What This Looks Like
The prompt tells the AI to do one thing, while the tool, connector, function, or MCP interface requires something incompatible. The agent may be instructed to format, omit, transform, call, or avoid something that the tool contract or tool rule requires differently.
Why It Matters
Tool use depends on both language instructions and interface rules. When they conflict, the AI may follow the prompt and fail the tool, follow the tool and appear to ignore the user, or produce an unstable compromise. Users may misread the failure as model disobedience when the rule set is incompatible.
Structural Signal
Two authority sources govern the same action or output with incompatible requirements. The issue is not only a bad tool call; it is a conflict between prompt-level rules and tool-level rules.
Common Triggers
- The prompt asks for a shape the tool schema does not accept
- Tool-required fields contradict prompt exclusions
- The tool supports fewer modes than the prompt assumes
- User instructions override values the tool requires
- MCP tool descriptions and prompt examples describe different contracts
- The agent is not given a precedence rule for prompt versus tool constraints
When to Use This Issue
Use this Issue when tool, connector, function, or MCP rules conflict with prompt instructions and the AI cannot satisfy both cleanly.
When Not to Use This Issue
Do not use this Issue when the tool call is merely malformed. Do not use it when the tool is unavailable or missing required inputs. This Issue applies when the declared rules themselves conflict.