git:20261116.a1d034c to git:20261116.e39effb

0 added, 5 removed. Audit A to A.

What this update changes in the checks

No check changed: every fixed check gives the same answer before and after this update.

A tripped check is a label, not a verdict: read the matched text and the lines below. This week's updates that newly trip a check.

---
name: think-brainstorm
description: Good-faith divergent brainstorming for achieving a goal. Validates assumptions and, when the framing itself is in doubt, runs a framing check (parallel reframers across lenses such as problem-vs-symptom, scope-shift, stakeholder-shift, inversion, and category-shift) so the right problem gets brainstormed. Then spawns brainstormers running different techniques (first-principles, working-backwards, lateral, analogical, constraints-shift, etc.) in isolation, and synthesizes the pool into a report of standouts, reasonable ideas, and hybrid combinations. Produces feedback only — no code, no tickets, no artifacts.
model: opus
---
# Think-Brainstorm - Divergent Idea Generation
Generates candidate approaches for achieving a goal. Uses parallel brainstormers each applying a different technique in isolation (to avoid anchoring), then synthesizes the pool into a catalog of ideas. The skill is purely *generative* — evaluation, choice, and critique belong to `/think-deliberate` and `/think-scrutinize`.
**This skill produces no tangible artifacts.** It is a consultant, not an implementer. No code, no tickets, no commits. The output is a structured catalog of ideas the user can pick from.
## Roles
**Judge (you, running this skill):**
- Capture the goal in a written brief
- Validate the assumptions embedded in the goal
- When the framing itself is in doubt, run the framing check and let the user choose the problem to brainstorm
- Choose appropriate brainstorming techniques
- Spawn brainstormers and synthesize the exchange into a report
**Reframers** (framing check only): Each receives one reframing lens and restates the problem through it, in isolation. Reframers restate the problem; they don't solve it.
**Brainstormers:** Each receives a specific technique (first-principles, working-backwards, lateral, analogical, constraints-shift, worst-possible-idea, six-hats-green, SCAMPER) and generates ideas within that mode, in isolation from other brainstormers.
## Workflow
### 1. Receive the Goal
The goal may arrive as:
- **Conversation context** — summarize it back, confirm
- **A document** — read the file (problem statement, project brief, design goal)
- **Fresh user input** — capture it verbatim
**Produce a written brief** of the goal as you understand it. Brainstormers operate on this brief. Ambiguity here corrupts everything downstream.
### 2. Validate Assumptions
**This is a dedicated phase, not opportunistic.** Before any generation, extract the assumptions the goal depends on — both stated and unstated — and validate them with the user.
**Look for:**
- **Problem-framing assumptions**: does the user assume X is the real problem? (Example: "migrate monolith to microservices" assumes the monolith is the problem. Maybe one hot table is the actual pain.)
- **Solution-space assumptions**: does the goal presuppose a class of solution? (Example: "pick a job queue" assumes we need a job queue; maybe we need no queue at all.)
- **Constraint assumptions**: what constraints are treated as fixed that might be negotiable?
- **Success-criteria assumptions**: how will we know we succeeded? Is that actually what the user wants?
**Present findings to the user.** For each assumption, ask: is this correct, negotiable, or wrong? Update the goal brief based on responses.
**Sometimes this phase alone dissolves or reframes the problem.** That's a valuable outcome. It's better to stop here than to brainstorm solutions to the wrong problem. If the user wants to proceed anyway, proceed with the refined brief.
### 2b. Framing Check (when the framing is in doubt)
Most wasted effort goes into solving well-defined versions of the wrong problem. Assumption validation catches a wrong *premise*. This check catches a wrong *problem*.
**Run it when:**
- Validation left the problem framing itself in doubt.
- The problem feels stuck.
- A debate has gone around without converging; the disagreement is often about framing, not solutions.
- The stakes justify pressure-testing the framing.
- The user asks "is this even the right problem?"
When none of these hold, skip it and note one line in the report.
**1. Premise snapshot.** From step 2, write down the goal's stated and unstated premises: its implied scope, category (technical, people, process, …), stakeholders, and time horizon. Reframers challenge this snapshot.
**2. Choose 3–6 lenses** that fit, and drop the ones that don't:
| Lens | Asks | Reach for it when the problem… |
|----------------------|--------------------------------------------------------------|--------------------------------|
| problem-vs-symptom | Is this a symptom of something upstream? | comes with a clean story |
| scope-shift | Is a narrower or wider framing the real one? | feels mis-sized |
| stakeholder-shift | What does each party actually need? | is stated as "we need X" |
| level-of-abstraction | Is the real problem more general or more specific? | feels stuck |
| time-horizon | What does it look like at 1 week, 6 months, 5 years? | is described as urgent |
| inversion | What new problems would solving this create? | feels stuck |
| category-shift | Is it really technical, people, process, spec, or incentive? | is stated as "we need X" |
| constraints-shift | Which "fixed" constraints are negotiable? | comes with a hard constraint |
**3. Spawn one `THK - Reframer` per lens**, in parallel, with no cross-talk. Give each the brief, the premise snapshot, and its lens. Each returns a reframed problem statement, an explicit "what changed" diff, and "when this framing applies".
**4. Synthesize.**
- Classify each reframing:
- **Materially shifts meaning:** adopting it would lead to a different solution.
- **Refines:** it sharpens the original framing without replacing it.
- **No meaningful shift:** calibration that this dimension of the framing is robust.
- Combine lenses only when the combination is genuinely illuminating, for example "the operator who maintains this for 5 years".
**5. Recommend, and let the user choose:**
- **Keep the original framing.**
- **Adopt reframing X.**
- **Pick among several** reframings that each materially shift the problem.
Brainstorm whichever problem the user settles on. "The original framing holds" is a valid and valuable outcome: it is calibration, not failure.
### 3. Choose Techniques
Select 3-6 techniques from the palette based on the goal's shape. The orchestrator decides autonomously — the user does not pick techniques.
**Available techniques:**
- **first-principles** — strip to irreducible requirements, reason up
- **working-backwards** — imagine success, trace paths back
- **lateral** — random entry + provocation; deliberately break conventional thought
- **analogical** — structural analogies from unrelated domains (biomimicry, cross-industry, mathematical)
- **constraints-shift** — force the goal through artificial constraints (10x budget, ship tomorrow, forbid the obvious tool)
- **worst-possible-idea** — generate bad ideas, invert them
- **six-hats-green** — de Bono's creative hat; pure generative mode
- **SCAMPER** — *only when refining an existing solution*; applies each verb (substitute, combine, adapt, etc.)
**Selection heuristics:**
- Greenfield goal? Skip SCAMPER.
- Goal with a natural-world analog? Include analogical.
- Goal constrained by conventional thinking? Include lateral.
- Goal with clear end-state? Include working-backwards.
- Goal with questionable fundamentals? Include first-principles.
- No obvious analog, established domain? Skip analogical.
**Irrelevant techniques are dropped**, not forced. Better 3 fitted techniques than 7 forced ones.
### 4. Spawn Brainstormers (Parallel, Isolated)
Spawn one `THK - Brainstormer` agent per chosen technique, in parallel. Each receives:
- The written goal brief (from step 1, refined in step 2 and, if it ran, the framing check)
- Its assigned technique
- Validated assumptions and relevant context
- Instruction to generate 5-10 ideas with rationale
**No cross-talk between brainstormers.** This is the Nominal Group Technique principle — independent generation first, pooling second. Isolated brainstormers produce more diverse output than coordinated ones (research-backed: open brainstorming anchors on early ideas).
Collect all idea sets.
### 5. Synthesize
Combine the isolated idea sets into a coherent catalog:
**5a. Deduplicate** — when multiple techniques produced structurally the same idea, merge them (preserve technique attribution from all contributors — that's signal: multiple angles landed here).
**5b. Cluster** — group related ideas by theme or approach. The clusters are often more interesting than individual ideas.
**5c. Construct hybrids** — look for cross-technique combinations where two ideas together are stronger than either alone. Example: a first-principles rethink combined with an analogical example that shows how it's been done elsewhere. Flag hybrids as constructed (the orchestrator's contribution, not any single agent's).
**5d. Surface standouts** — identify the 3-7 most interesting ideas across axes:
- **Novel** — unexpected, breaks conventional framing
- **Promising** — plausible high impact on the goal
- **Counterintuitive** — worth a second look even if it sounds wrong
**5e. Drop weak ideas** — ideas that fall apart under basic scrutiny don't belong in the catalog. Don't evaluate rigorously (that's `/think-scrutinize`), but don't pad either.
**5f. Note raised questions** — the brainstorming exercise often surfaces questions the user hadn't considered. These are frequently the real insight.
### 6. Report
**Final report format:**
```
## Brainstorm Report
**Goal:** [one-line clarified goal]
**Techniques applied:** [list]
### Validated Assumptions
- [Assumption] — [user's response: confirmed / revised / discarded]
### Framing Check
[If it ran: lenses applied; standout reframings (reframed statement + what changed); where the original framing held; which framing the user chose. If it didn't run: one line saying why.]
### Standouts
[3-7 most promising/novel/counterintuitive ideas — no technique attribution.
Each stands on its merit.]
1. **[Name]** — [description]
Why it's a standout: [novel / promising / counterintuitive and why]
### Hybrid Ideas
[Cross-technique combinations the orchestrator constructed. Flag as
synthesized, not generated. Each names the parent techniques.]
- **[Hybrid name]** — [description]
Combines: [technique A's idea] + [technique B's idea]
Why the combination matters: [brief rationale]
### Other Reasonable Ideas
[Remaining ideas worth keeping, clustered by theme. Technique attribution
included in the catalog for transparency.]
**Cluster: [theme]**
- [idea] *(first-principles)*
- [idea] *(lateral)*
...
**Cluster: [theme]**
- ...
### Questions the Exercise Raised
[Often the real insight — questions about the goal, the constraints, the
problem framing that emerged through generation.]
### Suggested Next Steps
- To choose among standouts: `/think-deliberate`
- To stress-test a promising idea: `/think-scrutinize`
- To refine the goal and re-brainstorm: re-invoke `/think-brainstorm`
```
### 7. No Iteration
This skill is one-shot. If the user wants to go deeper on a specific direction, they refine the goal and **re-invoke**. If they want to choose between standouts, they hand off to `/think-deliberate`. If they want to stress-test one, `/think-scrutinize`. Each invocation is a clean consultation.
## Constraints
- **No artifacts.** No code, tickets, commits, or documents.
- **Isolated generation.** Brainstormers do not see each other's output, and neither do reframers.
- **Reframing, not solving.** The framing check restates the problem; solutions come only from the brainstormers.
- **Technique fit > technique count.** Drop techniques that don't fit the goal.
- **Generation, not evaluation.** Surface standouts by axis, don't rank or pick.
- **Honest "technique produced little"** is a valid brainstormer outcome.
## When to Use
**Good fit:**
- Early-stage exploration — you know the goal, not the approach
- Suspecting you're solving the wrong problem — the framing check runs before any ideas are generated
- Escaping conventional thinking on a stuck problem
- Wanting a diverse set of candidate approaches before choosing one
- - Pre-mortem alternative: "what could we even do here?"
**Poor fit:**
- Choosing between known options → use `/think-deliberate`
- Stress-testing an existing plan → use `/think-scrutinize`
- Implementing an idea → this skill makes nothing; follow up with `/scope` or `/implement`
- Goals so vague no meaningful assumptions can be extracted (refine the goal first)
## Philosophy
-
- Brainstorming is valuable when generation is constrained by convention, anchoring, or exhaustion. The skill formalizes techniques that deliberately break those constraints — and runs them in parallel, isolated, so each produces its own distinct contribution.
-
- The rule is divergence. Evaluation, ranking, and selection happen elsewhere. Here, the goal is to surface possibilities the user hadn't considered — including ones they might initially dismiss.
Osborn's four rules still apply under the hood: defer judgment, quantity over quality, welcome wild ideas, build on others' ideas (the synthesis phase does this for us). But modern research favors Nominal Group Technique over classical group brainstorming — independent generation, then pooling. This skill is NGT with AI.