fix(members): enqueue the rules read/write audits inline, not via to_thread

[CrysisDeu]

Problem / Motivation

Main is red on test/test_members_dm_thread.py::TestDenialAuditOffload::test_no_members_sel_audit_is_offloaded (seen on #8816's merge ref, Backend Tests (Windows) 3). Two PRs crossed: #8741 (#8608) warmed the SEL singleton at startup and inverted the members AST guard so that no log_api_access may sit inside an asyncio.to_thread lambda; #7235 then landed the member rules GET/PUT handlers with exactly that wrapper around their success audits (handlers/members.py lines 581 and 716).

What changed

The two rules audits become direct enqueues, guarded the way the deny-path audits in the same file already are (try/except Exception → debug log; an audit must never change the outcome). Comments updated to name the startup warm (#8608) instead of the first-touch initialization that no longer happens here. No behaviour change for the request: the audit still records the same operation/outcome/resources.

Tests

Per the owner's standing instruction no suite was run locally; CI verifies. black --check and flake8 clean on the file. Unblocker for #8816 and any PR whose merge ref includes both #8741 and #7235.

Related Issues

Cross-merge of #8741 and #7235. Unblocks #8816.

Checklist

• One commit, Conventional Commits title
• Existing tests pass (CI)

[github-actions]

Design Review (Fable 5) — ✅ PASS

Design-level review of 7268629692d8645d6db1555197bdffa2db5fca6c — updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

Design-Verdict: PASS

Minimal cross-merge repair that restores the file's own established audit pattern, pinned by the existing AST guard; nothing to redesign.

The two rules-audit sites now match the deny-path sites in the same file byte-for-pattern (direct enqueue, try/except → debug log), log_api_access without critical=True is a non-blocking enqueue post-warm (sel.py log() docstring), and the inverted guard in test_no_members_dm_thread.py already owns the invariant this change satisfies. Root cause (two PRs crossing) is correctly fixed at the violating call sites rather than by weakening the guard.

[DESIGN-REVIEWED] 7268629

[github-actions]

First Principles Review (Fable 5) — ✅ PASS

Premise-level review of 7268629692d8645d6db1555197bdffa2db5fca6c — why this exists and whether the shipped surface is the smallest honest version. Updated in place on each push. A BLOCK verdict blocks PR readiness; PASS/CONCERNS are advisory.

All claims verified. The remaining to_thread-wrapped audits in cron.py (3 sites) and md_notebook (line 1414) are critical=True synchronous writes, which the AST guard's docstring explicitly exempts — with one exception at md_notebook/server.py:1428, a non-critical audit still paying the pointless hop. The fix itself is pinned by the AST guard test test_no_members_sel_audit_is_offloaded, and the try/except guard matches the four existing audit sites in the same file.

First-Principles-Verdict: PASS

Unwedges a red main by deleting two thread hops the startup warm made pointless; every line is either the fix or the file's established audit guard.

What this change ships

Intent: get main green again after #8741's AST guard and #7235's rules handlers cross-merged — a FIX.

1. Rules-read success audit runs inline, no worker-thread hop — justified, pinned by the AST guard test
2. Rules-write success audit runs inline — justified, same guard
3. A failing audit on either route now logs at debug instead of failing the request — declared, matches 4 sibling guarded audits in members.py
4. Comments now cite the startup warm instead of the removed first-touch init — declared

The change is pure subtraction plus the guard the file already uses everywhere else (grepped log_api_access in members.py: 4 pre-existing sites, all try/except Exception → debug). No new surface, config, or export. Root cause (two green PRs whose merge violates an invariant only their union exhibits) is CI-topology, out of any diff's scope; this fix sits at mechanism level, the right level available.

Watch

• One sibling of the same root cause stays unfixed: md_notebook/server.py:1428 wraps a non-critical log_api_access in to_thread post-warm (grepped to_thread[…]log_api_access; the other 4 remaining wraps are critical=True synchronous writes the AST guard docstring exempts). Fine to leave out of a red-main unblocker, but it is the last leftover of the Offload SEL first-touch init at gateway startup, then delete the 18 per-site to_thread audit wrappers #8608 conversion.

[FIRST-PRINCIPLES-REVIEWED] 7268629

[github-actions]

Opus 4.8 Review — ✅ no blocking findings

Reviewed 7268629692d8645d6db1555197bdffa2db5fca6c — this comment is updated in place on each push.

Review details

No findings.

[OPUS-REVIEWED] 7268629

Verdict parsed from the review's SHA-scoped output markers for commit 7268629692d8645d6db1555197bdffa2db5fca6c.

False positive or not applicable? A repository writer can comment:
/ai-review override fable 7268629692d8645d6db1555197bdffa2db5fca6c: <one-sentence reason>

[github-actions]

GPT 5.6 Review — 🔴 changes requested (blocking)

GPT 5.6 found at least one blocking issue that must be resolved before merging 7268629692d8645d6db1555197bdffa2db5fca6c. Adjudication upheld 1 of 1 blocking finding(s); those must be resolved before merging.

This comment is updated in place on each push.

BLOCKING -- src/kiro_crew/dashboard/handlers/members.py:583 -- Failed SEL warm retries blocking initialization on the event loop
_sel().log_api_access( (also line 719)
Transient startup warm failure -> rules GET/PUT -> _init_locked filesystem work on the loop -> gateway requests and heartbeat stall.
Anchor: no-blocking-call-on-event-loop
Fix: Resolve _sel() with await asyncio.to_thread(_sel) before logging at both sites.
[BLOCK-MERGE] 7268629
[GPT-REVIEWED] 7268629

Adjudication (Opus 4.8) — is blocking on each finding proportionate?

The finding is anchored to no-blocking-call-on-event-loop (AUTOSDE.yaml:91-92), which carries blocking: true. I verified the mechanism: _sel() at members.py:583 and :719 runs directly on the event loop inside async handlers (no to_thread hop), and per sel.py:3349-3384 a failed startup warm leaves the first later touch to run _init_locked's blocking file I/O (key load/create, backward tail scan bounded by _SEGMENT_MAX_BYTES, stat) on the caller's thread — here the loop thread. That is exactly the "large synchronous file IO reachable from an async handler" the rule forbids.

Harm rung: MEDIUM-to-HIGH loop stall, but the anchor rule's blocking: true flag is authoritative and outranks the weighing. The remedy (wrap _sel() in await asyncio.to_thread at both sites) is cheap and local. UPHOLD under autosde-blocking-rule.

[ADJUDICATION] 7268629 total=1 uphold=1 downgrade=0
UPHOLD F1 src/kiro_crew/dashboard/handlers/members.py:583 reason=autosde-blocking-rule
[GPT-ADJUDICATED] 7268629

False positive or not applicable? A repository writer can comment:
/ai-review override gpt 7268629692d8645d6db1555197bdffa2db5fca6c: <one-sentence reason>