Available commands
Examples:
The available thinking levels depend on the model SDK. A provider can ignore a
level that its model does not support. Cost and context fields also display only
when the runtime exposes those measurements.
Conditional control commands
These commands are handled before the general command dispatcher:
Approvals are disabled by default. Enable and configure them before relying on
/approve or /deny; see Approvals.
Prompt skills are not slash commands
The current daemon does not wire/skill:name to direct skill loading. A string
such as /skill:weather is ordinary text and may be interpreted by the model,
but it is not a deterministic command. Prompt skills are discovered and offered
to the agent for model-selected use. See Prompt Skills.
Likewise, /verbose and /reasoning are parsed internally but their directives
are not currently consumed by the agent execution path. They are intentionally
omitted from the supported command table.
Scope and delivery
Commands are processed by Comis, not by a messaging platform’s native slash command system. Availability can still depend on daemon wiring, sender trust, session ownership, and whether a live SDK session exists. A command that needs a live execution surface reports that it is unavailable when the surface is absent. Deterministic command, approval, and gate responses return to the complete authenticated turn endpoint, including the configured channel instance, conversation, and source thread. A response to a threaded command never falls back to the channel root.Sessions
How conversations are stored and managed
Models
Configure providers and model selection
Prompt Skills
Create model-selected reusable instructions
Approvals
Enable human approval for selected actions