fix(#6): connect/read timeouts on every HTTP agent

No ureq call had a timeout, so a silent registry hung pull forever. One
oci::agent() builder (30 s connect + per-read) now backs registry agents,
local ones included, and the kernel downloads.
This commit is contained in:
2026-10-22 19:54:09 +02:00
parent 2231e4efe1
commit 56762648e3
3 changed files with 12 additions and 4 deletions
+1
View File
@@ -72,6 +72,7 @@ The VM is the boundary. The guest (including a prompt-injected agent or a malici
Auth realms must be https. For public registries, connections to loopback/private/link-local addresses are refused
(token realms and redirects included, IPv6 forms like NAT64/6to4 too). A registry counts as local only by name:
`localhost` or a private IP literal (`192.168.1.5:5000`). LAN registries addressed by a DNS name are refused; use the IP.
All downloads (registries, kernel) time out after 30 s without progress (connect or any single read).
- **pi config**: allowlisted items only. Symlinks are followed only into `~/.pi/agent` or `~/.agents/skills`, with no cycles
and no special files, capped at 256 MB.
- **exec**: sessions run as root before dropping to the user, so the guest helper accepts them only from the host's vsock CID,