Slash commands

Slash commands are special instructions that begin with / and are typed in the chat input. They let you quickly trigger specific workflows: compact the context, start a new task, install a template, or run your own automation.

Built-in commands

Command	Description
/goal	Opens the Goal panel or starts a durable objective that Kodik pursues across turns. Learn more
/compact	Summarizes and compresses the current conversation to free up context window space. Learn more
/newtask	Duplicates the current conversation into a new task (“Copy of …”) so you can branch off. Learn more
/create-skill	Launches a guided flow for creating a new Skill for the current project or globally
/create-command	Creates a new custom slash command (a .md file in .kodik/commands/)
/create-rule	Creates a new rules file for controlling agent behavior
/create-hook	Creates a lifecycle hook — a shell command tied to an agent event. Learn more
/install-template	Starts a guided template installation workflow from the Kodik Marketplace
/analyze-workspace	Analyzes the workspace and provides a summary of the project

Custom commands

Beyond the built-ins, you can define your own slash commands as Markdown files that Kodik discovers and makes available to the agent.

File locations

Commands are loaded from multiple locations. When names collide, the higher-priority source wins:

Location	Scope
<selected-workspace>/.kodik/commands/	Selected chat workspace (highest priority)
<selected-workspace>/.agents/commands/	Read-only workspace compatibility
~/.kodik/commands/	User-global
~/.agents/commands/	Read-only user compatibility
Installed plugin’s commands directory	Namespaced plugin command

The selected chat session determines the exact workspace root, including in multi-root and remote windows. .agents commands remain inspectable in Settings but cannot be changed there, and workspace compatibility commands stay inactive until the workspace is trusted. Other .agents files are ignored. Duplicate identifiers follow the table’s precedence; same-rank duplicates are diagnosed as ambiguous and are not added to the / menu.

On the first launch of the new storage layout, Kodik moves global ~/.kodik/Commands into lowercase ~/.kodik/commands before opening the workbench. Existing lowercase content wins; exact duplicates deduplicate and conflicts remain inspectable under ~/.kodik/.migration/global-assets-v1/. After the committed cutover, uppercase global or workspace command folders are inert. Move any command created there later into the lowercase owner manually.

Symbolic links inside owned command or plugin paths are not followed. Linked roots, categories, command files, plugin manifests, and plugin configuration are reported as inactive. Copy the command into the canonical lowercase directory instead of linking it when you want it to appear in the / menu or be managed by Kodik.

Command file format

A command file is a .md file whose filename (without the extension) becomes the command name. An optional YAML frontmatter block can supply metadata:

---
name: deploy
title: Deploy to staging
description: Builds the project and deploys to the staging environment
argument-hint: "[branch]"
mode: code
---

Build the project, run the tests, and deploy branch `$ARGUMENTS` to staging.
If tests fail, stop and report the issue.

Supported frontmatter fields:

Field	Description
name	Command name (defaults to filename without .md)
id	Stable command identifier (defaults to name)
title	Human-readable display name shown in the UI
description	Short description of what the command does
argument-hint	Hint text shown in the autocomplete menu
mode	Execution mode (e.g. code, plan, ask)

Plugin commands

Commands contributed by plugins are namespaced as pluginId:commandName. In the chat you can invoke them with or without the prefix. See Plugins & Marketplace for details.