Security
Guardrails that make swarm work safe, inspectable, and reversible.
How we're different
- Budgets and approvals are enforced by policy gates.
- Every action is logged in an audit trail (who/what/when/why).
- Verification harness reduces “it looked fine” failures.
Trust rule: AI drafts; humans decide for sensitive outcomes.
Controls (plain English)
RBAC + approvals
Role-based access and dual control for risky actions.
Budget governors
Caps per bead, per role, and per time window.
Immutable audit trail
Append-only timeline plus artifacts (diffs, logs, screenshots).
Secrets hygiene
Credentials never stored in the browser; keep secrets in env vars.
Data minimization
Short-lived session context; durable state is structured and scoped.
Incident playbooks
Escalation paths and rollback flows for failed merges.
Legal/Compliance disclaimer: Not legal advice; not a substitute for your compliance obligations.