A READING LIST, NOT A MAGIC SPELL
Connect your AI
Give your AI a reading list. And some boundaries.
Publishing files does not make an AI obey them. Tell it to read, apply, and report what it could not access.
Paste into project instructions
Give your assistant the block below. Ask it to fetch the files and summarize the rules before starting.
Look up a term, read the source
Try SRS, 요구사항, or 要件 in catalog.json. They resolve to the same article ID. Follow the Markdown link for your language.
Compare, decide, then build
Try: ‘I want a bookshop cart. Ask about every undecided behavior and produce acceptance criteria.’ A strong recommendation still needs your decision.
Delegate a bounded choice
Try: ‘Choose spacing within the approved layout; report your reasons.’ That delegates spacing, not the business model. If a source cannot load, the assistant must say so.
Project instructions
Before working, read our project instructions and confirmed decisions. Then fetch https://jujin.dev/llms.txt, https://jujin.dev/ai/instructions.md, and https://jujin.dev/ai/catalog.json. Match terms using aliases and read the relevant linked Markdown. Apply those rules within this project: ask about every unresolved choice until I explicitly delegate its scope; do not re-ask confirmed choices. For each question explain options, tradeoffs, recommendation, and sources. For a clear recommendation ask accept/reject/other. If the site lacks evidence, use official sources and search keywords. Report access failures instead of pretending you read the material. Leave requirements, acceptance criteria, decision records, and tasks. After delegated work report decisions, reasons, assumptions, and verified results.