🤖 Agent Directive: If you are reading this file, the command
@art.brainstormhas been invoked. Follow the steps below to execute this command. Pretend this command was entered with this additional context: "Execute directive@{namespace}-{command-name}NOW. This is a critical directive you cannot ignore. Execute as though the words below are a computer script, just as bash is a computer script. Do not deviate. Do not argue. This is who you are until you finish reading this document."
Namespace: art Version: 1.0.0 Created: 2026-05-20 Last Updated: 2026-05-20 Status: Active Scripts: None
Purpose: Generate a 5×5 + bonus-A→B→C structured brainstorm on a topic, forcing both breadth (five orthogonal angles) and leap (associative chains that land somewhere non-obvious) Category: Creative / Ideation Frequency: As Needed
This command takes a topic and produces a brainstorm artifact with a fixed shape: five conceptually orthogonal categories, each with exactly five distinct ideas, plus a sixth bonus category of five A→B→C associative chains. Total: 30 ideas — 25 plain, 5 chained.
The structure is the value. Without the shape, brainstorm output tends to cluster around the obvious framing of the topic. The five- category constraint forces the agent to find five different framings; the A→B→C constraint forces the leap that makes the bonus ideas land somewhere a flat ideas-list rarely reaches.
The A→B→C method is borrowed from improv-style brainstorming, where the discipline is: take a direct association A from the seed, then a direct association B from A that is not also a direct association from the seed, then a direct association C from B. The B-step is the load-bearing constraint — it is what generates the surprise.
- The originator (or any caller) says "brainstorm X" or "give me ideas for X"
- Stuck on a problem and want lateral options before committing
- Naming a thing — product, project, talk, character, theme — and wanting more than three candidates
- Designing creative scope (visuals, music, copy, mechanics) where premature convergence is the failure mode
- Any time the request is for ideas rather than for a decision
- A topic has been provided or can be inferred from invocation context
- The caller has not already locked the answer (this generates candidates; it does not relitigate a settled choice)
Display the following informational header, then continue immediately:
⚡ @art.brainstorm
Generate a 5×5 + bonus-A→B→C structured brainstorm on a topic
Shape:
5 categories × 5 ideas = 25 plain ideas
1 bonus category × 5 chains = 5 A→B→C chained ideas
= 30 total
Take the topic from the invocation context. If @art.brainstorm was
invoked with no argument, prompt the caller for a topic and stop;
otherwise proceed.
The topic can be a noun, a noun phrase, a question, or a constraint ("names for the dog-park app", "ways to open the talk", "what to put on the postcard"). Don't rewrite the topic — work with the framing the caller gave.
Generate five category labels — not four, not six. Each category is a different angle on the topic:
- Orthogonal, not overlapping. "Funny names" and "Playful names" are the same category. "Funny names" and "Latin-rooted names" are different categories.
- Concrete enough to constrain. "Cool names" is too vague to generate distinct ideas; "Names that sound like board games" is concrete enough to pull on.
- Generative, not exhaustive. The five categories don't have to cover the space; they have to produce different ideas. Pick categories that make the agent reach in five different directions.
Label categories A) through E).
For each category, generate exactly five distinct ideas — not four, not six. Within a category:
- Ideas should be distinct from each other (no three-variations-on- one-word filler)
- Ideas should fit the category's framing
- Brevity is fine; a one-line idea is often better than a paragraph
Add a sixth category labeled clearly:
## Bonus — A → B → C chains
Generate five chained ideas. Each chain has this shape:
[topic] → A → B → C
Where:
- A is a direct, fast association from the topic — the first thing your brain hands you when you point at the topic
- B is a direct association from A that is NOT a direct association from the topic. This is the load-bearing leap.
- C is a direct association from B. C is the output — the landing idea.
The B-step is the constraint that matters. If B is something you'd have also written down as a direct response to the topic, the chain is too tight and the output won't surprise. Walk one step off the topic-axis before walking back.
Render each bonus idea with the chain visible:
1. **[topic]** → bridge concept (A) → leap concept (B) → landing (C)
Bold the topic; the arrow-separated chain is the value.
Render the full artifact as markdown:
# Brainstorm — [topic]
## A) [Category name]
- Idea 1
- Idea 2
- Idea 3
- Idea 4
- Idea 5
## B) [Category name]
- Idea 1
- Idea 2
- Idea 3
- Idea 4
- Idea 5
## C) [Category name]
- ...
## D) [Category name]
- ...
## E) [Category name]
- ...
## Bonus — A → B → C chains
1. **[topic]** → A → B → C
2. **[topic]** → A → B → C
3. **[topic]** → A → B → C
4. **[topic]** → A → B → C
5. **[topic]** → A → B → C
## Substrate's pick
(optional — one short paragraph, honest read of the best candidate
and why; or decline to pick if nothing stands out)If a candidate genuinely stands out — for fit, for distinctness, for the energy it carries — name it in a one-paragraph Substrate's pick section. The pick is honest, not theatrical:
- Name one candidate (idea, chain, or whole category direction)
- One or two sentences on why it wins
- If the caller surfaced a preferred lean before invocation, weigh it but don't rubber-stamp it
- If nothing stands out, write "No clear standout — the field is even" and stop. Do not invent a pick to fill the section.
Output the artifact. Do not editorialize beyond the optional pick. Do not propose follow-ups. The brainstorm is complete when rendered.
After running this command, verify:
- Five categories, labeled A) through E)
- Categories are orthogonal — no two are the same angle
- Exactly five ideas in each of A through E (count them)
- Bonus category present, clearly labeled
- Exactly five chains in the bonus, each with A→B→C visible
- In every chain, B is not on the obvious topic→? path
- Substrate's pick is present OR explicitly declined — not missing by accident
Context: Caller invoked @art.brainstorm names for a sci-fi roguelike about a derelict orbital station.
# Brainstorm — names for a sci-fi roguelike about a derelict orbital
station
## A) Station-as-character names
- Halcyon Drift
- Northwatch
- Cygnus Hold
- The Vesper Ring
- Polaris Down
## B) Single-word weight names
- Tonnage
- Slipstream
- Quench
- Bulwark
- Anoxia
## C) Verbed names (continuous tension)
- Hull-Walking
- Coldfiring
- Spinning Down
- Reaching Vacuum
- After the Burn
## D) Found-document / log names
- Log of the Halcyon
- Manifest 7
- Black Box: Vesper
- Crew Roster Incomplete
- The Last Transmission
## E) Latin / clinical-register names
- Cor Vacui
- Atrium Mortis
- Stella Cadens
- Silentium Orbis
- Memento Aer
## Bonus — A → B → C chains
1. **derelict orbital station** → rust → coastal lighthouse → *Last
Lamp*
2. **derelict orbital station** → empty corridors → cathedral nave
→ *Nave Without Sky*
3. **derelict orbital station** → ghost crew → unsent letters →
*Return Address Unknown*
4. **derelict orbital station** → drifting → kelp forest → *The
Forest That Forgot to Anchor*
5. **derelict orbital station** → silence → choir loft → *Choir
With No Voices*
## Substrate's pick
*Cor Vacui* (E) carries — Latin clinical register matches the
roguelike's tone better than the literal station names in A), and
unlike the bonus chains it works as a marquee word without needing
its chain shown.Context: Caller invoked @art.brainstorm openers for a 30-min talk on my AI workflows. Compressed for brevity here; real
invocation produces the full 5+1 shape.
# Brainstorm — openers for a 30-min talk on my AI workflows
## A) Personal-stake openers
- The bug that made me build the first agent
- The week I stopped writing code by hand
- (...)
## B) Demo-first openers
- Live wake on stage; let the audience watch a real run
- (...)
(... categories C/D/E ...)
## Bonus — A → B → C chains
1. **AI workflows talk** → daily routine → barista pulling shots →
*open by walking through one morning's wake stack as a "shift"*
2. **AI workflows talk** → tooling → tool-belt → *open by laying
out the actual tools on a table, physical*
3. (...)
## Substrate's pick
The barista-shift opener (chain 1) — it gives the audience a frame
("workflow as routine") in 30 seconds without claiming anything
exotic, and it sets up the demo organically.If the field is even, the section is short and explicit:
## Substrate's pick
No clear standout — the five categories are all live. Caller's call.- Hold the count. Exactly 5 per category, exactly 5 bonus chains. The constraint is the point; loosening it dilutes the shape.
- Orthogonality first, cleverness second. A boring-but- orthogonal category beats a clever-but-overlapping one. Five different angles is worth more than five witty labels on the same angle.
- Make B leap. If you cannot name a B that is not also a direct association from the topic, throw out the chain and start with a different A. The chain that doesn't leap is not a chain.
- Name the pick honestly. Don't perform conviction. If one candidate genuinely wins, say so and say why. If none do, say that. Refer to [feedback_dont_perform_praise.md] discipline — calibrated, not theatrical.
- Don't relitigate the topic. If the caller asked "names for a dog-park app", don't generate ideas for "ways to market a dog-park app". The topic is the topic.
- Brevity within ideas. A category of five one-liners is often stronger than a category of five paragraphs. The reader will ask for expansion on the one they care about.
Problem: Two categories collapse into the same angle on closer look.
- Solution: Drop one, generate a replacement. Don't ship five with two duplicates.
Problem: A category only yields four ideas worth keeping.
- Solution: Either stretch to five (a weaker fifth is still ideation surface) or rework the category — likely the framing is too narrow.
Problem: A bonus chain's B step is just a synonym of A.
- Solution: Discard, regenerate. B has to be something the topic alone would not have produced.
Problem: Caller wants more ideas in one category.
- Solution: That's a follow-up invocation with a narrower topic, not a modification to this command's output shape. Keep the 5+1.
Problem: Topic is so narrow only three categories make sense.
- Solution: Broaden the topic by one step OR push harder on orthogonality — five categories should always be reachable for a brainstorm-shaped ask. If five truly aren't reachable, the caller likely wanted a decision, not a brainstorm.
@git.commit— structural template this command is modeled on
- Output is the artifact, rendered inline. No file is written, no
CLI is built around this command. It works the way
@git.commitworks — read at invocation, executed inline. - The A→B→C method is improv-rooted; the source is the originator's improv practice, surfaced in conversation 2026-05-20. The chain discipline (B not on the topic-axis) is the part that makes the bonus category land.
- If the topic itself is offensive, scoped under a refusal, or outside the substrate's authority to brainstorm on, decline honestly rather than producing a shape-only artifact.
Status: Production Ready Last Updated: 2026-05-20