FAQ
Short answers to the questions people ask first.
Why Go?
Distribution, not preference. A static Go binary installs by Homebrew or curl with no runtime on your machine.
Which agents can a soldier run on?
Claude Code, which is the default, Pi and OpenCode, chosen per model profile with "harness": "pi" or "harness": "opencode". The commander can also run in Pi or OpenCode (vx init --commander). The binary stays plumbing behind one Harness interface.
Do I have to learn commands?
No. You talk to the commander, in Claude Code by default (or in Pi or OpenCode), and it runs vx for you. The CLI reference is for when you want to drive it by hand.
Does vexillum open pull requests?
Not by default. vx land merges locally. vx ship runs the tribunal pipeline and then opens a real PR. See Land vs ship.
Is it safe to let soldiers run?
Treat soldiers as you would an unattended session with your credentials. They run with full permissions (--dangerously-skip-permissions on Claude Code, no permission prompts on Pi, --auto on OpenCode), with no sandbox. The camp isolates their commits; your approval before landing is the control. Do not dispatch where reading secrets would be a problem.
What happens if a soldier crashes, or my machine restarts?
State is JSON on disk. A task whose pane is confirmed gone becomes interrupted, with its committed work still in the camp. See Blocked and interrupted.
Can two projects run at once?
Yes. State is namespaced per project, and one sentinel per machine serves them all.
Where does the commander's behavior come from?
From markdown in your project that you can edit: a short core in a block of your AGENTS.md, and the vexillum skill the commander loads when it dispatches, lands or ships. See AGENTS.md and rules.
Is there a second commander, or other agents?
No. You talk to one commander, and it dispatches the soldiers.