Virtual Think Tank
Use this skill when the user needs a disciplined decision workshop rather than a direct recommendation.
Core Rules
- Do not make the final decision for the user.
- Ask only the minimum missing questions needed to frame the decision well.
- Separate facts, assumptions, and role-play clearly.
- State uncertainties and what evidence would change the conclusion.
- Prefer role-based voices over named people by default.
- Use named people only when the user explicitly asks for them, or when they are clearly relevant and you can give a confidence level plus a verification path.
- Prefer stakeholder voices when the decision is mainly organisational, commercial, or process-driven.
- End by forcing an explicit human choice, or a time-boxed defer decision.
Input Contract
Parse $ARGUMENTS first.
If the user pasted a structured brief, use it. Otherwise ask for the missing fields from the briefing template.
Resolve these fields:
mode: guided or full, default guided
panel_style: archetype, stakeholder, or named, default archetype
decision_subject
current_state
required_state
problem_statement
constraints
non_goals
decision_horizon
- optional
decision_evidence, such as incident counts, timing data, cost bounds, revenue at risk, or adoption metrics
- optional
context
- optional
sources
- optional
outside_lens
- optional
named_moderator, when panel_style: named
- optional
named_voices, a comma-separated list of 2 to 4 public voices when panel_style: named
- optional
named_outside_voice, when panel_style: named
Interpretation rules:
- If
panel_style is omitted and any named_* field is present, resolve panel_style to named.
- If the user explicitly asks for named voices, or provides any
named_* field, honour that request even for organisational or process decisions.
- If the user asks for
panel_style: named without supplying names, propose a balanced panel with substantive disagreement.
- If 3 or more critical fields are missing, stop and ask focused questions.
- If 1 or 2 non-critical fields are missing, make explicit assumptions.
- If the user asks for
full mode but the brief is too thin, refuse to fabricate and switch to clarification.
- If the user provides earlier phase outputs and asks to continue, resume from the next missing phase.
- In guided mode, when
decision_subject, problem_statement, and constraints are already present, ask no more than 3 compound clarification questions before stopping.
Phase Model
The workflow runs in five operational phases.
Phase 1, Frame the Decision
Produce:
- Clarifying questions, only if still needed
- Decision question
- Success criteria
- Failure modes
Do not ask broad questionnaires. Ask the fewest discriminating questions that sharpen the decision.
Prefer quantitative missing questions when the user has not provided any evidence, for example timings, incident counts, cost ceilings, revenue at risk, adoption figures, or operational load.
Phase 2, Build a Neutral Baseline
Explain the decision space in neutral terms.
Output shape:
- Definitions
- Typical benefits
- Typical costs and risks
- When it is a bad idea
- Open questions
Phase 3, Build the Panel
Choose the panel according to panel_style.
If the user supplied named participants, use those as anchors unless the panel is one-sided enough that it cannot produce a meaningful debate. In that case, keep the requested names where possible and add one contrasting voice rather than silently replacing the panel.
For archetype:
- Moderator
- Voice A
- Voice B
- Middle voice
- Optional outside lens
For stakeholder:
- Product or business owner
- Engineering or delivery lead
- Platform or operations
- Security, privacy, or risk
- Finance or commercial
- Support or live operations
- Neutral moderator
For named:
- Use real public voices only if clearly useful
- Prefer a balanced panel: one moderator, at least two substantively conflicting voices, one pragmatic or middle voice, and an optional outside lens
- If the user supplied names, honour them and rebalance only when the panel lacks genuine disagreement
- Avoid panels made of near-identical advocates or adjacent commentators who would say the same thing
- Mark each suggestion with confidence:
high, medium, or low
- Add a verification path for each named person
For every voice, provide:
- stance
- why this voice is useful
- likely bias
- what they would challenge
Phase 4, Run the Debate
Run one structured debate round.
Output shape:
- Opening statements
- Cross-examination
- Moderator synthesis
- Options on the table, usually
A, B, and optional C
- Decisive questions the user must answer
Rules:
- The voices must disagree in substance.
- They must challenge hidden costs, assumptions, and failure cases.
- They may propose a compromise if it is materially different from
A and B.
- If one option starts to look obviously dominant, force the strongest counterargument against it before the moderator synthesis.
- Avoid vague platitudes and generic balance.
Phase 5, Force the Human Decision
Summarise the options and require the user to choose A, B, C, or defer with a date.
Ask the user to state:
- why they chose it
- what they are giving up
- what would make them reverse the decision
- the review date
Then draft a short first-pass decision record using the repository ADR template and the virtual think tank ADR profile.
- Use the ADR template as the canonical structure.
- Keep the first draft proportionate and concise.
- Default
Status to Proposed unless the user says otherwise.
- Use
TBD for missing metadata rather than inventing it.
If the user has not chosen yet, leave the decision as pending and stop after the questions.
Mode Behaviour
Guided Mode
- Default mode
- Complete one phase at a time
- Stop after each phase and ask whether to continue, correct, or switch panel style
Full Mode
- Run phases 1 to 4 in one response only if the brief is complete enough
- Stop at phase 5 and wait for the user to choose
Output Discipline
- Keep each phase clearly headed
- Mark assumptions as
Assumption:
- Mark gaps as
Unknown:
- Mark decisive evidence as
Would change the conclusion if:
- Keep the writing direct, concrete, and short
- Do not present role-play as evidence
Recommended Invocation
Ask the user to paste the briefing template when they need a clean starting point.
Prefer panel_style: stakeholder for people, governance, and process decisions.
Prefer panel_style: archetype for technical and architectural decisions.
Use panel_style: named only when the value of real public viewpoints is worth the verification overhead.
If you want public figures on an organisational or process decision, set panel_style: named and fill one or more named_* fields. That explicit request overrides the default stakeholder recommendation.
Examples
1---2name: virtual-think-tank3description: Run a structured multi-step think tank for engineering, architecture, product, process, policy, vendor, or organisational decisions. Use when the user needs conflicting viewpoints, explicit trade-offs, and a forced human decision rather than a one-shot answer.4license: MIT5---67# Virtual Think Tank89Use this skill when the user needs a disciplined decision workshop rather than a direct recommendation.1011## Core Rules12131. Do not make the final decision for the user.142. Ask only the minimum missing questions needed to frame the decision well.153. Separate facts, assumptions, and role-play clearly.164. State uncertainties and what evidence would change the conclusion.175. Prefer role-based voices over named people by default.186. Use named people only when the user explicitly asks for them, or when they are clearly relevant and you can give a confidence level plus a verification path.197. Prefer stakeholder voices when the decision is mainly organisational, commercial, or process-driven.208. End by forcing an explicit human choice, or a time-boxed defer decision.2122## Input Contract2324Parse `$ARGUMENTS` first.2526If the user pasted a structured brief, use it. Otherwise ask for the missing fields from the [briefing template](./assets/briefing-template.md).2728Resolve these fields:2930- `mode`: `guided` or `full`, default `guided`31- `panel_style`: `archetype`, `stakeholder`, or `named`, default `archetype`32- `decision_subject`33- `current_state`34- `required_state`35- `problem_statement`36- `constraints`37- `non_goals`38- `decision_horizon`39- optional `decision_evidence`, such as incident counts, timing data, cost bounds, revenue at risk, or adoption metrics40- optional `context`41- optional `sources`42- optional `outside_lens`43- optional `named_moderator`, when `panel_style: named`44- optional `named_voices`, a comma-separated list of 2 to 4 public voices when `panel_style: named`45- optional `named_outside_voice`, when `panel_style: named`4647Interpretation rules:4849- If `panel_style` is omitted and any `named_*` field is present, resolve `panel_style` to `named`.50- If the user explicitly asks for named voices, or provides any `named_*` field, honour that request even for organisational or process decisions.51- If the user asks for `panel_style: named` without supplying names, propose a balanced panel with substantive disagreement.52- If 3 or more critical fields are missing, stop and ask focused questions.53- If 1 or 2 non-critical fields are missing, make explicit assumptions.54- If the user asks for `full` mode but the brief is too thin, refuse to fabricate and switch to clarification.55- If the user provides earlier phase outputs and asks to continue, resume from the next missing phase.56- In guided mode, when `decision_subject`, `problem_statement`, and `constraints` are already present, ask no more than 3 compound clarification questions before stopping.5758## Phase Model5960The workflow runs in five operational phases.6162### Phase 1, Frame the Decision6364Produce:6566- Clarifying questions, only if still needed67- Decision question68- Success criteria69- Failure modes7071Do not ask broad questionnaires. Ask the fewest discriminating questions that sharpen the decision.7273Prefer quantitative missing questions when the user has not provided any evidence, for example timings, incident counts, cost ceilings, revenue at risk, adoption figures, or operational load.7475### Phase 2, Build a Neutral Baseline7677Explain the decision space in neutral terms.7879Output shape:8081- Definitions82- Typical benefits83- Typical costs and risks84- When it is a bad idea85- Open questions8687### Phase 3, Build the Panel8889Choose the panel according to `panel_style`.9091If the user supplied named participants, use those as anchors unless the panel is one-sided enough that it cannot produce a meaningful debate. In that case, keep the requested names where possible and add one contrasting voice rather than silently replacing the panel.9293For `archetype`:9495- Moderator96- Voice A97- Voice B98- Middle voice99- Optional outside lens100101For `stakeholder`:102103- Product or business owner104- Engineering or delivery lead105- Platform or operations106- Security, privacy, or risk107- Finance or commercial108- Support or live operations109- Neutral moderator110111For `named`:112113- Use real public voices only if clearly useful114- Prefer a balanced panel: one moderator, at least two substantively conflicting voices, one pragmatic or middle voice, and an optional outside lens115- If the user supplied names, honour them and rebalance only when the panel lacks genuine disagreement116- Avoid panels made of near-identical advocates or adjacent commentators who would say the same thing117- Mark each suggestion with confidence: `high`, `medium`, or `low`118- Add a verification path for each named person119120For every voice, provide:121122- stance123- why this voice is useful124- likely bias125- what they would challenge126127### Phase 4, Run the Debate128129Run one structured debate round.130131Output shape:1321331. Opening statements1342. Cross-examination1353. Moderator synthesis1364. Options on the table, usually `A`, `B`, and optional `C`1375. Decisive questions the user must answer138139Rules:140141- The voices must disagree in substance.142- They must challenge hidden costs, assumptions, and failure cases.143- They may propose a compromise if it is materially different from `A` and `B`.144- If one option starts to look obviously dominant, force the strongest counterargument against it before the moderator synthesis.145- Avoid vague platitudes and generic balance.146147### Phase 5, Force the Human Decision148149Summarise the options and require the user to choose `A`, `B`, `C`, or `defer` with a date.150151Ask the user to state:1521531. why they chose it1542. what they are giving up1553. what would make them reverse the decision1564. the review date157158Then draft a short first-pass decision record using the repository [ADR template](../../../docs/adr/ADR-nnn_Any_Decision_Record_Template.md) and the [virtual think tank ADR profile](./assets/decision-record-template.md).159160- Use the ADR template as the canonical structure.161- Keep the first draft proportionate and concise.162- Default `Status` to `Proposed` unless the user says otherwise.163- Use `TBD` for missing metadata rather than inventing it.164165If the user has not chosen yet, leave the decision as pending and stop after the questions.166167## Mode Behaviour168169### Guided Mode170171- Default mode172- Complete one phase at a time173- Stop after each phase and ask whether to continue, correct, or switch panel style174175### Full Mode176177- Run phases 1 to 4 in one response only if the brief is complete enough178- Stop at phase 5 and wait for the user to choose179180## Output Discipline181182- Keep each phase clearly headed183- Mark assumptions as `Assumption:`184- Mark gaps as `Unknown:`185- Mark decisive evidence as `Would change the conclusion if:`186- Keep the writing direct, concrete, and short187- Do not present role-play as evidence188189## Recommended Invocation190191Ask the user to paste the [briefing template](./assets/briefing-template.md) when they need a clean starting point.192193Prefer `panel_style: stakeholder` for people, governance, and process decisions.194195Prefer `panel_style: archetype` for technical and architectural decisions.196197Use `panel_style: named` only when the value of real public viewpoints is worth the verification overhead.198199If you want public figures on an organisational or process decision, set `panel_style: named` and fill one or more `named_*` fields. That explicit request overrides the default stakeholder recommendation.200201## Examples202203- [Examples overview](./examples/example-01-examples-overview.md)204- [Microservices decision, named figures, full mode, VS Code Copilot Chat](./examples/example-02-microservices-named-figures-full-vscode-chat.md)205- [Team Topologies decision, named figures, interactive mode, GitHub Copilot CLI](./examples/example-03-team-topologies-named-figures-interactive-copilot-cli.md)206- [Product mindset transition, stakeholder voices, guided mode, VS Code Copilot Chat](./examples/example-04-product-mindset-stakeholder-voices-guided-vscode-chat.md)