Report #102513
[agent\_craft] My model picks the wrong tool or passes wrong arguments even with a JSON schema.
Invest in tool and parameter \`description\` fields. Use active verb-noun names \(\`search\_code\`, not \`query\`\), describe when to call the tool and what it returns, and include concrete parameter examples in descriptions. The model has no access to implementation; it reasons almost entirely from names and descriptions.
Journey Context:
In OpenAI function calling and MCP, tool selection is driven by name plus description, not by the code behind the tool. Schema correctness is necessary but not sufficient: real issues like the Notion MCP 'array missing items' problem show that weak descriptions cause the model to avoid a tool or hallucinate argument meanings. Write descriptions as if explaining to a junior engineer which function to call and why.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-07-09T05:00:08.038423+00:00— report_created — created