Guides
Eight ways kcode changes, from a setting to a handler inside the agent loop.
- 01
Settings and appearance
kcode takes model defaults, context budgets, memory, permissions and notifications from its settings, along with themes, fonts, shaders, tile behavior and key bindings.
- 02
Rules, prompts, and output styles
kcode reads project conventions, replacement sections of the system prompt and output styles from Markdown files, which shape what the model reads but not the tools it can run.
- 03
Commands
kcode runs your own slash commands, such as one that writes a standup note from recent commits, reviews a change against your checklist or runs a plugin action.
- 04
Skills: guides and programs
kcode runs two kinds of skill: a Markdown guide the agent follows, with its scripts and files, and a Lua program that kcode runs itself in a sandbox, with checked inputs and results.
- 05
Tools and plugin runtimes
kcode runs tools written in Lua and connects tool servers through MCP, for example to read tickets from your issue tracker or query an internal service.
- 06
Handlers in the agent loop
kcode lets a plugin attach handlers at named points in the agent loop, for example to record when a turn finishes, reject a tool call or lower a request’s output-token limit before it reaches the provider.
- 07
Hooks, schedules and triggers
kcode starts work on a schedule or an event, such as a weekly dependency review, or a finished CI build whose result goes into a running session for the agent to investigate.
- 08
Collaborators and working environments
kcode gives each kind of agent its own tools, such as read-only ones for a researcher and editing ones for an implementer, and runs independent changes in separate worktrees with their own setup and service ports.