AigentFlow Desktop is in pre-release. This documentation reflects the current build and will change before general availability.

Reference

Commands

Typed commands, most of which the app answers itself without spending a turn.

Type a slash in the instruction bar and the command list appears. Most are answered by the app directly. Some are rewritten into a governed prompt and sent to the model. A few belong to the underlying command-line tools and are declined, with the AigentFlow equivalent named.

Everyday

Command	What it does
`/help`	The command list
`/status`	Where the session stands
`/cost`	What this session has spent
`/context`	How full the context window is
`/clear`	Start again in a new tab, leaving this one alone
`/stop`	Interrupt the current turn
`/model`	Change the model

Control

Command	What it does
`/mode`	Show the session mode, or switch to the one you name
`/permissions`	What the agent may do right now, and under how many rules
`/plan` `/review` `/guarded`	Switch to that mode
`/working` `/autopilot` `/discussion`	Switch to that mode
`/handover`	Write a handover before the context window fills
`/worker`	Launch a worker on a defined task

Memory

Command	What it does
`/learn`	Remember a rule for this project, or `--global` for all
`/memory`	List what has been learned here
`/forget`	Remove learnings matching a keyword

Working the project

These become a governed prompt rather than a direct answer, so they spend a turn. They cover discovery and design (/discover, /map, /stack, /risks, /drift, /requirements, /architecture, /decide, /scope), planning (/milestone, /phases, /task, /estimate, /plan), building (/feature, /fix, /refactor, /debug) and checking (/review, /test, /verify, /diff, /retro, /summarize).

Commands you may know from the command-line tools, such as /init, /config, /hooks, /resume or /theme, are declined rather than passed through. The app names what to use instead, because those commands would change state the app is responsible for.