Give this archive to your AI.

Your assistant already answers with what it can find. This makes it answer with what it could never have found: 404 documented solutions that worked from an angle nobody expected — filed by the shape of the problem, not the industry, so a restaurant's empty afternoon reaches an airline's 1985 fare system.

why a search engine cannot do this

Search matches the words of your problem, and the answer you need is filed under somebody else's vocabulary. Nobody solving an empty dining room types "yield management"; nobody whose safety spec goes unread types "compliance canary". You cannot search for a move you have never imagined — that is the whole reason this archive exists.

install

curl -fsSL https://genius.wiki/skill/SKILL.md \
  -o ~/.claude/skills/genius-wiki/SKILL.md

Any agent that reads markdown skills works — Claude Code, Claude Desktop, or your own harness. There is nothing to sign up for and no key to manage.

or call the API directly

GET https://genius.wiki/api/genius/search
  ?q=our restaurant is empty every afternoon
  &limit=6

Open, unauthenticated, GET-only. Each result carries the move, what it earned, the problem-shapes it belongs to, its sources, and one line saying what business situation it transfers to.

what it looks like in a conversation

you Nobody on site actually reads the safety spec before signing. How do I know which contractors did?

your aiVan Halen had this problem in 1982 and solved it with a bowl of M&Ms. Their tour rider — hundreds of clauses about rigging loads and power draw — buried one absurd line: no brown M&Ms backstage. Brown M&M in the bowl meant the promoter had skimmed the contract, and Roth called a full line-check of the staging. It cost nothing and reliably predicted real technical errors. For you: plant one trivial, checkable instruction inside the serious document, and its state tells you whether the serious part was read.

the honest limits

It answers with precedent, not with your situation — the traps listed on each shape are where transfers usually die.
An empty result means the archive has not filed your shape yet. The archive is growing daily; it is not finished.
Best practice is deliberately excluded. If what you need is the standard answer, ordinary search serves you better.

Sibling archive: mistake.club — that one tells your AI how it went wrong before. Run both: check the failures before you commit, and come here for a way through.