A Claude Code slash command that discovers net-new product ideas for any industry by transplanting proven mechanisms from other industries. Based on the analogy-mining research line (Kittur/Hope, SOLVENT, "Accelerating Innovation Through Analogy Mining"): abstract a pain point into domain-neutral terms, search distant domains for mechanisms that solved the abstract problem, and adapt the mechanism to the target industry.
Each run:
- Parses the target industry, optional focus area, and idea count N from the arguments (asks if missing).
- Researches the industry's current pain points via web search, requiring at least one or two from the end-user or worker perspective, not just the operator side.
- Builds a mechanism pool by searching 3-4 deliberately distant domains (epidemiology, maritime shipping, gaming economies, ...) for how they solve abstracted versions of those pain points.
- Generates candidates one at a time: mechanism + source industry -> specific use in the target industry.
- Runs each candidate through two gates:
- Relevance: does it solve a real problem for a real business in this industry? If kept, the source mechanism's actual operating details (unit economics, who pays whom, failure modes) are fetched so the transplant is faithful.
- Novelty: at least three differently-phrased searches for the combination inside the target industry, including startup/funding coverage. Existing elsewhere is fine; existing in the industry is a reject. Kept ideas get a High/Medium/Low novelty confidence, never a claim of certainty.
- Writes the kept ideas as cards (name, pitch, transplant, problem, novelty evidence, fit, why now) to
~/.claude/docs/innovation/, and dedupes against all prior runs so batches never repeat.
An attempt ceiling of 4xN guarantees termination.
cp commands/innovation-loop.md ~/.claude/commands/Requires Claude Code with web search enabled.
/innovation-loop <industry> [:: focus area] [N]
Examples:
/innovation-loop healthcare
/innovation-loop commercial real estate :: tenant experience 3
/innovation-loop insurance :: fraud
With no arguments it asks for the industry before running. Default N is 5.
If ~/.claude/docs/additionalContext/company.md exists, the loop reads it and targets ideas at that company: the "Fit" field on each card names which of the company's brands or channels would ship the idea. Without it, ideas target the industry generally. The file is free-form Markdown describing the company's products, brands, and goals.
One Markdown file per run in ~/.claude/docs/innovation/, named with the timestamp, industry, and focus area. Each idea is a card:
## MechaName
- **One-line pitch** - ...
- **The transplant** - [escrow milestones] from [freelance marketplaces] -> [progress-billed home renovation]
- **Problem it solves** - ...
- **Novelty** - Medium. Searches run: ... Closest adjacents: ...
- **Fit** - ...
- **Why now** - ...
Prior batches in that directory are grepped on every run so new batches never repeat an earlier mechanism-plus-use combination.
MIT. See LICENSE.