Repository navigation
codex TRACE conformance test fails on main; fix/codex-trace-pins was never pushed #73
Description
plugins/agentrust-codex/tests/test_codex_capture.py::test_signed_outputs_verify_and_pass_trace_level_zero fails on main, and has done throughout #65, #67, #69 and #72. Verified pre-existing each time by stashing the change under review and re-running on the same base, so it is not attributable to any of them.
TR-POL PASS policy.enforcement_mode is valid ('silent')
Result: FAIL (8 checks, 1 failure(s), 0 skipped)
A local branch named fix/codex-trace-pins exists on at least one machine with the message "bump every trace pin to the v0.2 cutover", and it was never pushed. I referred to it in several PR bodies as "the branch addressing this", which was wrong: nothing upstream is fixing it. Correcting that here so the next person does not go looking.
Two things to decide: whether the pin bump is the actual fix, and whether this should be quarantined with xfail(strict=True) in the meantime so the suite is green and a regression is distinguishable from this known failure. Right now every codex run reports a failure, which trains people to ignore the count.
Activity
imran-siddique commentedon Aug 3, 2026
Resolved on main, closing.
test_signed_outputs_verify_and_pass_trace_level_zerolives inplugins/agentrust-codex/tests/test_codex_capture.py, whichagentrust-codex-tests.ymlruns viapython -m pytest plugins/agentrust-codex/tests -q. That workflow has been green onmainsince 2026-08-01 17:47. The capture-core refactor carried the trace pin bump, so the v0.2 cutover this issue was waiting on has landed.Both open questions in the body are now moot: the pin bump was the fix, and no
xfailquarantine is needed because the suite is genuinely green rather than green-by-suppression.For the record, the same v0.2 cutover defect existed in a second place.
ramen-ai-cmcphardcodedtag:agentrust.io,2026:trace-v0.1and its conformance job had been failing on main since 2026-07-28, including the scheduled run at 09:38 today. That one is fixed in #85, which also unblocks #82, #83 and #84.