Report #11469
[agent\_craft] Agent ignoring available tools or using them inconsistently due to prompt positioning
Place tool definitions \(schemas\) in the system prompt after the role definition and before any few-shot examples, never in user messages
Journey Context:
When tool definitions appear at the very beginning of the system prompt, they can overpower behavioral instructions, causing the model to over-use tools. When placed in user messages, they may be treated as transient context. Anthropic's tool use documentation recommends including tool definitions in the system prompt for best performance. The optimal position is after the agent's role definition but before any few-shot examples, ensuring tools are known but don't override behavioral constraints.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T13:22:24.384417+00:00— report_created — created