[zzxwill]

When a review has no resumable session ID, a reply can establish a new session but previously discarded its ID. Save and reuse session IDs from both standalone and automatic replies, preserving reviewer/judge separation and legacy events. Correlate replies with their selected event snapshot so late or abandoned replies cannot overwrite a newer review or accepted session change.

Retain known or explicitly reported IDs across failures. Accept generated IDs only after successful execution and a valid, nonempty response; reject malformed, failed and empty whole/tail responses before storing unconfirmed IDs. Regression fixtures cover repeated replies, concurrency order, missing executables, nonzero exits and synchronized timeouts.

Validation on 92e3316b5566aed5cf9069d4cbbd25b428bc8262:

• 17 focused session/error/concurrency checks passed; fail-before evidence retained locally.
• node scripts/ci-check.mjs test: exit 0, 302 tests passed.
• node scripts/ci-check.mjs test:package: exit 0, 55 packaged files and installed CLI checks passed.
• Actual Jury: exit 0, NO NEW FINDINGS, clean after one round (96.7s). The final reviewer inspected the diff and passed syntax checks; its read-only sandbox prevented write-heavy tests, which passed independently above. Codex was configured for both roles; this clean first round did not require judge triage.

Exact executed review command:

GIT_AUTHOR_NAME=zzxwill GIT_COMMITTER_NAME=zzxwill node bin/jury.js review --dir /Users/bytedance/Documents/Codex/2026-10-20/task/codejury-post-142-audit --trunk master --reviewer codex --judge codex --rounds 2 --summary 'Review the focused reply-session persistence diff against master. Generated IDs require successful execution and a valid nonempty extracted report; known or explicitly reported IDs survive failures. Empty whole and tail responses are rejected in the shared result path. Concurrency tests cover abandoned/no-ID replies, completion order, stale snapshots and newer reviews. Independent final-head checks passed 302 tests and installed-package validation. The unchanged pre-existing run.json republish race is documented and deferred; do not implement it here. Assess new correctness regressions and preserve the focused scope.' --web=false --push=false > ../reply-session-review-evidence/jury-final-92e3316.log 2>&1

The unchanged, pre-existing run.json republishing race is documented and deferred; no fix for it is included. Separate master-CI diagnostics localized a package timeout to a web-launch test and console child but did not establish its root cause; no related source changes or retries are included. Live provider-session resumption is not claimed by local fake-provider regressions.