auggie 命令

常用示例

说明

auggie is a terminal-based AI coding agent built by Augment Code. It uses Augment's Context Engine for deep semantic understanding of codebases, going beyond simple text matching to comprehend project structure, dependencies, and relationships across large repositories. The tool operates in two primary modes. Interactive mode provides a full TUI with real-time streaming, visual progress indicators, vim-style keybindings, and observable tool execution. Non-interactive mode (--print) enables scripted automation for CI/CD pipelines, code review workflows, and headless environments. auggie supports sub-agents for specialized tasks such as security audits, test writing, and data analysis, and can execute multiple agents in parallel. Sessions are persistent and resumable, allowing conversations to be continued across terminal sessions or shared with teammates via the /share command. The CLI integrates with Model Context Protocol (MCP) servers for extended capabilities including GitHub, Linear, and Jira connectivity. It can also run as an MCP server itself, enabling integration with other tools. Multiple AI models are supported, and users can switch between them using the --model flag or the Option+M hotkey in the TUI. Custom slash commands can be defined as markdown files in .augment/commands/ or ~/.augment/commands/ directories, enabling reusable prompts for tasks like code review, bug fixing, and security analysis. User-specific rules in ~/.augment/rules/ and workspace rules customize agent behavior per project. The plugin system allows extending functionality through marketplaces, and an agent skills framework loads specialized domain knowledge from SKILL.md files following the agentskills.io specification.

参数

--print, -p

Run the prompt once and print the result to stdout, then exit. Useful for CI/CD pipelines and scripted automation.

--quiet

Return only the final output, suppressing intermediate messages and progress indicators.

--compact

Output tool calls, results, and response as single lines.

--output-format _FORMAT_

Output format for print mode (e.g., json).

--continue, -c

Resume the most recent conversation session.

--resume _ID_

Resume a specific session by ID or ID prefix.

-f

Filter session list to the current workspace only.

--dont-save-session

Skip saving conversation history for this session.

--delete-saved-sessions

Remove all saved sessions.

--model _NAME_

Select the model to use. Short names are supported (e.g., sonnet4.5).

--instruction _TEXT_

Provide an initial instruction for interactive mode.

--instruction-file _PATH_

Load the initial instruction from a file.

--workspace-root _PATH_

Set the workspace root directory.

--rules _PATH_

Append additional rules from a file.

--remove-tool _NAME_

Disable a specific tool for the session. Can be repeated.

--permission _SETTING_

Configure tool permissions at runtime.

--max-turns _N_

Cap the number of agent iterations in print mode.

--enhance-prompt

Improve prompts automatically before sending to the agent (non-interactive mode).

--image _PATH_

Attach an image to the prompt.

--shell _SHELL_

Set the shell to use for command execution.

--startup-script _PATH_

Specify a shell startup script.

--mcp

Run auggie as an MCP tool server.

--mcp-config _PATH_

Load MCP server configuration from a JSON file or inline JSON string.

--mcp-auto-workspace

Enable automatic workspace discovery in MCP mode.