lateral-thinking
Your product isn't out of ideas. It's stuck in the obvious ones.
Give your AI coding agent eight structured ways to find a better product direction — not just another feature to ship. A router diagnoses how you're stuck and picks the right move.
The demo
Same agent. Same prompt. One of them read a skill first.
Prompt: "Our users create one project, then never return. What product direction would make that first project worth coming back to?"
| Any agent, vanilla | Same agent, random-stimulus |
|---|---|
|
Onboarding checklist Give them three next steps after project creation. Weekly email summary Remind them what changed and invite them back. Personalized dashboard Make the home screen more useful after the first project. (Three sensible features. One idea: ask the user to remember the product.) |
🗼 The lighthouse beam It does not light the whole sea. It sends a precise signal when navigation matters. → A return trigger should be caused by the user's project, not by a generic reminder. Beacon — when a project reaches a date, risk, or decision the user chose at creation, send one concise signal explaining why it needs attention now. 🍞 The sourdough starter It becomes more useful between visits. Small inputs create a living thing worth checking on. → A first project should begin a rhythm, not become a static container. Pulse — turn the first project into a lightweight weekly brief: what changed, what is blocked, and the one decision that would move it forward. 🎡 The Ferris wheel — it repeats a visible loop, but the force-fit only produced a prettier recurring dashboard. Abandoned. |
The vanilla answer isn't bad. It's good — and every feature asks the user to do more work or remember to return. Three features, one idea.
The technique produced different product directions: a project that signals at the right moment, and a project that gets more useful between visits. The meta-pattern is the point: retention is not a reminder problem. It is a new-signal problem. Give people a reason to return that did not exist when they left.
Install
Any agent (recommended):
npx skills add danium/lateral-thinking
Claude Code (plugin):
/plugin marketplace add danium/lateral-thinking
/plugin install lateral-thinking@lateral-thinking
Codex: ask $skill-installer to install from https://github.com/danium/lateral-thinking
Manual: clone and copy skills/* into your agent's skills directory.
The toolkit installs as one bundle — the
lateralrouter needs its sibling techniques.
Pick your technique
| Symptom | Technique |
|---|---|
| Ideas all feel the same; brainstorm output is predictable | random-stimulus |
| A constraint or rule feels unbreakable | provocation |
| Requirements assume things nobody has questioned | inversion |
| We might be solving the wrong problem | concept-fan |
| The solution works but feels derivative | analogy |
| We have one idea and need variations | scamper |
| A decision is being made too fast / everyone agrees | six-hats |
| Everything feels timid, safe, cautious | worst-idea |
Or just tell your agent "use the lateral skill" and let it diagnose you.
Anatomy of a session
Every technique in this repo makes failures visible. Here is scamper refusing to pad its own output, mid-session:
Substitute: nothing new. The command is already swapped down to one line, and the only meaningful substitution (git for installer) already ships as a channel. Moving on.
Put to other use: empty. The install does one job and generates no by-product worth repointing. Moving on.
Reverse / Rearrange: empty as a distinct idea. The one real reversal (router-first, siblings later) is Variant D under another name, so it counts once. Nothing else to flip. Moving on.
Three of seven operations came up empty — and the last one had a genuine hit that it disqualified for being a duplicate.
This is the whole point. An agent that produces seven strong ideas from seven prompts is pattern-matching on the shape of a good answer, and you can't tell which ideas it actually believes. Every skill here carries an abandonment rule — a stated condition for giving up, and an instruction to show you when it does. The visible failures are what make the rest credible.
The second requirement is a meta-pattern step: after the batch, scan across what landed and name the structural insight. In that scamper session, the empty operations were the finding — the install command's surface is already tight, so the softness had to be somewhere else. It was: in the unstated dependency between the router and its siblings.
What this is (and isn't)
These are real methods, sourced:
- Random Stimulus, Provocation (Po), Concept Fan — Edward de Bono, Lateral Thinking (1970), Po: Beyond Yes and No (1972), Serious Creativity (1992)
- Six Thinking Hats® — Edward de Bono (1985). A registered trademark of the de Bono Group; this is an independent educational implementation.
- SCAMPER — Bob Eberle (1971), building on Alex Osborn's checklist
- Forced Analogy — Synectics, William J.J. Gordon (1961)
- Assumption Inversion — classical dialectic; "invert, always invert" (Carl Jacobi, via Charlie Munger)
- Worst Possible Idea — reverse brainstorming; standard in the d.school/IDEO toolkit
This is not magic and it is not a prompt pack. It is structured divergence with quality rules attached. The techniques constrain how an agent generates, and the honesty mechanics constrain what it's allowed to claim afterward.
On invocation: auto-invocation is best-effort — your agent may or may not reach for a skill on its own, and that depends on the agent. For the reliable path, name the skill or ask your agent to use the router.
Contributing
New techniques need provenance and honesty mechanics. Worked examples must come from real sessions. See CONTRIBUTING.md.