Turn one objective into a reviewed plan, then execute it with five parallel agents that never touch the same file. Nothing runs until you approve. Actions are risk-classified and gated on scored run history, so nothing executes autonomously until it has proven it can. Each agent can run in a kernel-isolated OpenShell sandbox with deny-all egress and a policy audit trail. Briefs, plans, and reports export as Markdown plus schema.org JSON-LD, so a directory of missions becomes a queryable history.
12 skills · 1 command · 6 agents · hooks
3