You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit c4ea03e
Browse filesBrowse the repository at this point in the historyBrowse files
feat(sandbox): add --permissive flag for audit2allow-style policy discovery
Add permissive mode to sandboxes, enabling a single-pass onboarding
workflow for discovering what network policy a workload needs.
When `openshell sandbox create --permissive` is used, the proxy logs
but does not enforce network policy denials at both L4 (TCP connect)
and L7 (HTTP method/path) layers. Denied connections succeed with full
credential injection, and each denial feeds the existing
DenialAggregator → mechanistic mapper → draft proposal pipeline. Users
review all discovered endpoints at once with `openshell policy draft
get`, approve them, and export the learned policy.
Filesystem (Landlock) restrictions are skipped in permissive mode.
Seccomp, process identity, and SSRF guards remain enforced.
Permissive mode is threaded through all enforcement paths: CONNECT L4,
forward-proxy L4, single-route and multi-route L7 relay, forward-proxy
L7 path selection, and WebSocket/GraphQL message-level inspection. The
permissive flag flows from SandboxSpec (proto) through the gateway's
GetSandboxConfigResponse to the sandbox supervisor, which sets it on
ProxyPolicy and L7EvalContext.
ClosesNVIDIA#1839
Signed-off-by: Russell Bryant <[email protected]>
0 commit comments