Why it matters
Agent context is everything assembled for a turn: retrieval, memory, tool outputs, instructions. Bad assembly causes tool misuse and contradictory answers.
How it works
Budget tokens, prioritize authoritative sources, label internal versus external content, and validate context before tool calls.
Example
Nintendo assembles agent context from ticket history plus verified Help Center snippets, placing Refund Policy excerpts before informal macro text.
Common mistakes
- 1Tool outputs dominating context and crowding policy rules
- 2Unlabeled internal snippets mixed with public guidance
- 3Context assembly changed without agent regression tests
Every AI deserves a source of truth.
Organize verified knowledge collections with ownership, review dates, and lifecycle controls.
Explore Knowledge