test(e2e): accept FAILED for #07 aout_custom_retry, matching its siblings - #152
Merged
Merged
Conversation
Suite 17 has 27 cases. 19 use validStatuses: BOTH (COMPLETED|FAILED), 5 use ["FAILED"] for raise, and only two use a strict ["COMPLETED"] — #7 aout_custom_retry and #9 aout_custom_fix. #7 is the outlier. Retry is model-dependent: the agent may or may not drop MARKER42 within max_retries, and since conductor-oss/conductor#1365 a retry that exhausts now escalates to raise rather than looping to the turn cap, so the run legitimately ends FAILED. Every other retry case in this file already accepts both outcomes (#3, #6, #12, #15, #21, #24). Aligning #7 with them. This does not weaken the check that matters: notContains "MARKER42" is still asserted whenever the run COMPLETEs, so a leak still fails the test. Deliberately NOT relaxing #27 tout_custom_fix, which also fails on 3.32.0-rc18. #9 aout_custom_fix stays strict and passes, asserting contains: "REDACTED" — so agent-output fix reliably applies and completes, while tool-output fix does not. That asymmetry points at the tool-output fix path server-side, and relaxing the assertion would hide it.
Codecov Report✅ All modified and coverable lines are covered by tests.
Flags with carried forward coverage won't be shown. Click here to find out more. 🚀 New features to boost your workflow:
|
shaileshpadave
approved these changes
Jul 29, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Suite 17: Guardrail Matrixhas 27 cases. TheirvalidStatusessplit:BOTH(COMPLETED|FAILED)["FAILED"]raisecases["COMPLETED"]#07 aout_custom_retry,#09 aout_custom_fix#07is the outlier. Retry is model-dependent — the agent may or may not dropMARKER42withinmax_retries— and since conductor-oss/conductor#1365 a retry that exhausts escalates to raise instead of looping to the agent turn cap, so the run legitimately endsFAILED. Every other retry case already accepts both (#3, #6, #12, #15, #21, #24).This doesn't weaken the assertion that matters:
notContains: "MARKER42"is still checked whenever the runCOMPLETEs, so a leak still fails.Not changed:
#27 tout_custom_fixIt also fails on server
3.32.0-rc18, and I've left it strict on purpose.#09 aout_custom_fixkeeps["COMPLETED"]and assertscontains: "REDACTED"— and it passes. So agent-outputfixreliably applies the fix and completes, while tool-outputfixdoes not. That asymmetry looks like a defect in the tool-output fix path rather than a flaky assertion, and relaxing#27would hide it.Worth someone with context on
GuardrailCompiler.compileToolGuardrailTasksconfirming whether the custom worker'sfixed_outputreaches the normalize script on the tool path. The script itself is fine — verified directly in GraalJS:fix+fixed_outputstaysfix,fixwithout it escalates toraise, and a passing guardrail is inert.How this was found
Running this bundle (
4.0.0-rc1) against conductor3.32.0-rc18: 3 failures, reproduced identically across 4 runs. On3.32.0-rc.14the same bundle is green, and #1365 (merged 2026-07-20) postdates the bundle release (2026-07-14). The third failure,Suite 16 test_guardrail_pass_no_interference, is also unaddressed here — a passing guardrail ends the SSE stream inerrorrather thandone, which likewise looks server-side.