Skip to content

test(ai): lock auto and verify orchestrator transport contracts - #109

Closed
cursor[bot] wants to merge 3 commits into
mainfrom
cursor/bc-3e19ce03-7fa4-49d9-8e2b-752d1a7d4644-b821
Closed

test(ai): lock auto and verify orchestrator transport contracts#109
cursor[bot] wants to merge 3 commits into
mainfrom
cursor/bc-3e19ce03-7fa4-49d9-8e2b-752d1a7d4644-b821

Conversation

@cursor

@cursor cursor Bot commented Aug 16, 2026

Copy link
Copy Markdown
Contributor

Summary

  • Keep citation-bearing post-chat and lineage adjudication off the auto source-scan list so a docstring mention of mode="auto" cannot satisfy ADR-0013.
  • Add monkeypatched transport tests that call ContextualOrchestratorPostChatClient.answer and ContextualOrchestratorAdjudicationClient.judge and assert the outbound payload is mode="verify".
  • Name both checked-judgment exceptions in the Unreleased changelog, and stop telling operators that every other client still uses mode="route".

This branch is #106 plus the contract fix. Prefer this PR over merging #106 as-is.

Product contract

  • Ordinary product adapters keep mode="auto".
  • Post-chat and adjudication keep mode="verify" on the wire.
  • A source scan for route remains, but auto/verify membership is no longer inferred from comments.

Test plan

  • pytest tests/test_contextual_orchestrator_default_policy.py — 2 passed, 7 subtests
  • Isolated post-chat and adjudication transport assertions (mode == "verify", citation id, confidence 0.91)
  • Required PR checks
Open in Web View Automation 

cursoragent and others added 2 commits August 16, 2026 14:43
Keep ADR-0013 and drop the colliding ADR-0005 copies. Record one
Unreleased changelog entry, restore the runtime-adapter and
post-evaluation transport regressions, and correct the leftover
post-chat docstring so it no longer describes a forced route.

Co-authored-by: Seongho Bae <[email protected]>
Keep post-chat and adjudication off the auto source-scan list so a
docstring mention cannot satisfy the product default. Assert both
checked-judgment clients send mode=verify on the wire.

Co-authored-by: Seongho Bae <[email protected]>
The live client already sends mode=auto. Keep the research note
aligned with ADR-0013 so operators do not reintroduce route.

Co-authored-by: Seongho Bae <[email protected]>

@cursor cursor Bot left a comment

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Stale comment

No blocking production defect on this head. Ordinary adapters already send mode="auto"; post-chat and adjudication already send mode="verify". The new answer/judge monkeypatches close the #106 hole where a docstring mode="auto" greened post-chat as an auto client.

Residual: tests/test_contextual_orchestrator_default_policy.py still whole-file substring-scans while its messages say "executable source." Membership can still be satisfied by a comment unless the assertion requires the payload literal "mode": "auto" / "mode": "verify".

Prefer #113 (5eeaa7f) over merging this draft. It keeps these transport tests, requires those payload literals, and ships the related-node business-caption slice (v0.75.0) that #92 left unmergeable.

CodeRabbit CLI 0.7.3 is installed here but coderabbit auth status --agent is not_authenticated (automatic login timed out). This is not a CodeRabbit review.

Do not merge #106 as-is. Do not treat queued Checks as a blocker for the next caption/contract work; they are not evidence of success.

View PR

Open in Web View Automation 

Sent by Cursor Automation: Fix Issues

@cursor cursor Bot left a comment

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Findings

No production-payload defect on this head. The answer() / judge() wire tests close the original #106 post-chat-as-auto hole.

The source scan is still the weaker #106 a815440 form: _AUTO_MARKERS / _VERIFY_MARKERS can go green from a docstring. Closed #113 mixed this lock with related-node captions; do not reopen that mix.

What to do next

Close this draft in favor of #117. That PR is #106 plus payload-literal scans only.

#117

Open in Web View Automation 

Sent by Cursor Automation: Fix Issues

Comment on lines +54 to +57
self.assertTrue(
_contains_any(source, _AUTO_MARKERS),
f"{relative} must request mode=auto in executable source",
)

Copy link
Copy Markdown
Contributor Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

This _contains_any(..., _AUTO_MARKERS) check still treats a docstring mode="auto" as enough. #117 requires the payload literal "mode": "auto" (and "mode": "verify" for the verify list), with the post-evaluation typed-default exception.

Close this draft in favor of #117 rather than merging this scan.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants