Commands

code-assist supports slash commands — typed as /command in the prompt. Commands are either local (executed immediately on the client) or prompt (expanded into an API query).

Command Types

Type	Enum Value	Behavior
Local	local	Runs client-side logic, returns TextCommandResult, CompactCommandResult, or SkipCommandResult
Prompt	prompt	Expands to API messages via get_prompt_for_command(), then submitted to the model
Local JSX	local-jsx	Like local, but returns rich UI components (TUI widgets)

Command Sources

Commands can be loaded from multiple sources, identified by CommandLoadedFrom:

Source	Description
bundled	Ships with code-assist core
managed	Organization-managed commands
skills	User-defined skills (see Skills)
plugin	Loaded from installed plugins
mcp	Provided by MCP servers
commands_DEPRECATED	Legacy command format

Built-in Commands

Session Management

Command	Type	Description
/clear	local	Clear the conversation history and start fresh
/compact	local	Compact the conversation to reduce token usage. Returns CompactCommandResult with the summarized context
/resume	local	Resume a previous session. Supports entrypoints: cli_flag, slash_command_picker, slash_command_session_id, slash_command_title, fork
/status	local	Show current session status — model, turns, token usage, cost

Permission & Mode Control

Command	Type	Description
/permissions	local	View and manage permission rules (allow/deny lists)
/allowed-tools	local	List tools currently allowed without confirmation
/plan	local	Toggle plan mode — the model can only read and analyze, not write
/model	local	Switch the active model (e.g., claude-sonnet-4-6, claude-opus-4)

Configuration

Command	Type	Description
/config	local	View or modify runtime configuration
/settings	local	Open the settings file in your editor
/memory	local	View loaded CLAUDE.md files and memory entries
/env	local	Show environment variable overrides

Development Workflow

Command	Type	Description
/review	prompt	Review the current diff or staged changes
/commit	prompt	Generate a commit message and create a commit
/pr	prompt	Create a pull request with generated title and body
/test	prompt	Run the project's test suite and analyze results
/lint	prompt	Run linters and fix issues
/simplify	prompt	Review changed code for reuse, quality, and efficiency

Code Understanding

Command	Type	Description
/explain	prompt	Explain a file, function, or concept in the codebase
/search	prompt	Search the codebase for a pattern or concept
/diagram	prompt	Generate a mermaid diagram of architecture or flow

Agent & Task Management

Command	Type	Description
/agent	local	Spawn a sub-agent with a given prompt
/task	local	Create or manage background tasks
/loop	local	Run a command on a recurring interval (e.g., /loop 5m /test)
/schedule	local	Create or manage scheduled remote agents (cron triggers)

Help & Debugging

Command	Type	Description
/help	local	Show available commands and usage
/version	local	Show the code-assist version
/debug	local	Toggle debug mode (verbose API logging)
/cost	local	Show cumulative token usage and estimated cost
/doctor	local	Run diagnostics (API connectivity, tool health, etc.)

Command Anatomy

Every command is a CommandBase dataclass with these fields:

@dataclass
class CommandBase:
    name: str = ""                  # Slash command name (without /)
    description: str = ""           # Short description
    command_type: CommandType        # local, prompt, or local-jsx
    is_hidden: bool = False         # Hidden from /help listing
    aliases: list[str] = []         # Alternative names
    argument_hint: str | None       # Placeholder for arguments
    when_to_use: str | None         # Hint for the model on when to invoke
    immediate: bool = False         # Execute without waiting for Enter
    is_sensitive: bool = False      # Redact from logs
    user_invocable: bool = True     # Can the user type this directly
    availability: list[str] = []    # "claude-ai" and/or "console"

Using Commands

Basic usage

> /help

With arguments

> /commit -m "fix: resolve null pointer in query loop"

Prompt commands with context

> /review src/code_assist/core/query.py

TIP

Prompt commands are expanded into API messages before being sent to the model. The model sees the expanded content, not the raw slash command. This means prompt commands can leverage the full tool ecosystem.

Command Result Types

Local commands return one of three result types:

TextCommandResult

TextCommandResult(type="text", value="Current model: claude-sonnet-4-6")

Displayed directly to the user.

CompactCommandResult

CompactCommandResult(
    type="compact",
    compaction_result={"summary": "...", "token_reduction": 4200},
    display_text="Compacted conversation: saved 4,200 tokens",
)

Replaces the conversation history with a compacted version.

SkipCommandResult

SkipCommandResult(type="skip")

Command handled silently — no output shown and no message sent to the model.

Command Availability

Some commands are restricted by user type:

• claude-ai — available to OAuth subscribers using the Claude AI service.
• console — available to users with a Console API key.
• Commands with an empty availability list are available to everyone.

WARNING

Commands marked disable_non_interactive = True cannot be used in headless/CI mode. Commands marked supports_non_interactive = True explicitly work in that context.