Tools

The model chooses these tools itself. Permissions decide which calls run.

Tool	What it does
`glob_files`	Finds file paths that match a pattern.
`grep_files`	Searches file contents.
`read_file`	Reads text files and images.
`write_file`	Creates a file, or replaces one it has read.
`edit_file`	Changes part of a file. It refuses if the file changed since the last read.
`shell`	Runs commands in your shell. Dev servers keep running across prompts.
`web_search`	Searches the web.
`web_fetch`	Reads a public URL.
`subagent`	Hands a task to another agent with its own context.
`skill`	Reads a skill’s instructions.
`install_skill`	Installs a skill.
`capability_search`	Finds installed skills and MCP tools.
`mcp_call`	Calls a tool on an MCP server.
`ask_user_question`	Asks you a question and waits.
`read_tool_result`	Reads the full output of an earlier tool call.

Commands run with AI_AGENT=clue in their environment, so your tools can detect clue.