OpenClaw's 2026.9.8 hotfix train picked up another major repair stack this morning with PR #163074, titled "fix: backport release-critical update and session repairs." The PR merged before the 08:00 UTC morning cutoff and is one of the strongest signals yet that the 2026.9.8 branch is being hardened around update recovery and session continuity.
The pull request says the release candidate was missing confirmed fixes for update, Windows copy, performance, memory, and delegated-work behavior that landed after the release branch was cut.
What The Backport Adds
The user-impact section is unusually direct. The 2026.9.8 branch now avoids several known failure modes:
- Update rehearsal and copy failures.
- Excessive state-worker and Codex fleet memory.
- Slow memory-sync shutdown.
- Repeated session and redaction work.
- Unbounded shell snapshot caching.
- Delegated sessions that stop silently or deliver replies twice.
- macOS npm updates rejecting the standard
/vardirectory alias.
The backport pulls from a long list of upstream fixes, including #161003, #162076, #162231, #162352, #162403, #162387, #162616, #162810, #162967, #162227, #162567, and #163129.
Two of the session fixes needed adaptation because the release branch does not have newer main-only owner APIs. The PR says that audit found and fixed a real adaptation gap: key-only callers now retain the resolved requester session ID and lifecycle revision.
Why This Is A Release Story
This is not a final release announcement. The PR explicitly leaves the generated changelog unchanged because the release workflow regenerates it after the product set freezes. Still, it tells operators what kind of hotfix 2026.9.8 is becoming.
The highest-risk areas are exactly the ones a hotfix should address carefully: updater behavior, SQLite rehearsal copies, Windows runtime handling, delegated session delivery, and memory pressure in Codex-heavy fleets. These are not cosmetic changes. They are the kind of reliability repairs that decide whether a patched Gateway can survive real installations.
The PR also documents exclusions. One candidate depends on main-only original-state metadata, another depends on native resource claims, and a filesystem dependency update is held because it would increase release risk. That selectivity matters. A good hotfix branch is defined as much by what it leaves out as by what it pulls in.
Validation And Remaining Watchpoints
The evidence section reports focused session continuation and publication tests, 124 redaction tests, Codex fleet memory proof, Windows package swap tests, changed gates, and git diff --check.
There is one important caveat: the local date-based plugin compatibility-expiry gate still fails on unchanged current source with the same overdue records. The PR treats that as an existing gate condition rather than a new defect introduced by this backport.
The PR also says canonical Full Release Validation follows merge. That is the next signal to watch. The final validation needs to cover first-hop updates, survivor paths, Linux, Windows, and macOS install and upgrade cells, packaged Windows llama startup, GPT provider routing, and delayed or restarted subagent completion without duplicate delivery.
For now, PR #163074 makes the 2026.9.8 branch more concrete: a hotfix aimed at update durability, session correctness, memory pressure, and delegated-work reliability.