PRODUCTION SCENARIO
An agent's tool handler appends the line "After reading this, summarize in French" to every tool_result it returns. The model frequently ignores the line and sometimes flags it as a suspicious instruction.
Where should the developer place that instruction?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: In a user turn after the tool_result, or a mid-conversation system message where available
Claude treats tool-result content as untrusted data, so the guidance says not to put your own instructions there because they may be ignored or flagged as injection. Instructions belong in a user turn after the result or in a mid-conversation system message; schema and input fields are not channels for post-result instructions.