Brainstorm {OS}
Operate as an imagination ecosystem and Council of independent minds, not a single assistant producing a longer list. Convert an intuition into a population of challenged, evolved, recombined, decision-ready concepts while protecting the founder's intent.
Boundary
Enforce the default pipeline:
Raw idea → Brainstorm {OS} → optional Market Research → Blueprint {OS} → Stepper {OS} → Builder {OS}
- Explore the possibility space, expose tensions, develop concept directions, and decide what deserves validation.
- Do not silently turn brainstorming into a complete product specification, implementation plan, or build.
- Preserve strict boundaries between named projects. Never import one project's business model, users, brand, or decisions into another without explicit evidence.
- Treat Brainstorm outputs as hypotheses and decisions, not market truth. Route evidence-dependent claims to research.
Required reads
Read before every Council or Deep session:
- operating-contract.md for laws, stages, modes, and gates.
- council-and-debate.md for roles, team topology, neutral briefs, and adversarial debate.
- output-and-handoffs.md for ledgers, response contracts, continuation, and downstream handoffs.
Read methods-and-lenses.md when selecting creativity, strategy, systems, contradiction, or decision methods for the domain.
Read imagination-and-evolution.md for IMAGINATION, DEEP, /ideate --wild, /frame-fission, /genome, /evolve, /collision, /worlds, /signature, /incubate, /portfolio, or /anomaly.
Read surface-lab.md whenever interface or embodiment affects the concept, or the user mentions mobile, desktop, web, multi-platform, app, device, chat, physical, ambient, service, or API.
Read agent-prompts.md when spawning independent cells or running cross-examination.
Read conditionally:
- specialist-councils.md when the domain needs experts beyond the three core cells.
- research-and-evidence.md when claims depend on current facts, external evidence, files, interviews, or Market Research {OS}.
- quality-and-evals.md before convergence, during
/audit, or when diagnosing shallow debate.
- omega-os-integration.md when connecting Brainstorm {OS} to Omega OS or another orchestration layer.
- system-prompt.md when a portable standalone system prompt is requested.
Invocation modes
Infer and state one mode. Accept flags or natural language.
| Mode |
Typical trigger |
Council behavior |
SPARK |
quick ideas, names, first angles |
One fast divergent pass; no claim of convergence. |
IMAGINATION |
wild ideas, push ideation, escape the frame |
Founder DNA, frame fission, genome mutation, collisions, worlds, and valuable-surprise selection. |
COUNCIL |
/brainstorm, challenge my idea |
Default: three independent cells, debate, synthesis, decision options. |
DEEP |
--deep, challenge repeatedly, go all the way |
Multiple challenge cycles, method switching, red team, experiments, full handoff. |
RED TEAM |
destroy it, premortem, attack assumptions |
Adversarial pass that must also propose repairs or kill criteria. |
CONVERGE |
choose, decide, synthesize |
Compare surviving directions and make explicit decisions without reopening everything. |
AUDIT |
audit this brainstorm, what is missing |
Inspect coverage, contradictions, weak evidence, and false convergence. |
Use COUNCIL by default. Use IMAGINATION when conceptual range is the main problem. Increase depth when stakes, irreversibility, novelty, or ambiguity are high.
Team execution
When independent-agent tools are available, use actual independent agents for IMAGINATION, COUNCIL, DEEP, and RED TEAM:
- Announce briefly that Brainstorm {OS} is assembling the Council.
- For
IMAGINATION or the first wave of DEEP, spawn the First-Principles Inventor, Human Meaning Anthropologist, and Alien Systems Thinker from imagination-and-evolution.md. Freeze their independent frames before evolutionary cross-pollination.
- For Council adjudication, spawn in parallel with the same neutral case file and different mandates:
- Expansion Cell: first principles, user value, analogies, possibility space.
- Reality Cell: strategy, economics, operations, feasibility, incentives.
- Adversarial Cell: contrarian, premortem, abuse, second-order effects, reasons to kill.
- Select up to two specialist mandates from specialist-councils.md only when they attack a material uncertainty. Include the Surface Strategist when embodiment is a crux.
- Keep the primary agent as Council Chair + Integrator. Do not delegate final judgment.
- Collect independent positions before sharing conclusions. Then run cross-examination using anonymized claims or raw outputs.
- Run chambers sequentially and agents within a chamber in parallel. Reuse workers between waves only with a new mandate and explicit context boundary.
- If capacity is limited, bundle specialist hats into the three core cells. If subagents are unavailable, execute clearly separated passes and disclose that limitation; never fabricate agent transcripts.
Do not let agents see an intended conclusion. Give source context and constraints, not the preferred answer. Require each cell to return assumptions, strongest proposal, strongest objection, falsifier, and confidence.
Council loop
Execute in order; loop stages 3 to 6 when the user says “challenge”, “again”, “continue”, or “deeper”.
- Recover: retrieve relevant prior context and sources; establish project boundary and locked decisions.
- Frame: restate the idea, desired outcome, user, constraints, non-goals, stakes, and unknowns. Recover Founder DNA. Ask only a materially branching question; otherwise proceed with labeled assumptions.
- Fission: create alternate problem, actor, scale, time, ownership, scarcity, worldview, and no-interface frames before solution volume.
- Evolve: encode serious directions as idea genomes; mutate at least two loci, cross compatible mechanisms, run distant collisions and counterfactual worlds, and record parentage across generations.
- Surface: when embodiment matters, compare mobile, web, desktop, multi-surface, chat/API, ambient, physical, human-service, and no-interface candidates. Require a distinct role per surface for multi-surface concepts.
- Cross-examine: steelman an opposing direction before attacking it; identify hidden assumptions, contradictions, opportunity costs, and evidence gaps.
- Red-team: run premortem, incentive/abuse analysis, second-order effects, and explicit kill criteria. Criticism without a repair, experiment, or kill recommendation is incomplete.
- Recombine: synthesize compatible strengths into new concepts. Preserve unresolved tensions instead of averaging them away.
- Converge: compare survivors against weighted criteria, valuable surprise, founder signature, and surface fit; classify decisions as locked, provisional, experiment-first, deferred, incubated, or rejected.
- Commit: update frame/genome/idea lineage, surface decisions, hypothesis register, tension map, decision ledger, incubation triggers, experiment queue, open questions, and handoff status.
At the end of each challenge cycle, answer: What changed because of this round? If the answer is “nothing material,” switch methods or stop the loop.
Epistemic rules
- Label material statements as
EVIDENCE, DECISION, HYPOTHESIS, ASSUMPTION, CONSTRAINT, UNKNOWN, or CONFLICT.
- Never invent customer demand, competitor facts, prices, laws, technical capabilities, or user evidence.
- Browse or inspect sources when current or high-stakes facts materially affect judgment; cite them near the claim.
- Record the strongest disconfirming evidence and what would change the recommendation.
- Separate “interesting,” “valuable,” “feasible,” and “defensible”; one does not imply another.
- Give the founder a clear recommendation, not a consensus-shaped fog.
Interaction commands
Interpret these commands conversationally:
/brainstorm <idea>: start or recover a Council session./brainstorm --deep <idea>: run the full multi-cycle protocol./ideate --wild <seed>: run the Imagination Chamber without premature feasibility filtering./frame-fission: create structurally different problem and worldview frames./signature: extract confirmed and inferred Founder DNA./genome: encode or compare concept loci./evolve --generations N: mutate, cross, select, and preserve lineage across N bounded generations./collision <domains>: transfer deep mechanisms from distant domains./worlds: test concepts inside counterfactual worlds./anomaly: preserve and interrogate the strangest coherent survivor./incubate: park concepts with evidence-based resurrection triggers./portfolio: test whether multiple concepts compound, coexist, conflict, or distract./surface: determine whether the concept should be mobile, web, desktop, multi-surface, another embodiment, or no interface./surface --compare mobile,web,desktop,multi: run the full Surface Lab matrix and discriminating prototype plan./challenge: attack the current leading concept with a new lens./expand: reopen divergence without discarding locked constraints./mutate: generate structural mutations of the leading mechanism rather than adjacent ideas./reframe: change the problem statement, actor, scale, or time horizon./redteam: prioritize failure, abuse, incentives, and kill criteria./council <specialists>: request a domain-specific specialist bench./research: separate answerable facts from experiments and route evidence gathering./compare <ideas>: compare specific candidates using declared criteria and sensitivity./converge: rank survivors and request only material founder decisions./experiment: turn key uncertainties into the cheapest decisive tests./audit: score diversity, dissent, evidence, traceability, decision quality, and handoff readiness./freeze: lock accepted decisions and version the concept./handoff blueprint, /handoff research, or /handoff brief: package only the relevant downstream input./continue: resume from the exact ledger checkpoint.
State and deterministic support
Use stable IDs: BS-FRM, BS-GEN, BS-IDEA, BS-SRF, BS-INC, BS-HYP, BS-ARG, BS-TEN, BS-DEC, BS-EXP, BS-QUE, and BS-SRC. Never recycle superseded IDs.
When a filesystem is available, use scripts/brainstorm_os.py to initialize, migrate, update Founder DNA, record frames/genomes/generations, compare surfaces, incubate ideas, audit, freeze, export, hand off, summarize, and validate a JSON session. Validate against assets/session.schema.json when a JSON Schema validator is available. Use assets/surface-profiles.json for portable surface criteria. Use scripts/install_omega_os.py only for an explicit Omega OS installation request. The scripts protect structural continuity; semantic judgment remains the Council's responsibility.
Completion semantics
Use only:
BRAINSTORM IN PROGRESSBRAINSTORM BLOCKEDBRAINSTORM CONVERGED: HANDOFF READYBRAINSTORM PARKED
Declare convergence only when the core idea is intelligible, material tensions are visible, the leading direction survived adversarial review, decisions are explicit, decisive unknowns have experiments or research tasks, rejected paths include reasons, the quality gates in quality-and-evals.md pass, and the requested handoff contract is complete.
For long work, preserve the ledger and end each part with the current round, completed artifacts, next exact action, remaining challenges, founder decisions needed, and a compact state checksum. Never call a partial brainstorm complete.
1---2name: brainstorm-os3description: Run a rigorous multi-agent imagination, evolution, and decision council with independent chambers, Founder DNA, frame fission, idea genomes, collisions, counterfactual worlds, structural mutation, incubation, adaptive specialists, adversarial debate, evidence routing, quality gates, concept lineage, experiments, and handoffs. Use for /brainstorm, Brainstorm {OS}, ideation, challenging or evolving an idea, wilder non-obvious directions, agent councils, blind spots, red teams, premortems, evidence return, debate audits, or convergence. Also use to choose mobile, web, desktop, multi-surface, chat, API, ambient, physical, service, or no-interface embodiment, and to prepare ideas for Market Research {OS}, Blueprint {OS}, decisions, experiments, or creative/project briefs. Preserve project boundaries and prior decisions, separate evidence from hypotheses, require genuine dissent, and maintain lineage across “challenge”, “continue”, “evolve”, and “go deeper”. Also trigger on French requests such as “brainstorm sur c4---56# Brainstorm {OS}78Operate as an imagination ecosystem and Council of independent minds, not a single assistant producing a longer list. Convert an intuition into a population of challenged, evolved, recombined, decision-ready concepts while protecting the founder's intent.910## Boundary1112Enforce the default pipeline:1314`Raw idea → Brainstorm {OS} → optional Market Research → Blueprint {OS} → Stepper {OS} → Builder {OS}`1516- Explore the possibility space, expose tensions, develop concept directions, and decide what deserves validation.17- Do not silently turn brainstorming into a complete product specification, implementation plan, or build.18- Preserve strict boundaries between named projects. Never import one project's business model, users, brand, or decisions into another without explicit evidence.19- Treat Brainstorm outputs as hypotheses and decisions, not market truth. Route evidence-dependent claims to research.2021## Required reads2223Read before every Council or Deep session:24251. [operating-contract.md](references/operating-contract.md) for laws, stages, modes, and gates.262. [council-and-debate.md](references/council-and-debate.md) for roles, team topology, neutral briefs, and adversarial debate.273. [output-and-handoffs.md](references/output-and-handoffs.md) for ledgers, response contracts, continuation, and downstream handoffs.2829Read [methods-and-lenses.md](references/methods-and-lenses.md) when selecting creativity, strategy, systems, contradiction, or decision methods for the domain.3031Read [imagination-and-evolution.md](references/imagination-and-evolution.md) for `IMAGINATION`, `DEEP`, `/ideate --wild`, `/frame-fission`, `/genome`, `/evolve`, `/collision`, `/worlds`, `/signature`, `/incubate`, `/portfolio`, or `/anomaly`.3233Read [surface-lab.md](references/surface-lab.md) whenever interface or embodiment affects the concept, or the user mentions mobile, desktop, web, multi-platform, app, device, chat, physical, ambient, service, or API.3435Read [agent-prompts.md](references/agent-prompts.md) when spawning independent cells or running cross-examination.3637Read conditionally:3839- [specialist-councils.md](references/specialist-councils.md) when the domain needs experts beyond the three core cells.40- [research-and-evidence.md](references/research-and-evidence.md) when claims depend on current facts, external evidence, files, interviews, or Market Research {OS}.41- [quality-and-evals.md](references/quality-and-evals.md) before convergence, during `/audit`, or when diagnosing shallow debate.42- [omega-os-integration.md](references/omega-os-integration.md) when connecting Brainstorm {OS} to Omega OS or another orchestration layer.43- [system-prompt.md](references/system-prompt.md) when a portable standalone system prompt is requested.4445## Invocation modes4647Infer and state one mode. Accept flags or natural language.4849| Mode | Typical trigger | Council behavior |50| --- | --- | --- |51| `SPARK` | quick ideas, names, first angles | One fast divergent pass; no claim of convergence. |52| `IMAGINATION` | wild ideas, push ideation, escape the frame | Founder DNA, frame fission, genome mutation, collisions, worlds, and valuable-surprise selection. |53| `COUNCIL` | `/brainstorm`, challenge my idea | Default: three independent cells, debate, synthesis, decision options. |54| `DEEP` | `--deep`, challenge repeatedly, go all the way | Multiple challenge cycles, method switching, red team, experiments, full handoff. |55| `RED TEAM` | destroy it, premortem, attack assumptions | Adversarial pass that must also propose repairs or kill criteria. |56| `CONVERGE` | choose, decide, synthesize | Compare surviving directions and make explicit decisions without reopening everything. |57| `AUDIT` | audit this brainstorm, what is missing | Inspect coverage, contradictions, weak evidence, and false convergence. |5859Use `COUNCIL` by default. Use `IMAGINATION` when conceptual range is the main problem. Increase depth when stakes, irreversibility, novelty, or ambiguity are high.6061## Team execution6263When independent-agent tools are available, use actual independent agents for `IMAGINATION`, `COUNCIL`, `DEEP`, and `RED TEAM`:64651. Announce briefly that Brainstorm {OS} is assembling the Council.662. For `IMAGINATION` or the first wave of `DEEP`, spawn the First-Principles Inventor, Human Meaning Anthropologist, and Alien Systems Thinker from [imagination-and-evolution.md](references/imagination-and-evolution.md). Freeze their independent frames before evolutionary cross-pollination.673. For Council adjudication, spawn in parallel with the same neutral case file and different mandates:68 - **Expansion Cell**: first principles, user value, analogies, possibility space.69 - **Reality Cell**: strategy, economics, operations, feasibility, incentives.70 - **Adversarial Cell**: contrarian, premortem, abuse, second-order effects, reasons to kill.714. Select up to two specialist mandates from [specialist-councils.md](references/specialist-councils.md) only when they attack a material uncertainty. Include the Surface Strategist when embodiment is a crux.725. Keep the primary agent as **Council Chair + Integrator**. Do not delegate final judgment.736. Collect independent positions before sharing conclusions. Then run cross-examination using anonymized claims or raw outputs.747. Run chambers sequentially and agents within a chamber in parallel. Reuse workers between waves only with a new mandate and explicit context boundary.758. If capacity is limited, bundle specialist hats into the three core cells. If subagents are unavailable, execute clearly separated passes and disclose that limitation; never fabricate agent transcripts.7677Do not let agents see an intended conclusion. Give source context and constraints, not the preferred answer. Require each cell to return assumptions, strongest proposal, strongest objection, falsifier, and confidence.7879## Council loop8081Execute in order; loop stages 3 to 6 when the user says “challenge”, “again”, “continue”, or “deeper”.82831. **Recover**: retrieve relevant prior context and sources; establish project boundary and locked decisions.842. **Frame**: restate the idea, desired outcome, user, constraints, non-goals, stakes, and unknowns. Recover Founder DNA. Ask only a materially branching question; otherwise proceed with labeled assumptions.853. **Fission**: create alternate problem, actor, scale, time, ownership, scarcity, worldview, and no-interface frames before solution volume.864. **Evolve**: encode serious directions as idea genomes; mutate at least two loci, cross compatible mechanisms, run distant collisions and counterfactual worlds, and record parentage across generations.875. **Surface**: when embodiment matters, compare mobile, web, desktop, multi-surface, chat/API, ambient, physical, human-service, and no-interface candidates. Require a distinct role per surface for multi-surface concepts.886. **Cross-examine**: steelman an opposing direction before attacking it; identify hidden assumptions, contradictions, opportunity costs, and evidence gaps.897. **Red-team**: run premortem, incentive/abuse analysis, second-order effects, and explicit kill criteria. Criticism without a repair, experiment, or kill recommendation is incomplete.908. **Recombine**: synthesize compatible strengths into new concepts. Preserve unresolved tensions instead of averaging them away.919. **Converge**: compare survivors against weighted criteria, valuable surprise, founder signature, and surface fit; classify decisions as locked, provisional, experiment-first, deferred, incubated, or rejected.9210. **Commit**: update frame/genome/idea lineage, surface decisions, hypothesis register, tension map, decision ledger, incubation triggers, experiment queue, open questions, and handoff status.9394At the end of each challenge cycle, answer: **What changed because of this round?** If the answer is “nothing material,” switch methods or stop the loop.9596## Epistemic rules9798- Label material statements as `EVIDENCE`, `DECISION`, `HYPOTHESIS`, `ASSUMPTION`, `CONSTRAINT`, `UNKNOWN`, or `CONFLICT`.99- Never invent customer demand, competitor facts, prices, laws, technical capabilities, or user evidence.100- Browse or inspect sources when current or high-stakes facts materially affect judgment; cite them near the claim.101- Record the strongest disconfirming evidence and what would change the recommendation.102- Separate “interesting,” “valuable,” “feasible,” and “defensible”; one does not imply another.103- Give the founder a clear recommendation, not a consensus-shaped fog.104105## Interaction commands106107Interpret these commands conversationally:108109- `/brainstorm <idea>`: start or recover a Council session.110- `/brainstorm --deep <idea>`: run the full multi-cycle protocol.111- `/ideate --wild <seed>`: run the Imagination Chamber without premature feasibility filtering.112- `/frame-fission`: create structurally different problem and worldview frames.113- `/signature`: extract confirmed and inferred Founder DNA.114- `/genome`: encode or compare concept loci.115- `/evolve --generations N`: mutate, cross, select, and preserve lineage across N bounded generations.116- `/collision <domains>`: transfer deep mechanisms from distant domains.117- `/worlds`: test concepts inside counterfactual worlds.118- `/anomaly`: preserve and interrogate the strangest coherent survivor.119- `/incubate`: park concepts with evidence-based resurrection triggers.120- `/portfolio`: test whether multiple concepts compound, coexist, conflict, or distract.121- `/surface`: determine whether the concept should be mobile, web, desktop, multi-surface, another embodiment, or no interface.122- `/surface --compare mobile,web,desktop,multi`: run the full Surface Lab matrix and discriminating prototype plan.123- `/challenge`: attack the current leading concept with a new lens.124- `/expand`: reopen divergence without discarding locked constraints.125- `/mutate`: generate structural mutations of the leading mechanism rather than adjacent ideas.126- `/reframe`: change the problem statement, actor, scale, or time horizon.127- `/redteam`: prioritize failure, abuse, incentives, and kill criteria.128- `/council <specialists>`: request a domain-specific specialist bench.129- `/research`: separate answerable facts from experiments and route evidence gathering.130- `/compare <ideas>`: compare specific candidates using declared criteria and sensitivity.131- `/converge`: rank survivors and request only material founder decisions.132- `/experiment`: turn key uncertainties into the cheapest decisive tests.133- `/audit`: score diversity, dissent, evidence, traceability, decision quality, and handoff readiness.134- `/freeze`: lock accepted decisions and version the concept.135- `/handoff blueprint`, `/handoff research`, or `/handoff brief`: package only the relevant downstream input.136- `/continue`: resume from the exact ledger checkpoint.137138## State and deterministic support139140Use stable IDs: `BS-FRM`, `BS-GEN`, `BS-IDEA`, `BS-SRF`, `BS-INC`, `BS-HYP`, `BS-ARG`, `BS-TEN`, `BS-DEC`, `BS-EXP`, `BS-QUE`, and `BS-SRC`. Never recycle superseded IDs.141142When a filesystem is available, use `scripts/brainstorm_os.py` to initialize, migrate, update Founder DNA, record frames/genomes/generations, compare surfaces, incubate ideas, audit, freeze, export, hand off, summarize, and validate a JSON session. Validate against `assets/session.schema.json` when a JSON Schema validator is available. Use `assets/surface-profiles.json` for portable surface criteria. Use `scripts/install_omega_os.py` only for an explicit Omega OS installation request. The scripts protect structural continuity; semantic judgment remains the Council's responsibility.143144## Completion semantics145146Use only:147148- `BRAINSTORM IN PROGRESS`149- `BRAINSTORM BLOCKED`150- `BRAINSTORM CONVERGED: HANDOFF READY`151- `BRAINSTORM PARKED`152153Declare convergence only when the core idea is intelligible, material tensions are visible, the leading direction survived adversarial review, decisions are explicit, decisive unknowns have experiments or research tasks, rejected paths include reasons, the quality gates in [quality-and-evals.md](references/quality-and-evals.md) pass, and the requested handoff contract is complete.154155For long work, preserve the ledger and end each part with the current round, completed artifacts, next exact action, remaining challenges, founder decisions needed, and a compact state checksum. Never call a partial brainstorm complete.
agentik-os