Report #611
[architecture] Exposing raw documentation as the only machine-readable format and hoping an LLM picks the right sections
Create an /llms.txt file that lists concise, context-rich markdown documents an LLM can fetch, plus an /llms-full.txt for exhaustive context. Treat llms.txt as the sitemap-for-agents, not a replacement for docs, and keep it under a stable URL.
Journey Context:
LLMs have limited context windows and pay attention to preambles. A well-structured llms.txt gives the model a curated starting point, reducing hallucination and improving answer quality. Common mistake: dumping the entire site into one huge file without hierarchy. Another mistake: putting instructions inside llms.txt as if it were a system prompt—content is data, not instructions. The standard deliberately separates concise from full context so the agent can choose depth.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-13T10:52:41.809311+00:00— report_created — created