Repository navigation
bug: an unhashable value in an agent's operator-queue file silently stops answer delivery to that agent #3313
Description
Summary
services/operator_queue_service.py::_sync_agent raises a TypeError when an agent's queue entry holds a list or dict where a scalar is expected. Each poll cycle runs its syncs through asyncio.gather(..., return_exceptions=True), which discards the error without logging it. The sync fails again every 5 seconds, and no operator answers are written back to that agent while the bad entry is there.
| Trigger | Where it raises |
|---|---|
"id": ["x"] |
if req_id in open_by_rid is the one membership check with no isinstance guard |
An ingested entry's priority rewritten to a list |
changed_fields → _comparable_priority |
A list or dict priority on ingest |
_clamp_ingested_item. Its contract says it never raises; here the ask gets quarantined instead. |
Context
The queue file is written by the agent, which means by an LLM, so malformed shapes are realistic. One bad entry should cost only that entry, not the whole agent's answer channel. Found by the 2026-10-06 /edge-cases sweep.
Acceptance Criteria
- An entry with a non-scalar
id,priorityor other scalar field is skipped or quarantined on its own. Every other entry for that agent still syncs and gets its answers. -
_clamp_ingested_itemnever raises on any JSON value. - Exceptions from an agent's sync are logged, not silently dropped by
gather. - These strict-xfail tests in
tests/unit/test_ec_operator_queue_edges.pypass with their markers removed:TestUnhashableEntryValues::test_an_unhashable_id_does_not_stop_the_agents_answers::test_an_unhashable_priority_rewrite_is_a_change_not_a_crash::test_the_clamp_never_raises_on_an_unhashable_priority
Activity
fix(operator-queue): one malformed entry no longer stops answer deliv…