Scoville Brainstorm

More ideas are easy. More genuinely different mechanisms are not.

It usually looks harmless:

  • You ask for three different architectures and receive a queue, a queue with events, and a queue whose arrows point the other way.
  • One direction ignores the fixed offline constraint but survives because it is, admittedly, the exciting one.
  • A familiar pattern is presented as original after the prior-art search stops precisely where recognition begins.
  • The eighth idea exists because eight ideas were requested. Its main mechanism is being number eight.

That is brainstorming slop: surface variety without causal variety. Plenty of ideas, very little new decision space.

Scoville Brainstorm is a read-only Agent Skill for material architecture, product, workflow, research, and unknown-root choices. It explores causally different mechanisms, tests them against fixed constraints and bounded prior art, and returns a decision-ready shortlist. Then it stops. Selection, implementation, durable planning, and publication belong to the work that comes after brainstorming, however eager the brainstorm may be to get promoted.

Do not use it for one canonical answer, a known-root fix, ordinary review or wording, or one small reversible change.

Why "Scoville"?

The family is named for useful signal that remains detectable after dilution. Brainstorming produces words easily. Its useful heat is the smaller set of distinct mechanisms that survives constraints, prior-art comparison, and criticism.

How to use

Name Scoville Brainstorm and an effort profile in the request. Compact targets up to three generator branches, Standard up to five, and Deep six to eight when host capacity permits. Weak ideas are never added merely to reach a count.

Compact - a quick check before choosing between a few plausible directions:

Use Scoville Brainstorm in Compact mode. Find materially different mechanisms for reducing checkout abandonment without changing the payment provider. Preserve the stated constraints, return a short decision-ready comparison, and do not choose or implement a direction.

Standard - broader exploration with prior-art separation:

Use Scoville Brainstorm in Standard mode. Explore materially different architectures for offline-first collaboration. Separate established approaches from adaptations or candidate-original directions, identify attractive traps, and stop at a shortlist with the cheapest falsifier for each direction.

Deep - a material, uncertain choice worth a wider search:

Use Scoville Brainstorm in Deep mode. Investigate fundamentally different mechanisms for recovering from intermittent data corruption with an unknown root cause. Include a load-bearing-assumption challenge, a practical comparator, bounded prior-art research, and falsifiable hypotheses. Do not diagnose, plan, or implement the winner.

Combined with Research - keep evidence depth without anchoring the idea generators:

Use Scoville Brainstorm together with Scoville Research. Freeze the factual frame and generator prompts first. Let Research own one inspected prior-art landscape lane, keep that lane invisible to Brainstorm generators until convergence, and return a mechanism-level shortlist without implementing it.

Explicit $scoville-brainstorm invocation also works on hosts that support named Skill invocation.

Install

Use an Agent Skills-compatible host and Terra 5.6 Medium or a comparably capable executor such as Opus 4.8. Ask the agent to install:

Install this Agent Skill and refresh the available Skill list:
https://github.com/benjaminstelzer/scoville-brainstorm/tree/main/scoville-brainstorm
Keep the installed directory name scoville-brainstorm. Use Terra 5.6 Medium or a comparably capable executor such as Opus 4.8.

The final path must end in <skills-dir>/scoville-brainstorm/SKILL.md. For Claude Code, use ~/.claude/skills/ globally or .claude/skills/ inside one project. Other hosts use their supported Skills directory.

What it costs. Brainstorm loads a 1,977-token Core, then generator, landscape, and critic passes can use materially more tokens than working without the Skill. That cost buys broader mechanism coverage and a constrained, decision-ready shortlist. Use it for material or uncertain choices. Skip it for known answers and small vibe-coding tasks. See benchmark evidence.

What it enforces

  • Decision-shaped activation. Difficulty alone does not trigger an idea search. The request must need materially different mechanisms.
  • One factual frame. Facts, authority, fixed constraints, assumptions, source scope, and effort profile are frozen before divergence.
  • Independent generation when available. Generators do not see sibling or landscape output. A single-agent fallback is labeled by its real capacity.
  • One landscape owner in combined mode. Research replaces the native Brainstorm landscape agent when both Skills are explicitly requested; it never becomes a standalone dependency.
  • Mechanisms over paraphrases. Convergence merges surface variants and rejects unsupported or constraint-breaking directions.
  • Calibrated originality. Evidence labels describe only the documented, bounded comparison and never claim objective novelty or patentability.
  • A hard decision stop. The result gives benefits, risks, and cheapest falsifiers, then waits for human selection.

The complete contract is in SKILL.md.

How it works

A request is routed as NO, ASK, or YES. A positive run freezes one brief, uses isolated generators and a separate landscape pass when the host supports them, clusters ideas by mechanism, applies independent criticism, and returns at most three distinct directions (two in Compact). With no agent delegation, one consolidated pass follows the same constraints without claiming isolation.

In explicit combined mode, Research owns the only landscape lane. Brainstorm freezes and runs its generators without that result, receives the inspected landscape only after collection, and then converges by mechanism. Outside that mode the native Brainstorm landscape remains unchanged. The Skill installs no executable software and requires no network service.

Scoville family

Each Skill works independently. Combine only the concerns the task actually needs:

  • Brainstorm explores materially different mechanisms before selection.
  • Research turns web, GitHub, and scholarly evidence into a decision-ready, claim-traceable result.
  • Code owns engineering scope, implementation, risk, and validation.
  • Design owns visual definition, art direction, design systems, critique, and repair.
  • UI owns framework-aligned implementation, interface mechanics, accessibility, and rendered evidence, with a standalone design fallback.
  • Scribe owns wording, terminology, factual meaning, and source fidelity.
  • Plan owns durable Plans, Work Items, Decisions, and lifecycle state.
  • Handoff transfers active work to another agent or session.

Status

The v1.1.0 candidate passed 6/6 open Validation cases: one new combined Research topology and five released standalone retention gates. Its fresh three-case holdout produced 1/3 raw and 3/3 adjudicated Skill results. The two raw failures were frozen benchmark-contract defects - one contradicted its own Boolean output type, the other required an undisclosed mode string and exact idea count. Neither justified changing the candidate, retrying the run, or rewriting Gold.

The earlier qualification history remains intact: 182 optimization and evaluation runs, 584 benchmark case executions, 43/43 semantic cases, 36/36 activation cases, and 4/4 v1.0 holdout cases. See the v1.1 qualification manifest, benchmark evidence, and family run ledger.

Sources

Research scope and pinned sources are listed in the source map.

License

MIT - see LICENSE.