test(members): drive an increasing clock in roster order test (#8499)

[CrysisDeu]

Summary

test/test_members_dm_thread.py::TestMembersRoster::test_roster_orders_by_message_ts_not_file_mtime is flaky on the Windows CI shard. It appends one message to each of two freshly created transcript files back-to-back and asserts strict < on their last_active_ts. monotonic_transcript_ts only corrects against prior rows of the SAME file, so both appends take raw datetime.now() reads; on Windows' ~15.6 ms clock tick the reads collide and the assertion fails on equal timestamps. The production code is correct — cross-file ordering is not something it promises; the test assumed clock resolution it does not control.

Test-only fix: monkeypatch kiro_crew.history.datetime in the test with a strictly increasing, tz-aware clock (mirrors the existing prior art in test/test_history.py::test_recent_from_source_sorted_by_ts), so the chronological order the test asserts is encoded in the timestamps on every OS. Two refinements over the prior art, from pre-push review: the stand-in subclasses datetime.datetime so fromisoformat stays available while patched (keeps _parse_transcript_ts undegraded), and a comment marks the deliberate fake-message-clock vs real-os.utime-clock contrast so a later reader does not "align" them.

The os.utime mtime bump on the OLDER thread's file is unchanged — that is the behaviour under test (mtime must not reorder rows). The assertion stays strict <. No changes to src/kiro_crew/history.py or roster code.

Closes #8499

Testing

• isort --check-only, flake8, mypy src/kiro_crew/ — all pass locally
• Diff-scoped black gate and brand gate pass post-commit
• Full pytest runs in CI (local test execution is disabled on this host by operator policy)

Pre-push review

Two model-pinned read-only reviewers on the staged diff:

• GPT lane (gpt-5.6-sol): PASS, no findings
• Opus lane (claude-opus-5): PASS, 3 Low advisories — 2 applied (datetime subclass; two-clocks comment), 1 deferred (shared conftest fixture for the duplicated fake-clock helper, out of this fix's file scope)

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of 0330c88e7849468c590b18aeebe2e4844c8c6850 — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Deterministic clock injection attacks the actual flake cause (clock granularity), matching repo flake discipline and existing prior art — no design-level concerns.

[DESIGN-REVIEWED] 0330c88

[github-actions]

GPT 5.6 Review — ✅ no blocking findings

GPT 5.6 completed its review of 0330c88e7849468c590b18aeebe2e4844c8c6850 and found no blocking issues.

This comment is updated in place on each push.

Review details

No findings.
[GPT-REVIEWED] 0330c88

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 0330c88e7849468c590b18aeebe2e4844c8c6850: <one-sentence reason>

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed 0330c88e7849468c590b18aeebe2e4844c8c6850 — this comment is updated in place on each push.

Review details

The change is test-only and internally consistent: kiro_crew.history imports datetime as a name (from datetime import datetime), so monkeypatch.setattr("kiro_crew.history.datetime", _IncDateTime) correctly intercepts the two datetime.now().astimezone() reads in append; the _IncDateTime subclass keeps fromisoformat/.astimezone() working, and the incrementing clock makes the old-key row (base+1s) strictly precede the new-key row (base+2s), so the < assertion holds deterministically. No candidates to falsify and nothing grounded to add.

No findings.

[OPUS-REVIEWED] 0330c88

Verdict parsed from the review's SHA-scoped output markers for commit 0330c88e7849468c590b18aeebe2e4844c8c6850.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 0330c88e7849468c590b18aeebe2e4844c8c6850: <one-sentence reason>

[NicholasRBowers]

Tier 1 auto-approve: test (1 file). Criteria: no conflict, no requested changes, security path denylist clean, design-doc gate clean, SAST annotations clean, security checklist all-NO, AI reviewers green. Category: test-files-only change driving a strictly-increasing monkeypatched clock in test_members_dm_thread.py to fix a coarse-clock timestamp collision (#8499); no runtime code touched.