Session Commands

/help

Shows documentation for CLI-JAW commands. When called without arguments, it lists all available commands grouped by category. When a command name is provided, it shows detailed usage, arguments, and examples for that specific command.

Arguments:

/commands

Opens the command palette — an interactive, searchable list of every available slash command. The palette supports fuzzy matching, so you can type partial names to quickly find what you need. This is the fastest way to discover commands you have not used before.

The palette displays each command with its category, brief description, and keyboard shortcut (if one exists). Select a command to either execute it immediately or insert it into the input for further editing.

/status

Displays a comprehensive snapshot of your current session. This includes the active model and CLI backend, loaded skills, conversation token count, memory usage, active goals, connected integrations (Telegram, Discord, etc.), and server health. Useful for debugging or understanding why the agent is behaving a certain way.

Status fields include:

/clear

Clears the conversation history from the current session. This removes all previous messages from the context window, giving you a fresh start while keeping your loaded skills, memory, and session configuration intact.

With the all argument, it performs a deeper reset: skills are re-synced to defaults, employees are reset, MCP is re-synced, and the session state is cleared. Memory is still preserved.

Cleanup tiers (lightest to heaviest):

/purge

Deletes conversation history and all structured memory (memory/structured/). Skills, employees, and configuration are preserved. Use this when you want the agent to forget everything it learned about you and your project in this instance, but keep the tools and integrations intact.

This is stronger than /clear (which keeps memory) but lighter than /reset confirm (which also wipes skills and employees).

/compact

Compresses the current conversation context by summarizing previous exchanges while preserving key decisions, code changes, and important context. This is essential for long sessions where you are approaching the context window limit but do not want to lose the thread entirely.

The agent reads the full conversation, identifies critical information (decisions made, files modified, errors encountered, goals in progress), and produces a dense summary that replaces the full history. You can optionally pass instructions to guide what the summary should focus on.

Arguments:

/reset

Performs a full bootstrap reset — the most destructive cleanup option. Clears conversation history, deletes all structured memory, hard-resets skills to factory defaults, resets employees, and re-syncs MCP. Only settings.json (server config, auth tokens, CLI subscriptions) survives.

Because this is destructive, the command requires confirm as an argument. Without it, the agent shows what will be deleted and asks you to confirm.

Arguments:

What is deleted vs. preserved:

/steer

Redirects the agent mid-conversation without clearing history. Use this when the agent is heading in the wrong direction or you want to pivot to a different approach while keeping the existing context. The steer prompt is injected as a high-priority system-level instruction that overrides the agent's current trajectory.

This is more surgical than /clear — you keep the full conversation history and context, but the agent re-evaluates its approach based on your new direction. It is ideal for course corrections: "stop doing X, do Y instead."

Arguments: