You're viewing documentation for unreleased changes. View the latest release.

Commands

Type a slash command at the start of your message to trigger special actions. The commands below are built in.

Slash command autocomplete menu showing available commands

In addition, most skills can be run as /skill-name, so your own skills appear in the same menu. See the notes below for the exceptions.

Command Reference

CommandDescription
/reportCreate a Quarto document from the current conversation. Optionally add instructions after the command.
/notebookExport the conversation as a Jupyter notebook.
/savememoryCreate or update the project memory file. Add instructions to guide what gets saved. Use /savememory user to target the user-level memory file (~/.posit/assistant/AGENTS.md).
/compactSummarize the conversation to reduce token usage with Assistant’s built-in agent. Unavailable with agent engines. Previous messages are hidden but preserved.
/microcompactReplace eligible tool results in Assistant’s history with placeholders; does not compact an engine’s live session context.
/planEnter Plan mode to design an implementation approach before making changes.
/readonlyEnter Read-only mode for read-only exploration and questions — no file changes or direct R/Python code execution.
/newStart a new conversation with the current model, thinking effort, and web search settings, unless a fixed defaultModel is set.
/clearStart a new conversation with the current model, thinking effort, and web search settings, even if a fixed defaultModel is set.
/contextShow session context and token usage for the current conversation.
/pluginOpen the plugin & marketplace manager to browse, install, update, and enable plugins. Also available as /marketplace.
/mcpOpen the session panel, which lists your MCP servers and their status, and lets you sign in to any that need authentication.
/settings-reportShow which layer decided each setting and where each provider’s credentials come from. Secrets are never shown. See Checking what applies.
/aboutShow version and build information.
/trustMark the current workspace as trusted. Required for loading project-level memory files, skills, and settings.
/autoToggle Auto mode and Normal; trusted workspace and administrator permission required.
/restrictedToggle restricted mode in a trusted workspace. When enabled, tool calls you haven’t explicitly allowed in your settings require confirmation. Untrusted workspaces are always restricted.
/sandboxToggle sandbox mode for shell commands. Uses OS-level isolation on macOS/Linux; on Windows, gates active-shell commands with a built-in allowlist.
/yoloToggle automatic tool confirmation. When enabled, the assistant runs tools without asking.

Usage

Commands can be used on their own or with additional instructions:

/report
/report Focus on the visualization section and include the final plot code
/savememory Remember that this project uses tidymodels for all modeling work

Notes

  • Slash commands are different from the ! prefix, which runs shell commands and code from the chat input without a model turn. See Chat Features.
  • /report and /notebook use the report skill to generate documents.
  • /compact uses a low-tier model to summarize, so it completes quickly. The original messages are preserved — compaction just hides them from future requests.
  • /microcompact runs locally and does not send a message to the model. It is unavailable while the assistant is streaming.
  • Agent engines manage their own context and compaction; Assistant’s auto-compaction settings do not apply to them.
  • Most skills can be run as a slash command (/skill-name), including your own custom skills. Skills marked user-invocable: false are hidden from the menu, project skills need a trusted workspace, and if a skill’s name matches a built-in command, the built-in command takes precedence. See Skills for how invoking a skill differs from letting the assistant load it automatically.