fix(desktop): restore live reaction and Inbox edit parity with relay semantics

[devin-ai-integration]

Summary

Closes #150.

Reproduces and fixes the four failing Inbox edit/reaction smoke specs on clean origin/main (35af74019) by addressing two independent root causes:

• Product regression: 304173e42 (fix(agent): close attention recovery gaps #113) added a client-side relayEventMatchesFilter re-check that rejects events with no h tag from #h-scoped live subscriptions. The real relay routes those same events by derived channel_id (crates/buzz-core/src/filter.rs), so h-less live reactions (kind:7) were dropped after delivery. This fix makes the desktop filter relay-parity-safe: h-less events are accepted on #h filters, while an explicit mismatching h tag is still rejected.
• Harness fidelity gap: __BUZZ_E2E_PUSH_MOCK_FEED_ITEM__ never materialized a backing relay event, so fix(agent): close post-merge attention review gaps #114's verified get_event lookup for selected Inbox items failed and structural/auxiliary queries never ran. The mock bridge now materializes channel-scoped feed items generically and dedupes the mock event store by id (matching relay-store semantics).

Records the live-filter parity rule as D-046 in docs/crew/DECISIONS.md and updates docs/crew/STATE.md because restored live reactions are now shipped-state truth. The E2E mock-bridge fix applies the existing D-041 rule.

Related issue

Fixes #150.

Classification per failure (from clean-main reproduction)

Spec	Classification	Mechanism
messaging.spec.ts:1819	Product bug	Client #h filter rejects h-less kind:7 live reaction before thread-cache merge
inbox-reactions.spec.ts:36	Product bug	Same #h filter regression prevents Inbox live reaction update
inbox-edit.spec.ts:175	Harness-fidelity gap exposed by product fencing	Feed-only mock item has no stored event → get_event fails → verified Inbox context cannot fetch structural edits
inbox-edit.spec.ts:325	Harness-fidelity gap exposed by product fencing	Same mock feed/store split

Testing

Clean origin/main smoke baseline (/home/ubuntu/evidence/origin-main-full-smoke-baseline.log):

• 1005 tests / 945 passed / 53 failed / 7 skipped / 2899 s

After this change (/home/ubuntu/evidence/inbox-fix-full-smoke.log):

• 1005 tests / 949 passed / 49 failed / 7 skipped / 2788 s

Focused issue-150 rerun after renumbering (/home/ubuntu/evidence/inbox-fix-focused-four-after-renumber.log):

• 54 passed / 0 failed / 0 skipped (includes inbox-edit.spec.ts:175, :325, inbox-reactions.spec.ts:36, messaging.spec.ts:1819)

Local relay-parity unit verification after latest rebase (2026-08-11):

• desktop/src/shared/api/relayEventVerification.test.mjs: 6 passed / 0 failed / 0 skipped

Net change: the four issue-150 failures now pass. Remaining full-lane failures are the preexisting project/git and flaky UI failures from the baseline (e.g. channels.spec.ts:500 is tracked by #110).

Gates passed:

• pnpm --filter buzz check — clean
• pnpm --filter buzz typecheck — clean
• pnpm --filter buzz test — 5049 passed, 1 skipped, 0 failed
• just test-unit — 8 workspace crates passed, 0 failed
• just desktop-check — clean
• just desktop-test — 5060 passed, 0 failed, 1 skipped
• pnpm --filter buzz build:e2e — clean
• git diff --check — clean

Pre-existing CI failures not in issue scope

Desktop Smoke E2E (1) in earlier runs reports a cluster of shard-1 failures that also appear in the #156 base run (job 93676494903 — 8 failed, 2 flaky, 242 passed), including channels.spec.ts:500 (tracked by #110), channels.spec.ts:1951, channels.spec.ts:2108, channel-activity-popover.spec.ts, channel-agent-presence.spec.ts, and community-rail.spec.ts:1088. These are unrelated to live-reaction or Inbox-edit parity.

Desktop Smoke E2E (2) and Desktop Smoke E2E (3) passed in earlier runs. Desktop Smoke E2E (4) is expected to cancel at the 30-minute timeout (tracked by issue #109).

Current CI run (after rebase onto c36d10350)

The latest nuncio-crew-ci.yml runs on this branch are failing because required jobs are cancelled by GitHub Actions before they complete:

• Run 31499492424 (SHA 0a3fc0896): Desktop Fast was cancelled after pnpm check completed successfully.
• Run 31499688638 (SHA 921f4615c): Desktop Fast started just desktop-check and ran through many desktop unit tests before being cancelled at 2026-08-11T14:06:56Z (~50 s into the job). macos-arm, Project Relay, buzz-acp, and desktop-rust were also cancelled.

This matches the latest origin/main run (31499231392) and other recent main runs. Rebase/push runs 31500680911 (SHA a77a8b927) and 31501454678 (SHA d2d998d81) also cancel desktop-fast and macos-arm before completion. gh run rerun is unavailable from this integration (Resource not accessible by integration). This is an infrastructure-level cancellation pattern rather than a code/test regression specific to this PR.

Decision number

D-046 records the new live-filter parity rule. D-041 (already in main) covers the mock-bridge relay-semantics rule that the E2E fix applies. D-045 belongs to PR #160 and was not reused.

Link to Devin session: https://app.devin.ai/sessions/23c0c77ec9344e42addf538aa1bd43da
Requested by: @oscarlehuu

[devin-ai-integration]

🤖 Devin AI Engineer

I'll be helping with this pull request! Here's what you should know:

✅ I will automatically:

• Address comments on this PR. Add '(aside)' to your comment to have me ignore it.
• Look at CI failures and help fix them

Note: I can only respond to comments from users who have write access to this repository.

⚙️ Control Options:

• Disable automatic comment, CI, and merge conflict monitoring