followup(validation): on-box Docker stdio-broker smoke before enabling AUTOBOT_CODEEXEC_ENABLED #11596

Activity

1. mrveiss

   added

   priority: high

   backend

   testing

   chat-workflow

   on Jul 11, 2026
2. mrveiss

   added a commit that references this issue on Jul 11, 2026

   test(chat): on-box Docker stdio-broker smoke gate for code-exec (#11596…

   ...

   a206ff4
3. mrveiss commented on Jul 11, 2026

   

   mrveiss

   on Jul 11, 2026

   OwnerAuthor

   More actions

   Harness merged to Dev_new_gui in PR #11614, squash commit a206ff4: autobot-backend/tests/integration/test_codeexec_docker_smoke.py — 4 skip-gated integration tests (two-tool merge, no-network, budget abort, multi-frame demux) over the REAL executor+broker+Docker path.

   This is a GATE, not a passing check — it CANNOT run in CI or this environment (no Docker); it self-skips unless AUTOBOT_CODEEXEC_DOCKER_SMOKE=1 + a reachable daemon + the autobot/secure-sandbox:latest image. Run on a box:
   AUTOBOT_CODEEXEC_DOCKER_SMOKE=1 python3 -m pytest autobot-backend/tests/integration/test_codeexec_docker_smoke.py -v

   Two blockers remain before AUTOBOT_CODEEXEC_ENABLED=1 anywhere:

   1. discovery(code-exec): compose script has no clean result channel — stdout is owned by the broker RPC stream #11613 (HIGH) — writing the smoke surfaced a real gap: the compose script has no clean result channel (broker owns stdout for RPC), so the merged result is polluted. test_smoke_two_readonly_tools_merge_in_one_dispatch will FAIL on-box until discovery(code-exec): compose script has no clean result channel — stdout is owned by the broker RPC stream #11613 is fixed — by design.
   2. This smoke passing green on an actual box.

   Closing the deliverable (harness written+merged); the two blockers above remain tracked and gate enablement.
4. mrveiss

   closed this as completedon Jul 11, 2026