Does OpenClaw v2026.5.28 fix the Codex app-server crash propagation?

Yes. Codex app-server and helper failures no longer tear down shared runtime state. Shared app-server clients now survive startup and spawned-helper failures.

What security changes are in OpenClaw v2026.5.28?

Group prompt text is kept out of system prompt, repeated-dot hostnames are normalized, side-effecting command wrappers and unsafe Node runtime env overrides are blocked, no-auth Tailscale exposure is rejected, and node/device-role approvals require admin authority.

Is OpenClaw v2026.5.28 a stable release?

No, v2026.5.28-beta.2 is a pre-release. The latest stable is v2026.5.27. Beta users should expect potential issues.

What is the subagent workspace isolation improvement?

Subagents now maintain proper cwd/workspace separation, preventing cross-contamination between agent sessions. Hook context stays prompt-local instead of leaking across contexts.

← Back to dashboard

clawsmith.com/signal/openclaw-v2026-5-28-beta-agent-codex-runtime-recovery

📈 TrendsWide OpenLive

OpenClaw v2026.5.28-beta: Agent and Codex Runtime Recovery — Subagent Isolation, Session Lock Release, Stale Restart Prevention

OpenClaw v2026.5.28-beta.2 released May 29. Agent and Codex runtime recovery improvements: subagents keep cwd/workspace separation, hook context stays prompt-local, session locks release on timeout abort, stale restart continuations avoided, Codex app-server/helper failures no longer tear down shared runtime state. Security: group prompt text out of system prompt, repeated-dot hostnames normalized, unsafe Node runtime env overrides blocked. Part of ongoing stability focus post-rough-week.

Product Idea from this Signal

A vulnerability intelligence feed that aggregates AI agent security events across the OpenClaw ecosystem and delivers scored alerts within minutes of disclosure

453.5k ▲

OpenClaw accumulated 138 CVEs in 63 days at a pace of 2.2 new vulnerabilities per day, while 155,000 unprotected instances sit exposed on the internet. Existing CVE databases track millions of generic entries but none focus specifically on the AI agent ecosystem. The jgamblin/OpenClawCVEs GitHub tracker (135 stars) proves demand exists, but it is a static repo with no alerting, no scoring, and no API. This product aggregates all AI agent security events in real time from GitHub Security Advisories, ClawHub skill audits, NVD feeds, and exposed instance scans, scores each by exploitability and blast radius specific to agent deployments, and delivers prioritized alerts via webhook, Slack, RSS, or API within minutes of disclosure.

APISECURITYOPEN-SOURCESAASDEVTOOL

CompetitiveView Opportunity →