tool_calling — Claude Code prompt: tool-use rules
The tool_calling prompt for Claude Code and ChatGPT sets rules for using structured tools. It guides the agent to select a declared tool by its description and parameter schema.
MIT ★ 488 updated 2026-11-15
.claude/commands/tool_calling.mdWhat tool_calling does
The prompt tells the agent how to call tools declared in the current request. It should match a tool by its description and parameter schema, then use the exact declared name and namespace. Operation labels in prompts or descriptions do not guarantee a callable tool with the same name.
The prompt forbids inspecting simulator internals, privileged object state, service processes, or source files. It also rules out bypassing tools with shell commands. If a tool name is rejected, the agent should check the current declarations again. MCP resource listings are separate from callable-tool declarations, so an empty resource list does not show that declared tools are unavailable.
What it helps with
- select a tool by its parameter schema
- avoid guessing callable names from operation labels
- keep shell commands from bypassing declared tools
- distinguish MCP resources from callable tools
Example prompts for the agent
- Use the available tools according to their descriptions and parameter schemas.
- Find the declared tool for this operation and call it by its exact name.
- Recheck the current tool declarations if a call is rejected as unsupported.
How to install tool_calling
Manually: put the files in place
.claude/commands/tool_calling.mdHow to use
Paste the text at the start of a chat or into project instructions
Questions about tool_calling
How do I install tool_calling in Claude Code?
Save the prompt as .claude/commands/tool_calling.md.
What does tool_calling do?
It sets rules for selecting and calling structured tools from the declarations in the current request. It also bars shell-based workarounds and inspection of the listed internal data.
How does tool_calling determine whether MCP tools are available?
It relies on callable-tool declarations, not MCP resource listings. An empty resource list does not show that declared tools are unavailable.
Source text
tool_calling.md · MIT
Use the structured tools actually exposed in this session and follow their schemas. Do not inspect simulator internals, privileged object state, service processes or source files, and do not bypass the tools with shell commands. Tool names in prompts, skills and descriptions are operation labels, not fixed callable names. Select the matching tool from the declarations supplied with the current request by its description and parameter schema. Use its exact declared name and namespace structure; do not guess, add or remove a prefix based on documentation. Examples illustrate operations and arguments, not SDK or API call syntax. If a name is rejected as unsupported, recheck the current declarations before retrying or concluding that tools are unavailable. MCP resource listings are not the callable-tool list; empty resource listings do not establish that declared tools are unavailable.
© Open-RUA · MIT · Open-RUA/uni-robot-agent