first, then activate
each persona responds independently to the same target
preserve preset logic if a preset pack was requested
do not have personas argue with each other unless explicitly asked
do not synthesize beyond a minimal structural wrap unless explicitly asked

Each speaking persona should usually provide:

a short worldview framing
what it notices first about the target
its opinion, read, or instinctive move
the main concern, tension, or question from that worldview

Default Workflow

1. Anchor the target

Identify what the personas are being built for.

idea
plan
draft
decision
problem statement
branch set
source text

If the target is sprawling, compress it into a blunt one- or two-sentence anchor before casting. Do not spend the whole skill on clarification unless that is what the user actually needs.

2. Choose the cast logic

Decide whether the cast should be:

preset-based
broad worldview spread
domain-focused but internally diverse
user-specified types
reusable archetypes for downstream work

Prefer worldview diversity over credential theater.

3. Cast the personas

Create the personas as stable cards.

Check for blur:

are any two personas functionally the same?
are the differences load-bearing or cosmetic?
would these personas reliably disagree about important things?
would they notice different features of the same target?
does emotional register create a meaningfully different reaction where needed?

If the answer is no, rebuild the cast.

4. Activate only if asked

If the user wants speak, have each persona respond to the same anchored target.

Important rules:

each persona answers independently
no voting
no ranking by default
no forced consensus
no giant summarizer voice taking over the result

5. Preserve downstream usability

If the cast is likely to be reused:

keep persona cards stable
keep names memorable
keep worldview statements compact and explicit
avoid one-off references that make the cast non-portable

Response Shape

When responding with this skill, prefer one of these shapes.

Cast Output

Target: one or two sentences
Cast logic: why these personas were chosen
Personas: compact cards for each persona

Speak Output

Target: one or two sentences
Cast: compact cards or short intros
Voices: one section per persona with its independent take

Unless the user asks for it, do not end with an executive summary that flattens the differences.

Composability

This skill is intentionally self-contained.

do the cast cleanly enough that the personas can be reused later in any direction the user wants
keep the target explicit so later steps can activate the same cast without rebuilding it
avoid assuming what process comes next; that belongs to the user, not to this skill
keep preset and emotional-register choices visible so they can be reused later without guesswork
treat casting and speaking as portable outputs rather than as instructions for a larger pipeline

Boundaries

Do not pretend the personas are truly independent agents.
Do not fake diversity by swapping labels on the same worldview.
Do not add debate, voting, or consensus theater unless the user explicitly requests that kind of process.
Do not over-summarize the voices into your own preferred answer.
Do not let tone differentiation replace actual worldview differentiation.
Do not use a preset pack mechanically if the target clearly needs adaptation.

Failure Modes To Avoid

six personas who all sound like the same assistant in different hats
professions instead of worldviews
broad cards with no stable stance or usable downstream identity
speaking outputs that blur into one median answer
unnecessary synthesis that destroys the point of the cast
too many personas for the amount of real differentiation available
pseudo-diversity where the cast differs in biography but not in reasoning
emotional variety that is theatrical but not cognitively meaningful

Compact Prompts

Use these when you need a fast entry point.

Cast

Generate 6 distinct personas for thinking about this.
Cast a council of different worldviews for this target.
Give me reusable personas, not conclusions.
Use the political preset with 6 personas.
Use a psychology-schools preset and keep the cast reusable.

Speak

Generate 6 personas, then let each speak independently.
Roleplay several different minds reacting to this.
Have each persona give its own opinion without debate or synthesis.
Use the emotional-spectrum preset and let each persona react to this.
Give me 8 business-stakeholder personas and let them speak.

Default Deliverables

When the output is mature enough, end with one of these compact forms.

Cast deliverable

Target: one sentence
Cast logic: one or two sentences
Personas: 3 to 8 compact cards

Speak deliverable

Target: one sentence
Cast: 3 to 8 compact cards or labels
Voices: one short independent take per persona

Keep the result legible, reusable, and different. The win condition is not agreement. The win condition is a believable spread of minds.