TermAl

Feature Reference: Session Slash Commands

This document describes the slash-command behavior that TermAl currently ships in the composer.

Backlog source: docs/bugs.md

Status

Implemented for session controls and Claude-native slash commands.

What TermAl supports today

Typing / in the composer opens a session-control palette. The palette is scoped to the active session and uses the same settings APIs as the Prompt tab.

Supported commands:

• /model for Claude, Codex, Cursor, Gemini, and OpenCode
• /mode for Claude, Cursor, Gemini, and OpenCode
• /sandbox for Codex
• /approvals for Codex
• /effort for Claude and Codex
• /fast for Codex
• /mcp and /mcp verbose for Codex

Behavior

• Keyboard navigation works inside the palette.
• Enter applies the highlighted choice and closes the palette.
• Space applies the highlighted choice and keeps the palette open.
• The active choice stays aligned with the real selected session setting after live refreshes and setting changes.
• /model supports manual model-id entry for Claude, Codex, Cursor, and Gemini. OpenCode accepts only its live ACP model options plus Auto; see OpenCode ACP Integration.
• For live model lists, labels resolve to canonical ids before TermAl stores the selection.
• /mcp is a local status command, not a model prompt. It reads the owning Codex app-server’s configured MCP servers, authentication state, and tool counts. /mcp verbose also shows individual tool names and descriptions. Remote sessions proxy the read to the TermAl instance that owns the Codex runtime.

Live model integration

For Claude, Codex, Cursor, Gemini, and OpenCode, the slash palette can:

• trigger live model refresh when the model list is missing
• show inline loading state
• show inline error guidance
• offer retry without leaving the composer

What is not implemented yet

This is not full cross-agent slash-command parity. TermAl now discovers Claude’s native slash commands from the live runtime when available and falls back to .claude/commands prompt templates, but it still does not have equivalent native-command discovery for Codex, Cursor, Gemini, or OpenCode. The dedicated Codex /mcp status surface is implemented directly against the documented app-server API and does not imply general Codex native-command discovery.

Project command templates and future Claude skills are owned by the backend agent-command resolver, not by this session-control layer. See agent-slash-commands.md for $ARGUMENTS, optional Additional User Note handling, and the shared regular-send/delegation resolution contract.