fix(deps): raise molecule-ai-workspace-runtime floor to >=0.1.129 (closes internal#424) #5

Merged
devops-engineer merged 1 commits from fix/runtime-pin-floor-a2a-queued-parser into main 2026-05-15 22:54:39 +00:00

1 Commits

Author SHA1 Message Date
06739ba9e0 fix(deps): raise molecule-ai-workspace-runtime floor to >=0.1.129
All checks were successful
CI / test (3.11) (pull_request) Successful in 2m10s
CI / test (3.12) (pull_request) Successful in 2m10s
Pre-0.1.129 the runtime's inline A2A response sniffer in
``send_a2a_message`` checked only for ``result`` / ``error`` keys and
routed the poll-mode success envelope —
``{"status": "queued", "delivery_mode": "poll", "method": "..."}`` —
to the malformed branch:

    [A2A_ERROR] unexpected response shape (no result, no error): {...}

That string then propagated through ``tool_delegate_task`` into the
caller workspace's activity row, surfacing on canvas as the workspace's
task label and triggering a ~3s retry storm.

0.1.129 introduced ``a2a_response.py`` (SSOT typed parser with explicit
``Queued`` variant) + matching ``_A2A_QUEUED_PREFIX`` handling in
``a2a_tools_delegation.tool_delegate_task`` that falls back to the
durable ``/delegate`` + ``/delegations`` polling path — the correct
synchronous facade for poll-mode peers.

The existing ``>=0.1.110`` pin allowed a fresh ``pip install
codex-channel-molecule`` to resolve to a buggy wheel and reproduce the
incident. Raising the floor to 0.1.129 closes that window.

Adds a ``test_runtime_dependency_floor.py`` regression test that
parses ``pyproject.toml`` and asserts the lower bound stays at or
above 0.1.129 — bumps the version to 0.1.3 so a republish carries the
floor downstream.

Closes molecule-ai/internal#424.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
2026-05-15 15:44:25 -07:00